morphology and image analysis of some solar …
TRANSCRIPT
1
MORPHOLOGY AND IMAGE ANALYSIS OF SOME SOLAR
PHOTOSPHERIC PHENOMENA
By
ASMA ZAFFAR
Supervisor:
PROFESSOR (Dr.) MUHAMMAD RASHID KAMAL ANSARI
C0- Supervisor:
PROFESSOR (Dr.) MUHAMMAD JAVED IQBAL
Presented in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
Institute of Space and Planetary Astrophysics (ISPA)
University of Karachi (Pakistan)
2018
2
In the Name of Allah, the Most Beneficent, the most Merciful.
May Allah increase in my Knowledge
(Al-Quran Alkarim)
3
DEDICATED
TO
My Beloved PARENTS (Zafar Ali and Nasreen Zafar) and
Dr. Shaheen Abbas
4
CERTIFICATE
We accept the thesis as conforming to the required standard.
1. Internal Examiner (Supervisor):
2. External Examiner:
5
CERTIFICATE
It is to certify that the Ph.D. thesis entitled “Morphology and Image Analysis of
Some Solar Photospheric Phenomena by Asma Zaffar is completed under my
supervision.
Supervisor
Professor Dr. Muhammad Rashid Kamal Ansari
Co- Supervisor
Professor Dr. Muhammad Jawed Iqbal
6
CONTENTS
(i) Acknowledgment ix
(ii) Articles Appended to the Thesis x
(iii) Abstract (Englisn) xi
(iv) Abstract (Urdu) xiii
Chapter 0: Introduction 1
0.1 Birth of Sunspots 1
0.2 Sunspots 1
0.3 Group of Sunspots or Active Region 2
0.4 Sunspots Cycles 4
0.5 Solar-Climate Relationship 5
0.6 Enso- Sunspots Relationship 6
0.7 Our Approach 7
0.8 Thesis Outlook 8
Chapter 1: GARCH Model for Sunspot Cycles
11
1.1 Data Description and Methodology 12
1.2 Time Series Analysis With Stationary Process 13
1.3 Autocorrelation Function, Partial Autocorrelation Function 14
and Ljung-Box Q-Statistics Of Sunspot Cycles
1.4 Unit Root Test Of Sunspots Cycles Based On Akaike 24
Information Criterion (AIC)
1.5 Tests for Normality 29
1.6 Test of Adequacy 31
1.7 Diagnostic Checking Test 32
1.8 ARMA Model 33
1.9 AR (p)-GARCH (1, 1) Process Of Sunspot Cycle 39
7
1.10 ARMA(p, q) – GARCH (1, 1) Methods of Sunspot Cycles 50
1.11 Comparison Of Forecasting Evolution of ARMA, AR-GARCH 60
and ARMA-GARCH of Sunspot Cycles
1.12 Conclusion 68
Chapter 2: Fractal Analysis of Sunspots Cycles and 71
Enso Cycles and Determination of their Relationship
2.1 Data Description and Methodology 72
2.2 Mathematical Models and Fractal Dimension Relations of 76
Sunspots Cycles
2.3 Mathematical Relationship between Higuchi Fractal Dimension 88
and Spectral Exponent of Sunspots Cycles
2.4 Fractal Dimension and Hurst Exponent of Enso Cycles by 88
Using Self - Similar and Self - Affine Fractal Dimension
2.5 Comparative Analysis of Self- Similar and Self-Affine Fractal 95
Dimension of Sunspots Cycle and Active Enso Phenomena in
Same Time Interval
2.6 Conclusion 96
Chapter 3: A Comparative Study of Sunspot Cycles and 98
Enso Cycles Analyzing Probability Distributions and Heavy Tail
3.1 Data Description and Methodology 99
3.2 Probability Distribution Approach 100
3.3 Best Fitted Probability Distributions of Sunspot Cycle 104
3.4 Best Fitted Probability Distributions of Enso Cycle 110
3.5 Comparative Analysis of Best Fitted Distribution of Enso Active 116
Region and Sunspot Cycles in Same Duration
3.6 Probability Distribution of Fractal Dimension of Sunspot Cycles 118
and Enso Cycles
3.7 Solar Activity Cycles and Enso Activity Cycles in the Perspective 119
of Heavy Tails Parameter
8
3.8 Solar Activity Cycles and Enso Activity Cycle in the Perspective 123
Of Long Range Correlation
3.9 Unite Root Stochastic Process of Tail Parameter and Long Range 127
Correlation of Sunspot and Enso Cycles
3.10 Conclusion 129
Chapter 4: A Study of Active Regions of Rotating 131
Sunspots: Image Processing Approach
4.1 Data Description and Methodology 133
4.2 Fractal Dimensions, Wave Spectrum and Heavy Tail 136
Analysis of AR12192
4.3 Mathematical Morphology (MM) 137
4.4 Mathematical Morphology (MM) of AR12192 139
4.5 Image Segmentation of Rotating Sunspots Using Genetic 140
Active Contour
4.6 Active Region AR9114 and AR10696 147
4.7 Conclusion 150
Chapter 5: Conclusions and Future Work 152
5.1 Summary and Outlook 152
5.2 Principle Results 156
5.3 Future Outlook 158
References: 160
9
ACKNOWLEDGEMENT
ALLAH is the most merciful and beneficent, I am extremely thankful to give me the knowledge
and ability to complete this dissertation. Without the opportunities given by Almighty ALLAH I
could not have complete my dissertation successfully.
I express my gratitude to my supervisor Professor Dr. Muhammad Rashid Kamal Ansari and co-
supervisor Professor Dr. Muhammad Jawed Iqbal, for his animated guidance and encouragement.
I am very thankful for teachers, Professor Dr. M. Ayub Khan Yousufzai (University of Karachi),
Professor Dr. Zakaullah Khan (Late), Shuja Muhammad Quraishi (Sir Syed university of
Engineering and Technology). I specially acknowledged to Abdul Qayoom Bhutto (Pakistan
Meteorological Department), Dr. Muhammad Ali Virsani and Professor Dr. Shahid Shukat.
Specially Thanks to Dr. Fozia Hanif Khan, Dr. Frahat Naz Rehman, Talat Salma.
I would like to acknowledge to all my relatives, friends, class fellows, Colleagues, students,
Specially Abdul Moiz Khan, Syed Moutasim Ali, Javed Hameed Khilji, Adil Hameed, Salman
Shahid, Ovais Siraj, Sana waji, Fazal Akber, Bulbul Jan, Arsalan Khanzada, Ghazanfar Ali Khan,
Muhammad Younus, Muhammad Illyas.
Limitless thanks are due to my parents who motivated, encouraged and supported me throughout
the completion of the work. Many thanks to my sisters Dr. Hina Zafar, Saba Zafar and Farah Naz
who remained keen in every stage of completion of this work. In the end I once again thank
Almighty Allah for the completion of this work.
10
ARTICLES APPENDED TO THE THESIS
1. Asma zaffar, Shaheen Abbas and Muhammad Rashid Kamal Ansari (2017). The
Probability Distributions and Fractal Dimension of Sunspots Cycles associated with
ENSO Phenomena. Arabian Journal of Geoscience, Springer (Accepted)
2. Asma zaffar, Shaheen Abbas and Muhammad Rashid Kamal Ansari (2017). Study of
Sunspot cycles using Fractal Dimensions: Wave-Spectrum Scaling. Solar System
Research Springer (Accepted)
3. Asma zaffar, Shaheen Abbas and Muhammad Rashid Kamal Ansari (2017). An Analysis
of Heavy Tail and Long-Range Correlation of Sunspot and ENSO Cycles. Arabian
Journal of Geoscience, Springer (submitted)
4. Asma zaffar, Shaheen Abbas and Muhammad Rashid Kamal Ansari (2017). Model
Estimation and Forecasting of Sunspots Cycles through AR-GARCH Models.
Journal of Forecasting (submitted)
5. Asma zaffar, Shaheen Abbas and Muhammad Rashid Kamal Ansari (2017). A study of
largest Active Region AR12192 of 24th Solar Cycle using Fractal Dimensions and
Mathematical Morphology. Journal of Astronomy and Astrophysics (submitted)
6. Fozia Hanif, Asma zaffar, Shaheen Abbas and Muhammad Rashid Kamal Ansari (2017).
Image Segmentation of Sunspots Using Genetic active contour. Solar System Research
Springer (submitted)
7. Asma zaffar, Shaheen Abbas and Muhammad Rashid Kamal Ansari (2017). Comparison
of Forecasting Evolution of ARMA, AR-GARCH and ARMA-GARCH. Astrophysics
(submitted)
11
ABSTRACT
In this dissertation, sunspot cycles (1-23) analyzed together with the total sunspots observational
data from 1755 to 2008. Similarly, each ENSO cycle (cycle 1 to 21) and the total data from 1866
to 2012. In addition, the data of active period ranging from 1981 to 2000 are evaluated and the
relation between sunspot and ENSO are determined and an analysis is performed. The time series
of sunspot cycles has stationarity and holds linear behavior. ARMA (p, q), AR (p) - GARCH (1,
1) and ARMA (p, q) - GARCH (1, 1) models are analyzed and forecasting, evolution and test of
normality are performed for sunspot cycles. Least square estimator are used for ARMA models.
Whereas the quasi maximum likelihood estimator is used for AR (p) -GARCH (1, 1) and ARMA
(p, q) - GARCH (1, 1) models. AIC, SIC, and HQC are calculated for ARMA (p, q), AR (p)-
GARCH (1, 1) and ARMA (p, q) - GARCH (1, 1) models. Comparison of forecasting evolution,
such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute
Percentage Error (MAPE) and Theil’s U-Statistics test (U test) for ARMA (p, q), AR (p)-GARCH
(1, 1) and ARMA (p, q)-GARCH (1, 1) models is done. Self-similar and self-affine fractal
dimension (FDS and FDA) of the sunspot and ENSO cycles are analyzed. Active region of ENSO
during 1981 to 2000 corresponded to sunspot cycles 21 to 23. This data was partitioned
corresponding to these cycles. Distributions of both the sunspot cycles data and corresponding
ENSO data were analyzed and compared. Wave-spectrum and universal parameters are calculated
for sunspot cycles. The well-known Higuchi’s relation is also analyzed for sunspot cycles. Results
show that sunspot and ENSO cycles are persistent and long term correlation and dependence exists.
It was also concluded that self-similar (box counting method) is more appropriate than self-affine
(rescaled range analysis method). The wave-spectrum parameters such as the spectral exponent
and autocorrelation coefficient have significant range showing that sunspot cycles are strongly
correlated which shows that the sunspot are heavy tailed and have long range correlation.
Correspondingly ENSO cycles are also heavy tailed and long range correlated. This study also
shows that ENSO cycles are more persistent and strongly correlated than Sunspot cycles data. The
adequate probability distributions of sunspots and ENSO cycles are Generalized Pareto
Distribution (GPD) and Generalized Extreme Value Distribution (GEV). These distributions
behave heavy tailed on the right side. Same period of sunspot and ENSO active episode have best
fitted distributions which are also Generalized Pareto Distribution (GPD) and Generalized Extreme
12
Value Distribution (GEV). Moreover, self-similar and self-affine differencing parameter (ds and
dA), heavy tail parameters βS and βA and long range correlation parameters 𝛾𝑆 and 𝛾𝐴 are analyzed
for sunspot and ENSO cycles. The values of Heavy tailed and Long Range correlation parameters
show that the sunspot cycles and ENSO cycles are strongly correlated and each observation is
dependent on the previous observation. The image processing of largest active region AR12192 of
solar cycle 24 is performed using self-similar fractal dimensions (Hausdorff - Besicovich box
dimension and correlation dimension). Mathematical Morphology (MM) of the active region
AR12192 is also studied and analyzed. The persistency of AR12192 is also determined using Hurst
exponent and wave spectrum. The outcomes were analyzed and compared. This analysis
concluded that AR12192 is strongly correlated and spectral exponent and autocorrelation have
significant range. The transformation tools of mathematical morphology such as dilation, erosion,
closing and opening are also analyzed for AR12192. Image segmentation of rotating sunspots
AR9114 and AR10696 of solar cycle23 was performed by using genetic active contours. The
impact of negative and positive helicity and tangential velocity is discussed with the same values
of energy released and temperature.
13
CHAPTER 0
INTRODUCTION
The main source of terrestrial energy is the sun. It also governs the variations in the earth’s climate.
Sun has 1.99 x 1030 kg mass, 6.96 x 108 m radius and 3.86 x 1026 W photon luminosity. The average
temperature of the sun's surface is approximately 5810.928 K. The solar activity like solar flare,
coronal mass ejection and sunspots can be associated with the climate variations in the upper
atmosphere and with the surface of the earth (Anderson, 1990; Gray et al., 2010). The Sun spins
and rotates as a globe of gas not rotate like a solid. The Sun spins faster at its equator as compared
to its poles. The Sun rotates once every 27 days at its equator, but only once every 31 days at its
poles. This can be observed by watching the motion of sunspots and other solar features moving
with the Sun. The giant gas planets, Jupiter, Saturn, Uranus, and Neptune, also spin faster at their
equators than at their poles.
0.1: BIRTH OF SUNSPOTS
Different layers of the sun spin at different rates, creating a magnetic field for the solar sphere.
Convection currents create local magnetic fields in hot gas bubbles. Larger local magnetic fields
and bubbles rise to the surface. At the surface, north and south polarity is split into pairs of
disturbances. Large pairs usually create sunspots. Large sunspot groups often create flares and
mass coronal ejections.
0.2: SUNSPOTS
Solar activity is demonstrated by the dark spots on the surface of the Sun called Sunspots. The
number of Sunspots on the solar disc is considered to be a measure of solar activity. The yearly
average of sunspot areas has been recorded since 1700 (Kok K. et al, 2001). Number of Sunspots
have been recorded reliably since 1755. Sunspots are the dark spots that appear on the photosphere
because of the lower temperature there as compared to the remaining surface. The depth of the
photosphere is about 400 km and its surface reveal most of the solar radiation. Its inner and outer
boundary layer have 6,000 degrees K and 4,200 K respectively. The temperature of the sunspots
14
is about 4,600 K. Sunspot has an intense magnetic field. The central temperature of a sunspot is
3810.928 K. Galileo first made the drawings of sunspots in 1612 (TJO News, 2006). Sunspots are
associated with active regions. The active regions are areas of locally increased magnetic flux of
the Sun (Kevin R. M. et. al, 2014). Usually the sun’s surface is heated to the orange-white stage,
but the sunspot being cooler appear dark. The deeper regions of sunspots are opaque. The higher
and abundant less dense corona is only one-million as bright as the photosphere in light visible
(Stenning D et al. 2010). Normally sunspots appear in pairs in which one has a north magnetic
polarity and the other has a south magnetic polarity. The line joining the centers of these two spots
is nearly parallel to the sun’s equator (Choudhuri, 2008). Sunspots often appear near 30-35 degree
north and south of the sun’s hemisphere with higher latitudes. The sunspots life consists of either
days or one week or few weeks (Taylor, 1991). The number of sunspots always goes on increasing
and decreasing over time showing a cycles high to low and from low to high again. The number
of sunspots is calculated by the sum of individual sunspots and ten times the number of groups.
Since 2005 there is an active region in the sun that contains dozens of sunspots, including one of
the size of 17 earths. This spot is known as Benevolent Monster (Weiss, 2006).
Figure. 0.1 The internal and external structure of sunspots.
0.3: GROUP OF SUNSPOTS OR ACTIVE REGION
Sunspots appear individually and in groups. Sunspot groups are manually classified according to
a number of schemes (Stenning D et al. 2010). The area of sunspot group is related to magnetic
15
fields (E. Forg´acs-Dajka et al 2014). The sunspots growth rate has a linear relation to the
maximum area. The growth rate consists of two parts. First one is leading part and another one is
following part. Leading part is greater than the following part, but the cycle complete of following
parts as one day earlier to the following part. Leading sunspot has southern hemisphere and
following hemisphere has northern hemisphere (Muraközy J et al. 2012 and 2013). Leading
sunspots of all groups are situated closer to the solar equator as compared to the following sunspots
(Baranyi T 2001). Sunspots consist of four parameters viz. Starting point, amplitude, rising time
and asymmetry (between the rise to a maximum and the fall to a minimum is found to be very little
form cycle to cycle and can be fixed at a single). Over the solar sphere, sunspots do not exist
randomly, but develops in groups in the active regions. Active regions are developed by flares and
exist in east-west direction. Sunspots are visual indicator of active regions. Early observation of
magnetic field shows that the active regions of the sun are represented by a bipolar pattern
comprising of both leading and following sunspots. These two sunspots have opposite magnetic
polarities have southern hemisphere and northern hemisphere. Sunspots area plays an initial role
in the sunspot groups and variation in solar irradiance. Systemic and random errors may affect the
result of calculating the area of sunspots (Baranyi T et al 2001). A solar flare is a rapid and intense
variation of brightness and energy. Sunspots and active regions are mostly related with flare (Lin
et al 2005, Zirin and Ligget 1987, Shi and Wang 1994). The image below indicates complex and
numerous sunspot group.
Figure. 0.2 The Groups of sunspots.
16
Solar flares, solar storms and coronal mass ejection blast forth from active regions. When magnetic
energy is created in the solar atmosphere and near the sunspots is released, then solar flare occurs
on the surface of the sun. These energies are more than 1000 times stronger than the average
magnetic field of the sun. Though sunspots existing in active regions, but not all active regions
produce sunspots. In the same active region, different spots have opposite magnetic polarization.
Sunspots have a physical connection with X-rays in the active region. In the active region X-rays
exist in the corona just above a sunspot. X-rays and sunspots both have same magnetic field. The
active regions appear bright in the corona.
0.4: SUNSPOTS CYCLES
The solar magnetic activity cycle (solar cycle) is approximately periodic 11-year. The solar cycle
depends on the changes of the activity of the sun, solar material ejection and the level of solar
radiations. The appearance of solar cycle depends on the changes of sunspot numbers, flares and
other manifestation. For each cycle the maximum number of sunspots is known as solar maximum
and minimum number of sunspots as solar minimum. The first solar cycle is considered to be
starting from 1755 though, the solar cycles were first discovered by Samuel Heinrich Schwabe in
1843. The current solar cycle is the 24th cycle. This solar cycle was started from January 08, 2008.
The initiating pair of sunspots was observed at high latitudes. The maximum number of sunspots
in cycle 24 is 90 though it has some large sunspots too. Solar maximum existed in May 2014.There
were no sunspots in the duration of 2008 and 2009. The situation is very unusual in almost a
century. In 24 solar cycle, northern hemisphere has following sunspots of active regions with
negative polarity field while southern hemisphere has leading sunspots with positive polarity. It is
important to note that each cycle has different duration. Average duration of sunspot cycles is
slightly greater than 11 years. In each cycle the number of sunspots vary from maximum to
minimum and again back to maximum. Cycle 1 consists of 11.3 years, cycle 2 has 9 years, cycle
3 (9.3 years), cycle 4 (13.7 years), cycle 5 (12.6 years), cycle 6 (12.4 years), cycle 7 (10.5 years),
cycle 8 (9.8 years), cycle 9 (12.4 years), cycle 10 (11.3 years), cycle 11 (11.8 years), cycle 12 (11.3
years), cycle 13 (11.9 years), cycle 14 (11.5 years), cycle 15 (10 years), cycle 16 (10.1 years),
cycle 17 (10.4 years), cycle 18 (10.2 years), cycle 19 (10.5 years), cycle 20 (11.7 years), cycle 21
(10.3 years), cycle 22 (9.7 years), cycle 23 (11.7 years) and cycle 24 is going on which is start
from January 2008 and till 2018.
17
Figure. 0.3 The sunspot cycles indicate reversal of magnetic polarity during the solar cycle.
0.5: SOLAR-CLIMATE RELATIONSHIP
Solar activity plays an important role in the variation of global temperature. The solar-climate
relationship has established that the lower temperatures are associated with below average sunspot
activity. While the higher temperatures are associated with above average sunspot activity. It is
not easy to describe the direct relationship between the sunspot cycle and climate, but some events
of climate coincide with sunspot numbers. For instance, in the duration of 1645 to 1715, the
observation of sunspots was dramatically lower than previous one. This duration is called the
Maunder Minimum. The output of the reduction in solar energy in the minimum duration could
have a slight cooling of the Earth. The thin layer of gases, which surrounds the globe of the earth’s
atmosphere protects and sustains life on earth. The atmosphere is frequently in motion and sun
releases radiative energy. On the earth’s surface, the amount of heat is different between the
equator and the poles. The Earth also rotates itself. The atmosphere consists of moisture content,
temperature, airflow and pressure, which is usually described in terms of climate and weather.
Weather describes the local state of the atmosphere in the duration of time periods ranging from
minutes to days. Weather phenomena, such as heat-waves, rainfall, gales, snowfall and frost, these
are effective in daily life. Weather system brings uncertainties in the atmosphere from time to time.
The hotness of the environment are frequently increased and affected over the Cryosphere (e.g.
18
Ice caps, glaciers, etc.) and precipitation in numerous areas across the globe. The climate is
changing over frequently in time durations. The climate change can be expressed as a change in
the average climate from one period averaging to the next. There is no pattern in climate change,
some changes can be large and dramatic, as is clearly shown from reconstructions of temperature
changes between the ice ages and warmer interglacial periods.
Fig. 0.4 Curtsey by (David. Hath away @ NASA.GOV)
0.6: ENSO- SUNSPOTS RELATIONSHIP
The influence of the earth climatic condition of oscillations of solar activity is measurable only in
the long run duration. The Solar cycles (solar activity) and ENSO episode are correlated with each
other. Theory describes the relationship between sunspots and ENSO phenomena is premature,
but now is established by a collection of evidence that the solar cycle moderates wind field in the
stratosphere and troposphere (Anderson 1992, tnislay et.al, 1991) which reveal a potential
mechanism for the link between ENSO phenomena and the solar geometric effect. The change in
sunspot number can describe the association in which ENSO episode in more active when solar
activity is weak (Anderson 1990). The ENSO events can be reduced by long term solar changes
like Mander minimum, medieval grand maximum, similar solar activity is strong then ENSO
period is weak. For instance, the medieval warm period, solar activity is increased (Damon, 1988
19
and Anderson 1990). Enso phenomena effects on climate and weather. The strong ENSO episode
brings droughts in the Southeast Asia countries and in the USA these events make a reason to
bring heavy rain and the number of hurricanes and typhoons are increased in the eastern pacific.
The relationship between rainfall and ENSO in different world areas vary from area to area
(Bhalme et. al. 1984). The impact of ENSO of Indian Ocean varies from season to season (Singh
et. al. 1999). In the early nineties climatologist have intent to understand the relation of the Indian
monsoon with the ENSO is ordered to estimate the monsoon rainfall over Asia (wabster et.al 1992)
describe the review the historical background the relation of monsoon and ENSO some qualitative
studies on the coupling of ENSO year and rainfall over Pakistan (Arif et al. 1994; Chaudhary,
1998).
0.7: OUR APPROACH
The purpose of this study is to analyze the behavior and the relationship of sunspots data and to
study the relationship of each cycle with the other cycles. The stationarity of Sunspot cycles is
determined using second differences with the help of correlogram test which studies
autocorrelation function, partial autocorrelation function and Ljung-Q statistics. The stationarity
of sunspot cycles is verified by using unit root test with Augmented Dickey Fuller (ADF) test.
Sunspot data has a stationary nature indicating the existence of a white noise together with the
linear component. Descriptive study of sunspot cycles rejects the existence of normality. The
normality test is based on skewness, kurtosis and Jurque-Bera test. All sunspots cycles are
positively skewed (right tail) and leptokurtic (high peakness). Different appropriate ARMA (p, q)
models are selected by analyzing the least value of the Durbin-Waston test (DW). Least square
estimator is used to estimate the ARMA process. AIC, SIC, HQC and Log likelihood are also
investigated for each model. RMSE, MAE, MAPE and U test are used to predict the sunspots in
each cycle. One of the novelties of this study is to develop a suitable GARCH model for sunspots
data with appropriate ARMA model specifications. The Lagrange Multiplier test is used to detect
the presence of Autoregressive Conditional Heteroscedastic (ARCH) effect on sunspots data.
Moreover GARCH (1, 1) specification with AR (p) process and ARMA (p, q) - GARCH (1, 1)
model are also investigated in this study. Finally, a comparison of forecasting evolution of these
models is discussed. The other novelty of this study is to analyze the complexity of sunspots cycles
and ENSO cycles using fractal approach. Both the self-similar (Hausdorff – Besicovich method)
20
and self-affine (rescaled range analysis method) FDS and FDA approaches are used. Hurst
exponents exhibit that each sunspot cycles and ENSO cycles is persistent and long term
predictable. Universal scaling laws are used to explore the complexity of sunspots data. These laws
involve scaling exponents (βS, βA), spectral exponent (αs, αA) and autocorrelation coefficient (CA∇
and CS∇). It is concluded that self-similar approach is more appropriate than the self-affine
approach. The behavior of sunspot cycles and ENSO cycles is determined with the help of most
adequate probability distributions. The probability distributions are very helpful to understand the
cyclic behavior in future for the time series used.
It is found that Generalized Prato distribution and Generalized Mean value distribution are the
most appropriate probability distributions for sunspot cycles and ENSO cycles respectively. It is
to be noted that both these distributions are heavy tailed and long range predictable. Heavy tailed
analysis and long range correlation show that the time series data are strongly correlated and long
term predictable. Performing heavy tail and long range correlation analyses using self-similar and
self-affine techniques also confirmed that the time series data for sunspot cycles and ENSO cycles
are strongly correlated and long term predictable. Heavy tail analysis and long range correlation
indicated stationarity and linear trend. Again, the unit root test is used for verification of the results.
One of the objectives of this study is to analyze the images of solar active regions. For this purpose
the largest active region of 24th cycle viz. 12192AR is selected. This analysis is performed by using
fractal approaches and mathematical morphology. Fractal dimension are calculated by Hausdorff–
Besicovich method or box counting method and correlation dimension method. Universal scaling
laws, Hurst exponent, spectral exponent and autocorrelation coefficient are used for evaluation.
The basic principle of mathematical morphology (Dilation, Erosion, opening and closing) is used
to study the active region. Image segmentation of rotating sunspots is performed to understand the
active contour for edge detection for AR9114 and AR10696 of solar cycle 23rd are used to
understand the active contour with genetic algorithm. .
0.8: THESIS OUTLOOK
This dissertation exhibits the introduction of a brief review of the formation of sunspots, group of
sunspots, the active regions of sunspot and the role of sunspots in climate variability on the earth.
Solar Spots-ENSO relationship is also discussed.
21
This dissertation consists of 4 chapters. Chapter zero consists of an introduction and the last chapter
5 is devoted to conclusion and results.
In chapter 1 is concerned with the modeling and forecasting of sunspot cycles. First of all the data
is checked for stationarity. Then, various ARMA (p, q) models are developed with the particular
focus on least value of Durbin-Waston statistics test. Moreover, GARCH (1, 1) model specified
with AR (p) process and GARCH (1, 1) model specified by ARM A (p, q) process are analyzed
for each sunspot cycle. Forecasted values of ARMA (p, q), AR (p) - GARCH (1, 1) and ARM A
(p, q) - GARCH (1, 1) are also compared with RMSE, MAE, MAPE and U test.
Chapter 2 discusses the regular-chaotic and irregular-chaotic behavior in view of the fractal
dimension (FD) (calculated with both self-similar FDS Hausdorff–Besicovich and self-affine FDA
Rescaled Range analysis methods) for Sunspots and ENSO cycles. Higuchi’s Algorithm is also
applied to calculate the fractal dimension FDH for sunspot cycles. Hurst exponents (HES and HEA)
describe the persistency of each cycle. Persistency is calculated to know the correlation and long
term behavior of the time series data. Universal scaling parameter like a scaling exponent is also
evaluated for the sunspots cycles. Wave-spectrum relation like the spectral exponent (αs, αA) and
autocorrelation coefficient (CA∇ and CS∇) for sunspots cycles are also analyzed. Theoretical fractal
scale instruments are developed to find a relationship between Sunspot cycles and ENSO cycles
regarding a long term relationship. The comparison of the two methods shows that self-similar
fractal dimension is more appropriate than the self-affine fractal dimension.
Chapter 3 studies the best fitted and most adequate probability distributions of sunspot and ENSO
cycles. In this connection significant probability distribution is determined for 24 sunspot cycles
in which 24th cycle is still to complete ENSO data against cycles (1-23) is also studied accordingly.
Generalized Pareto and Generalized Extreme Value distribution are found to be most significant
for sunspots and ENSO cycles respectively. These distributions show long term persistence and
heavy tail behavior. The modeling of heavy tail and long term correlation of sunspots and ENSO
cycles are analyzed for the confirmation of heavy tail behavior. Unit root test is applied for
confirmation to linear and stationary behavior.
22
Chapter 4 consists of image analyses of the largest active region of 24th cycle viz12192AR. Fractal
technique are used to analyze the active regions. Both the box counting and correlation dimension
methods are used to evaluate fractal dimensions. Hurst exponent are analyzed to show the
persistency of active regions. Scaling parameters (scaling exponent, spectral exponent and
autocorrelation coefficient) are found to evaluate the active region. Mathematical morphology is
used to study the solar active regions. Image segmentation of rotating sunspot of active region
AR9114 and AR10696 of solar cycle 23 using genetic active contour. Helicity, Temperature,
tangential velocity and energy released are used as a parameter to determine active contour.
Chapter 5 is comprised of the conclusion and principal results of the dissertation along with the
future outlook.
Contents of chapters 1, 2, 3, and 4 are submitted for publication. Each chapter exhibits different
significant research study with consequent introduction, methodology, results and discussion.
23
CHAPTER 1
GARCH MODEL FOR SUNSPOT CYCLES
Sunspot cycles demonstrate solar activity. Solar activity is a feature of the Sun which is not
properly distributed in time. This fact can be observed from the variability of the number of
sunspots with respect to time. Usually they follow the 11 year’s cycle with a minimum and a
maximum (Schwabe, 1844). Schwabe was the first to explain an irregular behavior and
fluctuations of sunspot cycles as well as individual’s maximum intensity and shapes. The
maximum and minimum number of sunspots varies from cycle to cycle. For example a very few
sunspots were observed during the maunder minimum (1645 to 1715) (Thomas et. al, 2008).
Maunder minimum is considered to impose harmful effects on the earth. Long term variations in
solar activity bring a wide change in the climate of the earth. Whereas, short term variations may
cause fluctuations of certain meteorological parameters. In each 11-year cycle, the polarity of the
sun’s magnetic field reverses (Hale et. al, 1919). Cycles vary both in length and size. One of the
successful models to explain the development of sunspots is the model developed by Babcock
(Babcock 1961). Sunspots have an impact on earth’s climate and rainfall. The solar-climate
relationship has established that the lower temperatures are associated with below average sunspot
activity while the higher temperatures are associated with above average sunspot activity.
In chapter 1, describe stochastic autoregression and moving average (ARMA) process for
modeling time series stationary sunspot cycles mean monthly sunspot numbers (1-23 and 24 cycle
consists of a complete observation range from 1st to 23rd) and ENSO cycles (1-21, 22 cycle
complete data from 1 to 23 and 23 cycle belong to an active episode from 1981 to 2000). The
novelty of this chapter is analyzed the stationary generalized autoregressive conditional
heteroskedasticity GARCH (1, 1) stochastic volatility model with specification autoregressive AR
(p) process are used to estimate and forecast the sunspot cycles. GARCH (1, 1) specification with
ARMA (p, q) process also are used to estimate and forecast the sunspot cycles. Sunspot cycles are
forecasted with the help of a Root mean square error (RMSE), Mean Absolute Error (MAE) and
Mean Absolute Percentage Error (MAPE). The Gaussian quasi maximum likelihood estimation is
used to analysis these processes. The adequacy of each model depends on Akaike information
criterion (AIC), Bayesian Schwarz information criterion (BIC) and Hannan Quinn information
criterion (HIC). Best fitted model of sunspot cycles is chosen with compression of these techniques
24
with the help of Root mean square error (RMSE), Mean Absolute Error (MAE) and Mean Absolute
Percentage Error (MAPE).
1.1: DATA DESCRIPTION AND METHODOLOGY:
Time series sunspot cycle’s data are stationary at second difference. This study describes stochastic
autoregression and moving average (ARMA) process for time series modeling. Stationary sunspot
cycles mean monthly sunspot numbers (1-23 and 24 cycle consists of a complete observation range
from 1st to 23rd). The novelty of this chapter is analyzed the stationary generalized autoregressive
conditional heteroskedasticity GARCH (1, 1) stochastic volatility model with specification
autoregressive AR (p) process are used to estimate and forecast the sunspot cycles. GARCH (1, 1)
specification with ARMA (p, q) process also are used to estimate and forecast the sunspot cycles.
Sunspot cycles are forecasted with the help of a Root mean square error (RMSE), Mean Absolute
Error (MAE) and Mean Absolute Percentage Error (MAPE). The Gaussian quasi maximum
likelihood estimation is used to analysis these processes. The adequacy of each model depends on
Akaike information criterion (AIC), Bayesian Schwarz information criterion (BIC) and Hannan
Quinn information criterion (HIC). Best fitted model of sunspot cycles is chosen by diagnostic
checking with compression of these techniques with the help of Root mean square error (RMSE),
Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) and Theil’s U-
Statistics.
Sunspot cycles have an appropriate process by selection criteria based on the least value of the
Durbin-Waston stats (DW) are ARMA (2, 2), ARMA (2, 6), ARMA (3, 2), ARMA (3, 3), ARMA
(3, 4), ARMA (4, 3), ARMA (4, 4), ARMA (4, 5), ARMA (4, 6), ARMA (5, 2), ARMA (5, 3),
ARMA (5, 4), ARMA (5, 5), ARMA (5, 6), ARMA (6, 2), ARMA (6, 3), ARMA (6, 4), ARMA
(6, 5), ARMA (6, 6). Moreover, most appropriate model is ARMA (6, 4) process which is
frequently used in sunspot cycles. ARMA (5, 3), ARMA (5, 4), ARMA (3, 3), ARMA (4, 4),
ARMA (6, 5), ARMA (6, 2), these models are maximum use in sunspot cycles. Sunspot cycles
have appropriate GRACH (1, 1) process specification with AR (p) are estimation and forecast.
Sunspot cycles follow AR (2) - GARCH (1, 1) process and AR (3) - GARCH (1, 1). In ARAM (p,
q) - GARCH (1, 1) follows ARAM (1, 1) - GARCH (1, 1), ARAM (2, 2) - GARCH (1, 1), ARAM
(3, 2) - GARCH (1, 1), ARAM (3, 3) - GARCH (1, 1), ARAM (4, 2) - GARCH (1, 1), ARAM (4,
25
4) - GARCH (1, 1), ARAM (5, 1) - GARCH (1, 1), ARAM (5, 3) - GARCH (1, 1) and ARAM (6,
1) - GARCH (1, 1).
1.2: TIME SERIES ANALYSIS WITH STATIONARY PROCESS
Time series are very helpful in many disciplines of the solar physics, climate change and so on.
The stochastic component remains, periodicities and trend have been excluded from a time series.
The large range correlation in the context of the solar and terrestrial physics. Many time series in
the solar physics exhibits persistence, where successive value is positively correlated. Big values
trend to follow big whereas small observations followed the small trend. The correlation is the
statistical dependence of distantly and directly neighbor values.
If the data set is larger than time series analysis is used. Mathematically time series is described as
a series of vector Xt; t = 0, 1, 2, ….. Where t indicate the time elapsed (Raicharoen et. al. 2003,
Hipal et. al. 1994 and John et. al. 1997). Variable Xt is a random variable. Time series analysis
based on various goals. Descriptive is defined as identify the behavior trend of correlated data.
Explanation expresses as modeling and understanding the data. Forecasting explore the predication
the trend either long-term or short-term from previous patterns. Intervention analysis is the changes
in a single event with respect to time. Quality control defines the problem indicate specified size
of deviations.
In a time series, stationery series has over time constant mean, variance, autocorrelation and so on.
Consequently, in time shifting joint probability distribution does not change. Compulsory
condition for developing time series models that are better future forecasting. Moreover, the
mathematical complexity of the best fitted model reduces. Stationary process is strong as well as
a week. Strong stationary process Xt is defined as,
if Xt, Xt + 1, Xt + 2 , … , Xt + n and Xt + S, Xt + 1 + S, Xt + 2 + S , … , Xt + n + S
have some joint distribution of all t and n where s > 0.
Weak stationary explain as, if mean value E (Xt) is time independent and Cov (Xt, XS) depends on
the time difference |𝑡 − 𝑆| (time lag). The figure 1.1: depict the time series of monthly average
sunspot number from cycle 1st to cycle 23rd is plotted.
26
Figure 1.1: The monthly average sunspot number from cycle 1 to 23
1.3: AUTOCORRELATION FUNCTION, PARTIAL AUTOCORRELATION FUNCTION
AND LJUNG-BOX Q-STATISTICS OF SUNSPOT CYCLES
These functions are very useful in order to verify a time series model: Autocorrelation function
(ACF), Partial Autocorrelation function (PACF) and Ljung-Box Q-statistics (Reale 1998).
The autocorrelation function (AC) is used to analysis the linear predictability of the data set Xt.
ℵ(𝑠) = 𝐶𝑜𝑣 (𝑋𝑡,𝑋𝑡−𝑠 )
√𝑉𝑎𝑟(𝑋𝑡)𝑉𝑎𝑟(𝑋𝑡−𝑠) −1 ≤ ℵ(𝑠) ≤ 1 (1.1)
For stationary Var (Xt) = Var (Xt-s). Samples of AC are called a Correlogram.
Partial Autocorrelation function (PAC) defines the correlation between Xt and Xt-s after omit the
effect of the intervening variables Xt, Xt + 1, Xt + 2 , … , Xt – s-1 denoted by 𝜇𝑠𝑠.
𝜇𝑠𝑠 = ℵ(𝑠)− ∑ 𝜇𝑠−1,𝑗 ℵ𝑠−𝑗
𝑠−1𝑗=1
1− ∑ 𝜇𝑠−1,𝑗 ℵ𝑠−𝑗ℵ𝑠−1𝑗=1
(1.2)
Partial Autocorrelation function (PAC) is very helpful to identify the AC and PAC are used to
identify linear time series model.
Ljung-Box Q-statistics are used to analysis the effect of autocorrelation at lags.
Ljung-Box Q-statistics = N (N+2) ∑ℵ(𝑠)2
𝑁−𝑠
𝑠𝑗=1 (1.3)
Where ℵ(s) is the s-th autocorrelation and N is the number of observations. Ljung-Box Q-statistics
are also used as a test of white noise series.
27
Sunspot cycles have stationary in second difference. Differencing is the transformation which is
converted non-stationary the time series data into stationary. Table 1.1 depicts that Autocorrelation
(AC), Partial Autocorrelation (PAC) and Ljung-Box Q-statistics test of sunspot cycles. Each
sunspot cycle has a white noise in second difference. White noise is purely random and simple
stationary process in which sequence {∈𝑡} of uncorrelated random variables with ∈𝑡 = WN (0,𝜎2).
White noise is very useful to construct many processes. Probability is significant in each lag.
Figure 1.2 describe that ACF and PCF of each sunspot cycle (1-23) and 24 cycle based on complete
data set from 1st to 23rd cycles.
28
29
30
31
32
33
34
35
Figure 1.2: ACF and PACF of Sunspot Cycles (1-23, 24)
36
1.4: UNIT ROOT TEST OF SUNSPOTS CYCLES BASED ON AKAIKE INFORMATION
CRITERION (AIC)
The unit root test can be used for some stochastic process which cause problem is statistic inference
involving in time series. The random walk model
Yt = ρYt-1 + ut −1 ≤ ρ ≤ 1 (1.4)
In this model Yt belongs to the time series and ut represent the white noise error while ρ is the
coefficient. The equation (6) is used to regress the Yt on the lagged value Yt-1estimate for unit root
test. If the predictable value ρ identical to 1 statistically. The series comprises unit root and null
hypothesis rejected (McCarthy, 2015). Transform the equation 6 by subtraction Yt-1 both sides
Yt−Yt -1 = ρYt -1− Yt -1 + ut
Yt−Yt -1 = Yt -1(ρ− 1) + ut
∆ Yt = ɸYt -1 + ut (1.5)
Where ∆ and ɸ represent the operation of the first difference. The null hypothesis (ɸ = 0) for time
series non stationary condition represents the unit root (da silve Lopes, 2006).
The Augmented Dickey Fuller (ADF) test to solve the autocorrelation problem. This test is
implemented by augmenting three following equations, the dependent variable’s (∆Yt) adding
lagged values and Yt represents a random walk with
∆ Yt = 𝛾1+ 𝛾2t + ɸ Yt -1 + ut (1.6)
Where 𝛾2t and 𝛾1represents the intercept and trend respectively.
∆ Yt = ɸYt -1 + ut (1.7)
It shows neither intercept nor trend.
∆ Yt = 𝛾1+ ɸ Yt -1 + ut (1.8)
It shows intercept (𝛾1) only (Kwiatkowski et al., 1992; McCarthy, 2015).
We assume the null hypothesis H0: variables have a unit root (non-stationary data). The condition
where the H0 is rejected if p-value is less than 5%. The critical value of the absolute value of
Augmented Dickey Fuller (ADF) test is greater than at 1% and 5% significance level.
The Augmented Dickey Fuller (ADF) test for stationary explored that the sunspot cycles 1-23
(1755 to 2008) and complete from 1 to 23 cycle’s time series data. In case the data is not stationary
than behave like a unit root. Table 1.2 depicted that sunspot cycles has stationary nature and unit
root test rejected. The reason to reject H0 is p-value is less than 5% or the critical value of the
37
absolute value of Augmented Dickey Fuller (ADF) test is greater than at 1% and 5% significance
level. The stationary of sunspot cycles are calculated at second difference with based on Akaike
Information Criterion (AIC). Sunspot cycle (1-24) reject unit root test with Augmented Dickey
Fuller (ADF) test values -6.181642, -6.091491, -6.893381, -7.579517, -6.823644, -6.875331, -
6.695780, -8.558407, -6.745618, -6.933791, -5.950719, - 6.602668, -7.700490, -8.106940, -
7.087756, -10.74242, -6.524422, -7.167179, -7.122662, -6.852907, -7.960910, -7.082791, -
7.048743, -17.69189 respectively. Table 1.3 described that
38
39
40
41
Figure 1.4: Stationary sunspot cycle (1-24)
1.5: TESTS FOR NORMALITY
Normality test is very useful to determine a normally distribution of well modeled data set. Test
based on analysis the two numerical measures, shapes skewness and excess kurtosis. The data sets
are normally distributed if values are close to zero. The acceptance of Jurque-Bera test also focused
on skewness and kurtosis. Test of normality consists of skewness, kurtosis and Jurque-Bera test.
Skewness measures the degree of asymmetry of the data.
Skewness = ∑ (𝑋𝑖−�̅�)3𝑛
𝑖=1
(𝑛−1)𝑆3 (1.9)
Where �̅� is the mean, S is standard deviation and n is the number of values (Cryer et. al. 2008).
Data is normally distributed if the skewness value is equal to zero. Positive skewed (right tail) if
greater than zero and negative skewed (left tail) with less than zero.
Kurtosis measures the degree of peakness of the data. Kurtosis has been estimated as
Kurtosis = ∑ (𝑋𝑖−�̅�)4𝑛
𝑖=1
(𝑛−1)𝑆4 (1.10)
Where �̅� is the mean, S is standard deviation and n is the number of values of the time series data.
Kurtosis of normal distribution is called mesokurtic if it is equal to 3. Whereas, leptokurtic if the
value is greater than 3. Platokurtic if the value is less than 3.
Jurque-Bera Statistics Test is accepted with normality of the data with skewness is equal to zero
and excess kurtosis is also equal to zero. Jurque-Bera test as follows
Jurque-Bera test = 𝑛(𝑆𝑘𝑒𝑤𝑛𝑒𝑠𝑠)2
6 +
𝑛(𝐾𝑢𝑟𝑡𝑜𝑠𝑖𝑠)2
24 (1.11)
42
Jurque-Bera test statistics is approximately as Chi-squared distribution with two degrees of
freedom. Null hypothesis (HO) is a normal distribution with skewness is zero and excess kurtosis
is zero (which is same as a kurtosis is 3). Alternate hypothesis (HA) gives data is not normally
distributed.
1.5.1: Descriptive Statistics and Normality Test of Sunspot Cycles
Table 3 demonstrates the descriptive statistics of each Sunspot cycle. Mean, median, standard
deviation, skewness, kurtosis and Jurque-Bera test have been calculated of each cycle. 1st cycle
has 128 sunspot numbers with 43.81± 24.35. 2nd cycle follows 144 sunspot observations with
57.04± 39.21. 3rd cycle expresses 107 sunspot observations with 68.95± 52.12. 4th cycle represents
169 sunspot observations with 59.92 ± 54.44. 5th cycle explores 148 sunspot observations with
2305 ± 18.71. 6th cycle has 160 sunspot observations with 17.76 ± 17.98. 7th cycle follows 119
sunspot observations with 40.06 ± 26.01. 8th cycle reveals 120 sunspot observations with 65.39 ±
48.31. 9th cycle represents 139 sunspot observations with 59.37 ± 38.41. 10th cycle explores 144
sunspot observations with 45.86 ± 32.57. 11th cycle expresses 140 sunspot observations with 53.59
± 46.81. 12th cycle has 142 sunspot observations with 32.55 ± 27.15. 13th cycle represents 148
sunspot observations with 37.54 ± 30.63. 14th cycle follows 136 sunspot observations with 32.68
± 26.33. 15th cycle expresses 114 sunspot observations with 46.53 ± 33.97. 16th cycle signifies
125 sunspot observations with 39.79 ± 28.73. 17th cycle follows 125 sunspot observations with
57.64 ± 40.52. 18th cycle has 122 sunspot observations with 74.43 ± 54.71. 19th cycle expresses
129 sunspot observation with 89.04 ± 71.52. 20th cycle represents 140 sunspot observations with
60.26 ± 37.29. 21th cycle explores 119 sunspot observations with 83.57 ± 57.22. 22th cycle has
124 sunspot observations with 76.15 ± 58.50. 23th cycle follows 140 sunspots observations with
57.94 ± 42.67. 24th cycle reveals 3030 sunspot observations with 52.26 ± 44.60. Each sunspot
cycle shows that positive skewed (right tail). Whereas different cycles have kurtosis behavior.
Sunspots cycle 15 explore that approximate mesokurtic (3.021) normally distributed. Moreover,
maximum sunspot cycles follow platokurtic which represents that flat tail (peakness). Sunspot
cycles 6th, 9th and 24th evaluate leptokurtic (heavy tail). In Jurque-Bera test, each sunspot cycles
rejected the null hypothesis (sunspot cycles are normally distributed by any mean and variance).
43
1.6: TEST OF ADEQUACY
Selection for the appropriate and adequate process of ARMA (p, q), AR (p) – GARCH (1, 1) and
ARMA (p, q) – GARCH (1, 1) for accurate and significant test are very important. Test of
adequacy consists of Durbin- Watson (DW) statistics test, Akaike information criterion (AIC),
Bayesian Schwarz information criterion (SIC) and Hannan Quinn information criterion (HQC) and
Log maximum likelihood test. The selection of appropriate sunspot cycles is based on the
minimum value of the Durbin- Watson (DW) statistics test for each method.
Durbin- Watson (DW) statistics is a test for measuring the linear association between adjacent
residual from a regression model. The hypothesis of Durbin- Watson statistics is 𝜏 = 0 is the
specification.
Ut = 𝜏 Ut-1 + ∈𝑡 (1.12)
Durbin-Watson (DW) is equal to 2 shows that no serial correlation. If Durbin- Watson (DW) is
less than 2 indicate that positive correlation and range from 2 to 4 represent that negative
correlation. The series is strongly correlated if the value nearly approaches to zero.
Akaike information criterion used to compare the statistical models set to each other. AIC was
introduced by Hirotogu Akaike in 1973. It is the extension of the maximum likelihood principle.
The selection criterion is focused on the least value of AIC.
AIC = - 2Log (likelihood) + 2S (1.13)
Where S is the model parameter numbers. Likelihood is a measure of fit model. Maximum values
exhibits the best fit.
Schwarz criterion is used to select the best model among finite model. The appropriate model is
based on the least value of SIC. Schwarz criterion (SIC) was developed by Gideon E. Schwarz. It
is closely related to the AIC.
SIC = -2ln (Likelihood) + (S + S ln (N)) (1.14)
Where S is the model parameter numbers. N exhibits the number of observations.
In statistics, the Hannan-Quinn information is the criterion for model selection. HAQ is an
alternative to AIC and SIC.
HQC = -2 Log (Likelihood) + 2 (S + S ln (N)) (1.15)
Where S is the model parameter numbers. N exhibits the number of observations.
44
1.7: DIAGNOSTIC CHECKING TEST
In this study, forecasting performance and evolution of each sunspot cycle is checked by Mean
Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error
(MAPE) and Theil’s U-Statistics. A little description of these techniques is given in the following.
The Mean Absolute Error (MAE) is expressed as
MAE = 1
𝑛 ∑ |∈𝑡|𝑛
𝑡=1 (1.16)
Mean Absolute Error (MAE) measures the absolute deviation of forecasted values from real ones.
It is also known as Mean Absolute Deviation (MAD). It expresses the magnitude of overall error,
happened due to forecasting. MAE do not cancel out the effect of positive and negative errors.
MAE does not express the directions of errors. It should be as small as possible for a good
forecasting. MAE depends on the data transformations and the scale of measurement. MAE are
not the panelized for the extreme forecast error.
The Mean Absolute Percentage Error (MAPE) is defined as
MAPE = 1
𝑛 ∑ |
∈𝑡
𝑋𝑡|𝑛
𝑡=1 × 100 (1.17)
Mean Absolute Percentage Error (MAPE) provides the percentage of the average absolute error.
It is independent of the scale measurement. Whereas, dependent of the data transformations.
MAPE does not locate the direction of Error. Extreme deviation is not penalized by MAPE.
Opposite signed errors does not parallel to each other in MAPE.
The root mean squared error (RMSE) is defined as
RMSE = √1
𝑛∑ ∈𝑡
2𝑛𝑡=1 (1.18)
RMSE calculate the average squared deviation of forecasted values. Opposed sighed error do not
parallel to each other. RMSE provide the complete idea of the error happened during forecasting.
While forecasting, panelizes extreme error happens. In RMSE, the total forecast error is affected
by the large individual error. For instance, large error is more appropriate than small errors. It does
not reveal the direction of overall errors. RMSE is affected to the data transformation and change
of scale. RMSE is a good measure of overall forecast error.
Theil’s U-Statistics is defined as
U = √
1
𝑛∑ ∈𝑡
2𝑛𝑡=1
√1
𝑛∑ 𝑓𝑡
2𝑛𝑡=1 √
1
𝑛∑ 𝑋𝑡
2𝑛𝑡=1
0 ≤ 𝑈 ≤ 1 (1.19)
45
Where ft represent the forecasted value and Xt shows that the actual value. U is the normalized
measure of the total forecast error. U is equal to 0 exhibits the perfect fit.
1.8: ARMA MODEL
A statistical approach to forecasting involves stochastic models to predict the values of sunspot
cycles by using pervious once. In the linear time series, two methods are frequently used in
literature, viz. Autoregressive AR (p) and Moving Average MA (q) (Jenkins et.al. 1970 and Hipal
et. al. 1994). ARMA models are developed by (Jenkins et. al 1994). An ARMA model is the
combination of an idea of Autoregressive AR (p) and Moving Average MA (q) process. The
concept of ARMA process is strongly relevant in volatility modeling. ARMA model is wieldy
used for forecasting the future values. Autoregressive process (AR) is developed by (yule, 1927).
In stochastic process, Autoregressive process AR (p) can be expressed by a weighted sum of its
previous value and a white noise. The generalized Autoregressive process AR (p) of lag p as follow
Xt = α1 Xt-1 + α 2 Xt-2 + … + α p Xt-p + ∈t (1.20)
Here εt is white noise with mean E (∈t) = 0, variance Var (∈t) = σ2 and Cov (∈t -s, ∈t) = 0, if s ≠ 0.
For every t, suppose that 𝜏t is independent of the Xt-1, Xt-2, ….. 𝜏t is uncorrelated with Xs for each s
< t. AR (p) models regress is past values of the data set. Whereas, MA (q) model relates with error
terms as a descriptive variables (Hipal et. al. 1994). The generalized Moving Average process MA
(q) of lag q as follows
Xt = ∈t + β1 ∈t-1 + β2 ∈t-2+ … + βq ∈t-q (1.21)
The process Xt is defined by the ARMA model
Xt = α1 Xt-1 + α 2 Xt-2 + … + α p Xt-p + ∈t + β1 ∈t-1 + β2 ∈t-2+ … + βq ∈t-q (1.22)
With ∈t is an uncorrelated process with mean zero. The prediction of ARMA (p, q) process shows
the decay to be sinusoidally and exponentially to zero.
1.8.1 ARMA Models of Sunspots Cycles
ARMA models are widely used for the prediction of stationary stochastic second order process.
The ARMA model is a tool for analysis and understanding of the causal structures or getting the
predictions of future values in time series. The suitable models for sunspot cycles are selected
based on Durbin-Watson statistics test. Least square estimation is used to calculate ARMA
process. Coefficient of determination, R2 near to one demonstrates that data values in each cycle
depend on each other. Table1.4 depicts 5 best models for each sunspot cycle. Least square
46
estimation is used to calculate ARMA (p, q) model. The selection of models depends on Durbin-
Watson statistic test. The Durbin- Watson (DW) near to zero is describe the series are positively
correlated. The value 2 shows that the series has no correlation. If DW is ranged between 2 and 4
indicates negative correlation. Adequacy of the models is checked by AIC, SIC, HQC and
Maximum log likelihood tests. The forecasting evolution is checked by RMSE, MAE, MAPE and
U tests. Sunspot cycle (Aug 1755 - Mar 1766) shows that best fitted models are ARMA (4,4),
ARMA (4,5), ARMA (5,4), ARMA (5,5) and ARMA (6,5). According to Durbin- Watson (DW)
test ARMA (6, 5) model possesses the least value of DW (0.798) which implies that the 1st cycle
is strongly correlated. ARMA (6, 5) model is also the best model according to log maximum
likelihood (-554.25). ARMA (4, 5) and ARMA (4, 4) models have least value 8.559, 8.648 and
8.595 of AIC, SIC and HQC respectively and show a best fit. Coefficient of determination, R2
demonstrates that data values in each cycle depend on each other. Forecasted evaluation of 1st
cycle describes that RMSE with least value 23.01596 is most favorable for ARMA (5, 4).
According to MAE with the value 18.43576 shows that best fitted model is ARMA (4, 5). MAPE
with the least value 55.73550 shows that ARMA (4, 4) is the best fitted model. U test with least
value 0.251178 describes that ARMA (6, 5) is best fitted. Sunspot cycle (Mar 1766 - Aug 1775)
exhibits that best fitted model are ARMA (2, 2), ARMA (2, 6), ARMA (3, 2), ARMA (3, 3) and
ARMA (5, 3). ARMA (5, 3) process has least value DW (0.719) which is best fitted and strongly
correlated and predictable. Appropriate model according to log maximum likelihood is also
ARMA (5, 3) model with -547.40. ARMA (2, 2) model has least value 9.402, 9.498 and 9.441 of
AIC, SIC and HQC respectively and show a best fit. Forecasted evolution define that RMSE, MSE
and U test with least values 37.92649, 29.55789 and 0.309054 have best fitted model ARMA (5,
3) respectively. MAPE with least value 145.5506 shows that best fitted model is ARMA (2, 2).
Sunspot cycle (Aug 1775 - Jun 1784) exhibits that best fitted model are ARMA (5, 4), ARMA (5,
5), ARMA (5, 6), ARMA (6, 4) and ARMA (6, 5). ARMA (6, 5) model has least value DW (0.846)
and log maximum likelihood with -508.60 which is shown that best fitted and strongly correlated
and predictable. ARMA (5, 4) process has an appropriate model with least value 9.334, 9.434 and
9.375 of AIC SIC and HQC respectively. Forecasted evolution demonstrates that RMSE with
57.08266 has appropriate model is ARMA (6, 5). According to MAE and U test with least value
43.6452 and 0.411658 shows that best fitted model are ARMA (6, 4) respectively. MAPE with
least value 102.4356 shows that the best fitted model is ARMA (5, 5). Sunspot cycle (Jun 1784 -
47
Jun 1798) exhibits that best fitted model are ARMA (3, 3), ARMA (5, 4), ARMA (5, 6), ARMA
(6, 4) and ARMA (6, 5). According to DW and log maximum likelihood with 0.904 and -784.12
shows that ARMA (6, 5) process is best fitted respectively. ARMA (3, 3) process is best fit with
least value 8.496, 8.570 and 8.526 of AIC SIC and HQC respectively. Forecasted evolution define
that RMSE and MAE with 53.38909 and 41.59299 have appropriate model is ARMA (5, 6)
respectively. According to MAPE with 141.12189 define the appropriate model is ARMA (3, 3)
and U test with 0.477880 shows the best fitted model is ARMA (5, 4). Sunspot cycle (Jun 1798 -
Sep 1810) indicate that best fitted models are ARMA (3, 3), ARMA (4, 4), ARMA (5, 3), ARMA
(6, 2) and ARMA (6, 4). ARMA (6, 4) process has least value DW (0.794) which is best fitted and
strongly correlated and predictable. Suitable model according to log maximum likelihood with -
584.79 is ARMA (5, 3) model. AIC SIC and HQC with smallest values 7.278, 7.359 and 7.311
described that most appropriate model is ARMA (3, 3). Forecasted evolution define that RMSE
and MAE with 19.61195 and 16.77907 have appropriate model is ARMA (5, 3). According to
MAPE with 161.6836 shows that the appropriate model is ARMA (3, 3) and U test with 0.416733
describe that the best fitted model is ARMA (6, 4). Sunspot cycle (Sep1810 - Dec 1823) describe
the best fitted models are ARMA (3, 4), ARMA (6, 2), ARMA (6, 4), ARMA (6, 5) and ARMA
(6, 6). ARMA (6, 5) model is most appropriate with the smallest value of DW with 0.794 and AIC
SIC and HQC with 7.944, 8.021 and 7.975 respectively. According to log maximum likelihood
with -636.60 shows that the best model is ARMA (6, 2) model. Forecasted evolution define that
RMSE and MAE with 17.58113 and 13.25282 have appropriate model is ARMA (6, 4). According
to MAPE with 251.9070 appropriate model is ARMA (6, 2) and U test with 0.418201shows that
the best fitted model is ARMA (3, 4). Sunspot cycle (Dec1823 - Oct 1833) shows that best fitted
models are ARMA (4, 4), ARMA (5, 2), ARMA (5, 3), ARMA (5, 5) and ARMA (6, 4). According
to Durbin-Weston (DW) ARMA (5, 5) model is the best model with least value of DW (0.926).
ARMA (6, 4) model is also the best fitted model according to log maximum likelihood (-523.20).
AIC SIC and HQC described that the appropriate model is ARMA (4, 4) with least value 8.628,
8.721 and 8.666 respectively. Forecasted evaluation describes that RMSE and MAE explore
ARMA (5, 3) are favorable with least value 24.84239 and 20.57404 respectively. MAPE has least
value 125.1342 by ARMA (5, 5). U test describes that ARMA (5, 3) is best fitted with least value
0.294667. Sunspot cycle (Oct1833 - Sep 1843) exhibits that best fitted model are ARMA (3, 3),
ARMA (4, 6), ARMA (5, 6), ARMA (6, 2) and ARMA (6, 3). ARMA (5, 6) process is the best
48
fitted model with the smallest value DW (0.767). Appropriate model according to log maximum
likelihood is also ARMA (4, 6) model with -572.46. AIC SIC and HQC described that ARMA (6,
2) process has least value 9.321, 9.413 and 9.358 respectively. Forecasted evolution define that
RMSE, MSE and U test have best fitted model ARMA (4, 6) with 47.89669, 37.64070 and
0.353636 respectively. MAPE shows that best fitted model is ARMA (3, 3) with 101.5025.
Sunspot cycle (Sep1843 - Mar 1855) shows that the best fitted models are ARMA (3, 3), ARMA
(4, 3), ARMA (4, 4), ARMA (5, 3) and ARMA (6, 4). Appropriate model is ARMA (6, 4) with
the least value of Durbin- Watson with 0.779 and log maximum likelihood with -655.49. AIC SIC
and HQC depict that ARMA (3, 3) process is suitable with the smallest values 9.159, 9.243 and
9.193 respectively. Forecasted evolution define that RMSE and U test have best fitted model
ARMA (4, 3) with 36.85102 and 0.296860 respectively. According to the MAE best fitted model
is ARMA (6, 4) with 26.67266. MAPE with 55.33579 shows the best fitted model is ARMA (3,
3). Sunspot cycle (Mar1855 - Feb 1867) depict that best fitted models are ARMA (4, 4), ARMA
(5, 3), ARMA (6, 2), ARMA (6, 4) and ARMA (6, 5). According to DW and log maximum
likelihood with 0.993 and -614.66 shows that ARMA (6, 5) process is suitable. AIC SIC and HQC
with 8.424, 8.486 and 8.449 described that the appropriate model is ARMA (4, 4). Forecasted
evolution demonstrates that RMSE, MAE and U test with 34.05871, 26.32748 and 0.318617 have
appropriate model is ARMA (6, 4). According to MAPE with 149.2235 shows that best model is
ARMA (3, 3). Sunspot cycle (Feb1867 - Sep 1878) indicate that best fitted models are ARMA (3,
3), ARMA (5, 3), ARMA (5, 4), ARMA (6, 4) and ARMA (6, 5). ARMA (6, 5) process is suitable
model with the least value DW (0.680) and log maximum likelihood with -649.72. AIC, SIC and
HQC with smallest value 9.002, 9.086 and 9.036 described that ARMA (3, 3) process is best fitted.
Forecasted evolution define that RMSE and MAE with 51.34360 and 40.82203 have appropriate
model is ARMA (6, 4). According to MAPE with 619.2955 shows that the appropriate model is
ARMA (3, 3) and U test with 0.453341reveal that the best fitted model is ARMA (5, 3). Sunspot
cycle (Sep1878 - Jun 1890) describe the best fitted models are ARMA (4, 4), ARMA (5, 3), ARMA
(5, 4), ARMA (6, 4) and ARMA (6, 6). ARMA (5, 4) process is the suitable model with the least
value of DW with 0.877 and log maximum likelihood with -606.7. AIC, SIC and HQC with
smallest value 8.513, 8.596 and 8.547 show that the appropriate model is ARMA (4, 4). Forecasted
evolution define that RMSE, MAE and U test with 26.94070, 22.33445 and 0.384873 have
appropriate model is ARMA (5, 3). According to MAPE with 317.31740 shows that the
49
appropriate model is ARMA (4, 4). Sunspot cycle (Jun1890 - Sep 1902) reveal that the best fitted
models are ARMA (4, 3), ARMA (4, 4), ARMA (5, 4), ARMA (6, 2) and ARMA (6, 4). ARMA
(6, 4) model has the least value of DW is 1.122 so is an appropriate model according to Durbin-
Watson state. Log maximum likelihood with -627.95 shows best model is ARMA (5, 4). AIC SIC
and HQC with least value 8.356, 8.437 and 8.389 described that best model is ARMA (4, 4).
Forecasted evolution define that RMSE and MAE with 34.15608 and 28.14172 have appropriate
model is ARMA (6, 4). MAPE with the least value 424.3318 shows that the suitable model is
ARMA (6, 2) and U test with 0.418577 is best fitted model is ARMA (5, 4). Sunspot cycle (Sep
1902 - Dec 1913) exhibits that best fitted models are ARMA (4, 4), ARMA (4, 6), ARMA (5, 3),
ARMA (5, 5) and ARMA (5, 6). ARMA (6, 4) model is an appropriate model with least value of
DW is 0.980.Log maximum likelihood with maximum value -602.26 shows the best model is
ARMA (5, 4). AIC SIC and HQC with least value 8.711, 8.797 and 8.756 shows that the best
process is ARMA (4, 4). Forecasted evolution shows that RMSE, MAE and U test with 26.33349,
21.89815 and 0.363862 have appropriate model is ARMA (4, 6). According to MAPE with
447.7918 describe that the best model is ARMA (4, 4). Sunspot cycle (Dec 1913 - May 1923)
indicate that best fitted models are ARMA (4, 3), ARMA (5, 3), ARMA (5, 4), ARMA (5, 5) and
ARMA (6, 4). ARMA (6, 4) process has least value DW with 0.687 is best fitted and strongly
correlated. The best model according to log maximum likelihood with -528.49 is ARMA (5, 5)
model. AIC SIC and HQC with smallest value 9.077, 9.123 and 9.115 described that ARMA (4,
3) process is most fitted. Forecasted evolution of RMSE, MAE and U test with 33.02818, 26.04595
and 0.329163 have appropriate model ARMA is (5, 3). According to MAPE with 124.6412 shows
that the appropriate model is ARMA (5, 5). Sunspot cycle (May 1923 - Sep 1933) describe the
best fitted models are ARMA (3, 3), ARMA (4, 4), ARMA (5, 3), ARMA (6, 4) and ARMA (6,
5). ARMA (6, 5) process appropriate model with smallest DW value (0.565) and maximum Log
likelihood value (-547.01). Suitable model ARMA (3, 3) is best fitted according to AIC SIC and
HQC with 8.545, 8.635 and 8.582 respectively. Forecasted evolution define that RMSE, MAE and
U test with 28.41330, 23.46526 and 0.340449 have appropriate model is ARMA (6, 5). According
to MAPE with 225.7652 define that the appropriate model is ARMA (4, 4). Sunspot cycle (Sep
1933 - Jan 1944) shows that best fitted models are ARMA (3, 3), ARMA (4, 4), ARMA (5, 3),
ARMA (6, 2) and ARMA (6, 4). ARMA (6, 4) model is an appropriate model with least DW value
(0.583) and maximum Log likelihood value with -601.22. AIC SIC and HQC with least value
50
9.553, 9.645 and 9.591 described that the appropriate model is ARMA (3, 3). Forecasted
evaluation describes that best fitted model in RMSE with 40.50563 is ARMA (6, 2). MAE with
least value 31.48136 demonstrates that the ARMA (5, 3) is favorable model. Appropriate model
in MAPE with least value 79.25677 is ARMA (4, 4). U test with least value 0.355584 describes
that ARMA (6, 4) is best fitted. Sunspot cycle (Jan 1944 - Feb 1954) exhibits that best fitted models
are ARMA (3, 3), ARMA (4, 4), ARMA (5, 3), ARMA (6, 2) and ARMA (6, 4). ARMA (6, 4)
process is an appropriate model with smallest DW value with 0.583 and maximum Log likelihood
value with -601.22. AIC SIC and HQC with least value 9.553, 9.645 and 9.591 described that best
fitted model is ARMA (3, 3). Forecasted evaluation describes that best fitted model in RMSE with
52.26395 is ARMA (6, 2). MAE with 44.02761 and U test with 0.366021 explore that ARMA (5,
3) is favorable model. Appropriate model in MAPE with least value 374.3422 is ARMA (4, 4).
Sunspot cycle (Feb 1954 - Oct 1964) shows that the best fitted models are ARMA (4, 3), ARMA
(5, 3), ARMA (5, 4), ARMA (6, 4) and ARMA (6, 5). Appropriate model is ARMA (6, 5) model
with least value DW with 0.663 and log maximum likelihood with -639.27. AIC SIC and HQC
with least value 9.560, 9.649 and 9.561 depict that best fitted model is ARMA (5, 3). Forecasted
evolution defines that RMSE with 81.95903 explore that the suitable model is ARMA (4, 3). MAE
with 64.21752 and U test with 0.483381 have best fitted model is ARMA (5, 3). According to
MAPE with 125.9374 shows that the best fitted model is ARMA (5, 4). Sunspot cycle (Oct 1964
- May 1976) depict that best fitted models are ARMA (3, 3), ARMA (4, 3), ARMA (5, 3), ARMA
(6, 2) and ARMA (6, 4). ARMA (5, 3) process has least value DW with 0.947 is best fitted and
strongly correlated. The best model according to log maximum likelihood with -626.80 is ARMA
(6, 4) model. AIC SIC and HQC with smallest value 8.645, 8.729 and 8.679 described that the
most appropriate model is ARMA (3, 3). Forecasted evolution reveals that RMSE, MAE and U
test with 38.99557, 32.35205 and 0.328025 have appropriate model is ARMA (5, 3). According
to MAPE with 88.14809 shows that the best model is ARMA (4, 3). Sunspot cycle (May1976 -
Mar 1986) indicate that best fitted models are ARMA (3, 3), ARMA (5, 2), ARMA (5, 3), ARMA
(6, 2) and ARMA (6, 4). An appropriate model is ARMA (6, 4) has least value of DW with 0.828
and log maximum likelihood with -582.02. AIC, SIC and HQC with smallest value 9.551, 9.644
and 9.589 described that the ARMA (3, 3) process is best fitted. Forecasted evolution describes
that RMSE, MAE and U test with 61.20026, 50.64551 and 0.375419 have appropriate model is
ARMA (5, 3). According to MAPE with 99.64765 shows that an appropriate model is ARMA (3,
51
3). Sunspot cycle (Mar1986 - Jun1996) describe the appropriate fitted models are ARMA (4, 5),
ARMA (5, 4), ARMA (5, 6), ARMA (6, 4) and ARMA (6, 5). The best fitted ARMA (6, 5) process
has least value DW with 0.834 and log maximum likelihood with -595.39 respectively. AIC, SIC
and HQC with smallest value 9.554, 9.645 and 9.591 shows that an appropriate model is ARMA
(4, 5). Forecasted evolution demonstrates that RMSE with 65.05685 explore the suitable model is
ARMA (5, 6). MAE and U test with 50.90771 and 0.433141 have best fitted model is ARMA (4,
5). According to MAPE with 120.4492 shows that the best fitted model is ARMA (6, 5). Sunspot
cycle (Jun1996 - Jan 2008) describe the best fitted models are ARMA (3, 3), ARMA (3, 4), ARMA
(4, 3), ARMA (5, 4) and ARMA (6, 4). ARMA (5, 4) process has the least value DW with 0.880
and log maximum likelihood with -638.59. AIC, SIC and HQC with smallest value 9.111, 9.195
and 9.145 shows that the appropriate model is ARMA (5, 4). Forecasted evolution define that
RMSE, MAE and U test with 43.79293, 35.46916 and 0.373450 have appropriate model is ARMA
(4, 3). According to MAPE with 178.4829 demonstrate that an appropriate model is ARMA (5, 4).
Sunspot cycle (Aug1755 - Jan 2008) describe the best fitted models are ARMA (5, 4), ARMA (5,
5), ARMA (6, 4), ARMA (6, 5) and ARMA (6, 6). ARMA (5, 4) process is best fitted with the
least value DW with 0.818 and log maximum likelihood and -13941.6. AIC, SIC and HQC smallest
value 9.106, 9.113 and 9.108 shows that the suitable model is ARMA (5, 5). Forecasted evolution
defines that RMSE and U test with 44.54175 and 0.370626 have best fitted model is ARMA (6,
4). According to MAE and MAPE with 35.30753 and 464.3263 shows that best fitted model is
ARMA (5, 5).
1.9: AR (p)-GARCH (1, 1) PROCESS OF SUNSPOT CYCLES
Autoregressive process (AR) is developed by Yule (Yule, 1927). An autoregressive process AR
(p) can be expressed by a weighted sum of its previous values and a white noise. The generalized
Autoregressive process AR (p) of lag p is expressed as follows.
Xt = α1 Xt-1 + α 2 Xt-2 + … + α p Xt-p + 𝜏t (1.23)
Here εt is white noise with mean E (𝜏t) = 0, variance Var (𝜏t) = σ2 and Cov (𝜏t -s, 𝜏t) = 0, if s ≠ 0.
For every t, it is supposed that 𝜏t is independent of the Xt-1, Xt-2, ….. 𝜏t is uncorrelated with Xs for
each s < t.
The GARCH (1, 1) process was developed by Bollerslev (1986). The generalized GARCH (p, q)
stochastic volatility model was developed by (Aquilar et. al. 2000, Kim et. al. 1998). The
52
generalized autoregressive conditional heteroskedasticity GARCH model is more appropriate to
analyze the fluctuation of variances. GARCH model is preferred to study the volatility clustering
and the relative volatility forecasts (Andersen et. al. 1998). GARCH (1, 1) process has better
forecasting ability as compared to other traditional models (Akgiray, 1989). GARCH (1, 1) process
is better to construct multi period long term forecasting. GARCH model is widely used for
modeling and forecasting of various other types of data which include economic and financial
modeling too. The GARCH (1, 1) process of Xt is represented as follows
𝜏t = 𝜎𝑡𝜖𝑡 (1.24)
Where ∈𝑡 ~ 𝐼𝐼𝐷(0, 1)
The model defines by itself as
𝜎𝑡2 = 𝛿 + 𝛽 𝜏𝑡−1
2 + 𝛾 𝜎𝑡−12 with 𝛿 + 𝛽 + 𝛾 ≥ 0 (1.25)
GARCH (1, 1) process is a covariance-stationary white noise process if and only if β + γ < 1. The
variance of the covariance-stationary process is represented as follows.
Var (Xt) = 𝛿
1−𝛽−𝛾 (1.26)
GARCH (1, 1) process is stationary if it holds stationary conditions. GARCH (1, 1) process mostly
follows leptokurtic (kurtosis greater than 3) which shows a heavy tailed behavior. GARCH (1, 1)
process specification with AR (p) is defined as follows (t = 0 ± 1, ±2, …).
Xt = α1 Xt-1 + α 2 Xt-2 + … + α p Xt-p + 𝜏t (1.27)
𝜏t = 𝜎𝑡𝜖𝑡 with ∈𝑡 ~ 𝐼𝐼𝐷(0, 1) (1.28)
𝜎𝑡2 = 𝛿 + 𝛽1 𝜏𝑡−1
2 + 𝛽2 𝜏𝑡−22 + ⋯ + 𝛽𝑝 𝜏𝑡−𝑝
2 + 𝛾 𝜎𝑡−12 (1.29)
Where E(𝜏t) = 0, variance Var (𝜏t | 𝜏𝑡−12 , 𝜏𝑡−2
2 … ) = σ2 and Cov (𝜏t -s, 𝜏t) = 0, if s ≠ 0.
Moreover, The Box-Jenkins methodology with GARCH approach is used to develop models, to
estimate the models and to forecast the sunspot cycle’s data.
Table 1.5 shows the results of performing the diagnostic checking test, forecast evaluation and
GARCH equations for AR (p)- GARCH (1,1) for Sunspot Cycles (1-24). Table 1.6 describes the
results of normality test for these models. The Gaussian quasi maximum likelihood estimation is
used to analysis AR (p)-GARCH (1, 1) model. Stationary GRACH (1, 1) model follows leptokurtic
(heavy tail) which represent the strong correlation among the spots and similarly each cycle of
sunspots. In this study AR (p)-GARCH (1, 1) process of each sunspot cycle has kurtosis value
greater than 3 (Leptokurtic) expect (8th, 16th, 17th) cycles follows the platokurtic flat tail. In this
53
study, the goodness of fit of the AR (p)-GARCH (1, 1) process is focused on residuals and more
specification with the standardized residuals. The Lagrange Multiplier, LM-test for
heteroskedasticity and Ljung-Box test is used for serial correlation. In stationary model the
coefficient of determination is used to predict the dependent variable from independent variable.
0 < R2 < 1, large values of R2 show that models demonstrate a closer fit to the time series data.
Durbin- Watson statistics test value of each sunspot cycle is less than 2 which shows that each
sunspot observation are correlated to each other. Akaike information criterion (AIC), Bayesian
Schwarz information criterion (BIC) and Hannan Quinn information criterion (HIC) values of each
cycle is also calculated. Sunspot cycles follow AR (2)-GARCH (1, 1) expect cycles (7th, 15th, 17th)
indicate AR (3)-GARCH (1, 1) model with 0.102 ± 0.995, 0.102 ± 0.997, 0.221 ± 0.973
respectively. Sunspot cycles GRACH (1,1) process specification with AR(p) follow positive
skewness (right tail) expect (2nd, 4th) cycles which is showing negative skewness (left tail) are
depicted in table 3. Sunspot cycle 5th is the best fitted model on the basis of Akaike information
criterion (AIC), Bayesian Schwarz information criterion (BIC) and Hannan Quinn information
criterion (HIC) which has least value 6.837493, 6.938750 and 6.878633 respectively. Each sunspot
value is correlated to other value. Jurque-Bera test failed in each cycle, which is shown that sunspot
cycles are not normally distributed. In the figure 1.6 is display the actual values, fitted values and
residual of Sunspot Cycle (Aug1755 - Jan 2008) as well as in the table is determined the forecasting
evolution of each sunspot cycle analysis in the view of diagnostic check, for each cycle Mean
Absolute Error (MAE) has least value. 6th cycle of sunspot has the smallest value of RMSE, MAE
and MAPE with 23.87355, 16.61689 and 81.56241 respectively. 7th cycle has least value of U test
with 0.342102. Figure 1.5 depicted that GARCH graph with a conditional standard deviation of
Sunspot Cycles (1-23 & 24 (complete sunspot cycle duration Aug1755 - Jan 2008)). Y-axis shows
that values of spots and X-axis represent that number of values. In figure 1.6 actual values, fitted
values and residuals of Sunspot Cycle (1-24) are displayed.
54
55
56
57
Figure 1.5: GARCH graphs with a conditional standard deviation of Sunspot Cycles
58
59
60
61
Figure 1.6: Actual values, Fitted values and Residual of AR– GARCH model of Sunspot Cycle
62
1.10: ARMA (p, q) – GARCH (1, 1) METHODS OF SUNSPOT CYCLES
The concept of ARMA models is strongly relevant in volatility modeling. The generalized
autoregressive conditional heteroscedastic (GARCH) models can be linked as ARMA models.
GARCH Models satisfy an ARMA equation with white noise. In time series, GARCH model
supposition that conditional mean is zero. Generally, conditional mean of ARMA model can be
structured. Identification of GARCH process focused on the square of residuals from the
appropriate ARMA models. Moreover, in the ARAM process the quasi maximum likelihood
estimation is nearly independent of their GARCH process. ARMA estimation and GARCH
estimation are strongly correlated if the ARMA – GARCH process has a skewed distribution
(Csyer et. al 2008). The ARAM process and GARCH process have similar behavior in forecasting.
ARMA – GARCH process provides a good estimation in time series data.
GARCH (1, 1) process specification with ARMA (p, q) is defined as follows (t = 0 ± 1, ±2, …).
Xt = α1 Xt-1 + α 2 Xt-2 + … + α p Xt-p + ∈t + β1 ∈t-1 + β2 ∈t-2+ … + βq ∈t-q (1.30)
𝜏t = 𝜎𝑡𝜖𝑡 with ∈𝑡 ~ 𝐼𝐼𝐷(0, 1) (1.31)
𝜎𝑡2 = 𝛿 + 𝛽1 𝜏𝑡−1
2 + 𝛽2 𝜏𝑡−22 + ⋯ + 𝛽𝑝 𝜏𝑡−𝑝
2 + 𝛾 𝜎𝑡−12 (1.32)
Where E(𝜏t) = 0, variance Var (𝜏t | 𝜏𝑡−12 , 𝜏𝑡−2
2 … ) = σ2 and Cov (𝜏t -s, 𝜏t) = 0, if s ≠ 0.
Moreover, The Box-Jenkins methodology with GARCH approach is used to develop models, to
estimate the models and to forecast the sunspot cycle’s data.
The novelty of this section to analyze the ARMA (p, q) -GARCH (1, 1) process of sunspot cycles.
The selection of ARMA (p, q) -GARCH (1, 1) model based on the least value of Darbin - Waston
statistics test (DW). Least DW value (< 2) shows that the each value of cycles is strongly correlated
and persistence to each other. AIC, SIC, HQC and Log likelihood also estimate to each cycle. In
Table 1.7, and 1.8 are depicted the GARCH (1, 1) model equations to specification ARMA (p, q)
model of sunspot cycles by diagnostic checking test and normality test. The Gaussian quasi
maximum likelihood estimation is used to analysis ARMA (p, q) -GARCH (1, 1) model. Lagrange
multiplier is used to verify the ARCH effect on following time series data. Ljung-Box test is used
for serial correlation of each sunspot cycle. The novelty of this research to analysis the conditional
mean and conditional variance effect on each sunspot cycle. The Correlation determination (R2)
of each cycle is range from 7.5 to 9.75 which are shown that each value are strongly correlated to
previous ones. The appropriate model which is frequently verified in sunspot cycles is GARCH
(1, 1) with specification ARMA (2, 2) cycles (1st, 4th, 12th, 13th, 14th, 15th, 16th, 19th, 20th, 23rd and
63
24th). ARMA (3, 3) -GARCH (1, 1) process follows with means and standard deviation
0.153±0.988, 0.190±0.983, 0.097±1.009 and 0.238±0.973 of most appropriate sunspot cycles
(5th, 6th, 7th and 15th) models respectively. ARMA (5, 1) -GARCH (1, 1) model follows cycles (2nd
and 11th) with 0.167±0.991 and 0.249±0.968 respectively. Cycle (Aug 1775 - Jun 1784)
represents ARMA (1, 1) -GARCH (1, 1) model with 0.197±0.983. ARMA (5, 3) -GARCH (1, 1)
process follows cycles (18th and 19th) with 0.258±0.965 and 0.258±0.966 respectively. Sunspot
cycle (Aug 1775 - Jun 1784) represents ARMA (1, 1) -GARCH (1, 1) model with 0.197±0.983.
Sunspot cycle (Oct1833 - Sep 1843) explore ARMA (3, 2) -GARCH (1, 1) process with
0.305±0.954. Sunspot cycle (Sep1843 - Mar 1855) explore ARMA (4, 2) -GARCH (1, 1) model
with 0.181±0.984. Sunspot cycle (Mar1855 - Feb 1867) explore ARMA (4, 4) -GARCH (1, 1)
model with 0.145±0.992. Sunspot cycle (Mar1986 - Jun1996) explore ARMA (6, 1) -GARCH (1,
1) model with 0.206±0.977. Sunspot cycle (Jun 1798 - Sep 1810) appropriate model ARMA (3,
3) -GARCH (1, 1) on the basis of least value of Akaike information criterion (AIC), Bayesian
Schwarz information criterion (BIC) and Hannan Quinn information criterion (HIC) which has
least value 7.075953, 7.197462 and 7.125322 respectively. According to log likelihood sunspot
cycle (Aug1755 - Jan 2008) best model ARMA (2, 2) -GARCH (1, 1) with maximum log
likelihood value -12808.63. Test of normality demonstrates that each sunspot cycles have positive
skewed expect sunspot cycles (4th and 19th) have negative skewed with -0.283 and -0.1425
respectively. Each sunspot cycle has kurtosis value greater than 3 (Leptokurtic) heavy tail expect
(2nd, 7th, 10th, 11th, 18th and 19th,) cycles follows platokurtic flat tail. Jurque-Bera test failed in each
cycle, which is shown that sunspot cycles are not normally distributed. Diagnostic Checking Test
is chosen with compression of these techniques with the help of Root mean square error (RMSE),
Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) and Theil’s U-
Statistics test. Figure 1.7 displayed that GARCH graphs of sunspot cycles (1-23 & 24 cycle
consists of complete time series data from 1855 to 2008) with conditional variance. In the figure
1.8 displayed that the actual values, fitted values and residual of sunspot cycle (1-24) as well as in
the table is determined the forecasting evolution of each sunspot cycle analysis in the view of
diagnostic check, for each cycle Mean Absolute Error (MAE) has least value. 6th cycle of sunspot
has the smallest value of RMSE, MAE and MAPE are 25.17780, 17.81173 and 79.46162
respectively. Each sunspot cycle has U test values less than 1 which is shown that are cycles are
correlated with others. 7th cycle has least value of U statistics test with 0.486173. It is a value is
64
near to zero describe that strongly correlated. Figure 1.9 depicted that the residual graph of
complete time series data ranges from 1st to 23rd cycles. Figure 1.9 shows that the residual graph
of the sunspot cycle (1-23). Figure 1.10 displayed that the standard residual graph of sunspot cycle
(1-23).
65
66
67
Figure 1.7: GARCH graphs with conditional variance of Sunspot Cycles
68
69
70
71
Figure 1.8: Actual values, Fitted values and Residual of ARMA– GARCH model of Sunspot
Cycle
72
Figure 1.9: Residuals ARMA – GARCH models of Sunspot Cycle (1-23)
Figure 1.10: Standard Residuals ARMA – GARCH models of Sunspot Cycle (1-23)
1.11: COMPARISON OF FORECASTING EVOLUTION OF ARMA, AR-GARCH AND
ARMA-GARCH OF SUNSPOT CYCLES
The forecasting evolution is based on RMSE, MAE, MAPE and U-test. The novelty of this section
focused on the comparison of the forecast evolution of ARAM, AR-GARCH and ARMA-GARCH
process of sunspot cycles.
73
1.11.1: Root Mean Square Error (RMSE) of ARMA, AR-GARCH AND ARMA-GARCH
of Sunspot Cycles
In this study, forecasting accuracy of ARMA, AR-GARCH and ARMA-GARCH models is
checked by RMSE. Table 1.9 depicts that the comparative study of the forecasting evolution of
ARMA (p, q), AR (p)-GARCH (1, 1) and ARMA (p, q)-GARCH (1, 1) process of each sunspot
cycle. Sunspot cycles (Aug 1755 - Mar 1766) has best fitted models are ARMA (6, 5), AR (2)-
GARCH (1, 1) and ARMA (6, 5)-GARCH (1, 1) process. Sunspot cycles (Mar 1766 - Aug 1775)
has expressed that the appropriate models are ARMA (5, 3), AR (2)-GARCH (1, 1) and ARMA
(5, 1)-GARCH (1, 1) process. Sunspot cycles (Aug 1775 - Jun 1784) has suitable fitted models are
ARMA (6, 4), AR (2) -GARCH (1, 1) and ARMA (1, 1) -GARCH (1, 1) process. Sunspot cycles
(Jun 1784 - Jun 1798) has best fitted models are ARMA (5, 4), AR (2)-GARCH (1, 1) and ARMA
(2, 2)-GARCH (1, 1) process. Sunspot cycles (Jun 1798 - Sep 1810) shows appropriate models are
ARMA (5, 3), AR (2)-GARCH (1, 1) and ARMA (3, 3)-GARCH (1, 1) process. Sunspot cycles
(Sep1810 - Dec 1823) express the suitable models are ARMA (3, 4), AR (2)-GARCH (1, 1) and
ARMA (3, 3)-GARCH (1, 1) process. Sunspot cycles (Dec1823 - Oct 1833) depict best fitted
models are ARMA (5, 3), AR (3)-GARCH (1, 1) and ARMA (3, 3)-GARCH (1, 1) process.
Sunspot cycles (Oct1833 - Sep 1843) shows best appropriate models are ARMA (4, 6), AR (2)-
GARCH (1, 1) and ARMA (3, 2)-GARCH (1, 1) process. Sunspot cycles (Sep1843 - Mar 1855)
has best fitted models are ARMA (4, 3), AR (2)-GARCH (1, 1) and ARMA (4, 2)-GARCH (1, 1)
process. Sunspot cycles (Mar1855 - Feb 1867) express best fitted models are ARMA (6, 4), AR
(2)-GARCH (1, 1) and ARMA (4, 4)-GARCH (1, 1) process. Sunspot cycles (Feb1867 - Sep 1878)
reveals best fitted models are ARMA (6, 4), AR (2)-GARCH (1, 1) and ARMA (5, 1)-GARCH (1,
1) process. Sunspot cycles (Sep1878 - Jun 1890) has suitable models are ARMA (5, 3), AR (2)-
GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Jun1890 - Sep 1902)
shows best fitted models are ARMA (5, 4), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1,
1) process. Sunspot cycles (Sep 1902 - Dec 1913) express appropriate models are ARMA (4, 6),
AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Dec1913 - May
1923) describe the best fitted models are ARMA (5, 4), AR (3)-GARCH (1, 1) and ARMA (3, 3)-
GARCH (1, 1) process. Sunspot cycles (May 1923 - Sep 1933) represent suitable models are
ARMA (6, 5), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles
(Sep 1933 - Jan 1944) has best fitted models are ARMA (6, 2), AR (3)-GARCH (1, 1) and ARMA
74
(2, 2)-GARCH (1, 1) process. Sunspot cycles (Jan 1944 - Feb 1954) describe the appropriate
models are ARMA (6, 4), AR (2)-GARCH (1, 1) and ARMA (5, 3)-GARCH (1, 1) process.
Sunspot cycles (Feb 1954 - Oct 1964) has best fitted models are ARMA (4, 3), AR (2)-GARCH
(1, 1) and ARMA (5, 3)-GARCH (1, 1) process. Sunspot cycles (Oct 1964 - May 1976) reveal best
fitted models are ARMA (4, 3), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process.
Sunspot cycles (May1976 - Mar 1986) express suitable models are ARMA (5, 3), AR (2)-GARCH
(1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Mar1986 - Jun1996) has best
fitted models are ARMA (5, 6), AR (2)-GARCH (1, 1) and ARMA (6, 1)-GARCH (1, 1) process.
Sunspot cycles (Jun1996 - Jan 2008) describe appropriate models are ARMA (6, 4), AR (2)-
GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Aug1755 - Jan 2008)
represent suitable models are ARMA (6, 4), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1,
1) process. Root Mean Square Error (RMSE) shows that ARMA (p, q) model is best appropriate
model as comparative to AR(p)-GARCH (1, 1) and ARMA (p, q)- GARCH (1, 1) models. Sunspot
cycles have least RMSE values in ARMA (p, q) process. Figure 1.11 displayed that RMSE
comparison of ARMA, AR-GARCH and ARMA-GARCH process of sunspot cycles.
Figure 1.11: RMSE of ARMA, AR-GARCH and ARMA-GARCH process of sunspot cycles.
1.11.2: Mean Absolute Error (MAE) of ARMA, AR-GARCH AND ARMA-GARCH of
Sunspot Cycles
The selection of best fitted model is based on MAE. Table 1.10: depict that sunspot cycles (Aug
1755 - Mar 1766) has best fitted models are ARMA (4, 5), AR (2)-GARCH (1, 1) and ARMA (6,
75
5)-GARCH (1, 1) process. Sunspot cycles (Mar 1766 - Aug 1775) has expressed that the
appropriate models are ARMA (5, 3), AR (2)-GARCH (1, 1) and ARMA (5, 1)-GARCH (1, 1)
process. Sunspot cycles (Aug 1775 - Jun 1784) has suitable fitted models are ARMA (6, 4), AR
(2) -GARCH (1, 1) and ARMA (1, 1) -GARCH (1, 1) process. Sunspot cycles (Jun 1784 - Jun
1798) has best fitted models are ARMA (5, 6), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH
(1, 1) process. Sunspot cycles (Jun 1798 - Sep 1810) shows appropriate models are ARMA (5, 3),
AR (2)-GARCH (1, 1) and ARMA (3, 3)-GARCH (1, 1) models. Sunspot cycles (Sep1810 - Dec
1823) express the suitable models are ARMA (6, 4), AR (2)-GARCH (1, 1) and ARMA (3, 3)-
GARCH (1, 1) process. Sunspot cycles (Dec1823 - Oct 1833) depict best fitted models are ARMA
(5, 3), AR (3)-GARCH (1, 1) and ARMA (3, 3)-GARCH (1, 1) process. Sunspot cycles (Oct1833
- Sep 1843) shows best appropriate models are ARMA (4, 6), AR (2)-GARCH (1, 1) and ARMA
(3, 2)-GARCH (1, 1) process. Sunspot cycles (Sep1843 - Mar 1855) has best fitted models are
ARMA (6, 4), AR (2)-GARCH (1, 1) and ARMA (4, 2)-GARCH (1, 1) process. Sunspot cycles
(Mar1855 - Feb 1867) express best fitted models are ARMA (6, 4), AR (2)-GARCH (1, 1) and
ARMA (4, 4)-GARCH (1, 1) process. Sunspot cycles (Feb1867 - Sep 1878) reveals best fitted
models are ARMA (6, 4), AR (2)-GARCH (1, 1) and ARMA (5, 1)-GARCH (1, 1) process.
Sunspot cycles (Sep1878 - Jun 1890) has suitable models are ARMA (5, 3), AR (2)-GARCH (1,
1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Jun1890 - Sep 1902) shows best
fitted models are ARMA (5, 4), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process.
Sunspot cycles (Sep 1902 - Dec 1913) express appropriate models are ARMA (4, 6), AR (2)-
GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Dec1913 - May 1923)
describe the best fitted models are ARMA (5, 3), AR (3)-GARCH (1, 1) and ARMA (3, 3)-
GARCH (1, 1) process. Sunspot cycles (May 1923 - Sep 1933) represent suitable models are
ARMA (6, 5), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles
(Sep 1933 - Jan 1944) has best fitted models are ARMA (6, 2), AR (3)-GARCH (1, 1) and ARMA
(2, 2)-GARCH (1, 1) process. Sunspot cycles (Jan 1944 - Feb 1954) describe the appropriate
models are ARMA (6, 4), AR (2)-GARCH (1, 1) and ARMA (5, 3)-GARCH (1, 1) process.
Sunspot cycles (Feb 1954 - Oct 1964) has best fitted models are ARMA (4, 3), AR (2)-GARCH
(1, 1) and ARMA (5, 3)-GARCH (1, 1) process. Sunspot cycles (Oct 1964 - May 1976) reveal best
fitted models are ARMA (4, 3), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process.
Sunspot cycles (May1976 - Mar 1986) express suitable models are ARMA (5, 3), AR (2)-GARCH
76
(1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Mar1986 - Jun1996) has best
fitted models are ARMA (4, 5), AR (2)-GARCH (1, 1) and ARMA (6, 1)-GARCH (1, 1) process.
Sunspot cycles (Jun1996 - Jan 2008) describe appropriate models are ARMA (3, 4), AR (2)-
GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Aug1755 - Jan 2008)
represent suitable models are ARMA (6, 4), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1,
1) process. The forecasting evolution of Mean Absolute Error (MAE) shows that ARMA (p, q)
model are best fitted and appropriate model as compared to AR(p)-GARCH (1, 1) and ARMA (p,
q)- GARCH (1, 1) models. Sunspot cycles have smallest values of MAE in ARMA (p, q) process.
Figure 1.12 depicts that the compression of MAE values of sunspot cycles by using ARMA, AR-
GARCH and ARMA-GARCH models.
Figure 1.12: MAE of ARMA, AR-GARCH and ARMA-GARCH process of sunspot cycles.
1.11.3: Mean Absolute Percentage Error (MAPE) of ARMA, AR-GARCH AND ARMA-
GARCH of Sunspot Cycles
Mean Absolute Percentage Error (MAPE) has variation in selection of best fitted model among
ARMA, AR-GARCH AND ARMA-GARCH process. Table 1.11: depict that sunspot cycles (Aug
1755 - Mar 1766) has best fitted models are ARMA (4, 4), AR (2)-GARCH (1, 1) and ARMA (6,
5)-GARCH (1, 1) process. Cycle 1 shows that best fitted model is an ARMA model with value
55.73550 among of these processes. Sunspot cycles (Mar 1766 - Aug 1775) has expressed that the
appropriate models are ARMA (2, 2), AR (2)-GARCH (1, 1) and ARMA (5, 1)-GARCH (1, 1)
process. Sunspot cycles (Aug 1775 - Jun 1784) has suitable fitted models are ARMA (5, 5), AR
77
(2) -GARCH (1, 1) and ARMA (1, 1) -GARCH (1, 1) process. Cycle 2 and 3 represent that the
appropriate model is AR-GARCH with value 115.7679 and 80.41010 respectively. Sunspot cycles
(Jun 1784 - Jun 1798) has best fitted models are ARMA (3, 3), AR (2)-GARCH (1, 1) and ARMA
(2, 2)-GARCH (1, 1) process. Sunspot cycles (Jun 1798 - Sep 1810) shows appropriate models are
ARMA (5, 3), AR (2)-GARCH (1, 1) and ARMA (3, 3)-GARCH (1, 1) models. Sunspot cycles
(Sep1810 - Dec 1823) express the suitable models are ARMA (6, 6), AR (2)-GARCH (1, 1) and
ARMA (3, 3)-GARCH (1, 1) process. Sunspot cycles (Dec1823 - Oct 1833) depict best fitted
models are ARMA (5, 5), AR (3)-GARCH (1, 1) and ARMA (3, 3)-GARCH (1, 1) process.
Sunspot cycles (Oct1833 - Sep 1843) shows best appropriate models are ARMA (3, 3), AR (2)-
GARCH (1, 1) and ARMA (3, 2)-GARCH (1, 1) process. Cycles 4, 5, 6, 7 and 8 reveal that the
best model is ARMA-GARCH with value 88.91661, 82.29495, 79.46162, 113.0161 and 79.03446
respectively. Sunspot cycles (Sep1843 - Mar 1855) has best fitted models are ARMA (3, 3), AR
(2)-GARCH (1, 1) and ARMA (4, 2)-GARCH (1, 1) process. Sunspot cycles (Mar1855 - Feb
1867) express best fitted models are ARMA (4, 4), AR (2)-GARCH (1, 1) and ARMA (4, 4)-
GARCH (1, 1) process. Cycle 9 and 10 express that best fitted model is ARMA model with value
55.33579 and 149.2235 respectively among of these processes. Sunspot cycles (Feb1867 - Sep
1878) reveals best fitted models are ARMA (3, 3), AR (2)-GARCH (1, 1) and ARMA (5, 1)-
GARCH (1, 1) process. Sunspot cycles (Sep1878 - Jun 1890) has suitable models are ARMA (4,
4), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Jun1890 -
Sep 1902) shows best fitted models are ARMA (4, 4), AR (2)-GARCH (1, 1) and ARMA (2, 2)-
GARCH (1, 1) process. Sunspot cycles (Sep 1902 - Dec 1913) express appropriate models are
ARMA (5, 5), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles
(Dec1913 - May 1923) describe the best fitted models are ARMA (5, 5), AR (3)-GARCH (1, 1)
and ARMA (3, 3)-GARCH (1, 1) process. Sunspot cycles (May 1923 - Sep 1933) represent suitable
models are ARMA (4, 4), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Cycles
11, 12, 13, 14, 15 and 16 analyze that the best model is ARMA-GARCH with values 163.9832,
105.1265, 108.7426, 117.5639, 79.70615 and 116.2730 respectively among of these process.
Sunspot cycles (Sep 1933 - Jan 1944) has best fitted models are ARMA (4, 4), AR (3)-GARCH
(1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Cycle 17 shows that best fitted model is ARMA
model with value 79.25677 among of these models. Sunspot cycles (Jan 1944 - Feb 1954) describe
the appropriate models are ARMA (4, 4), AR (2)-GARCH (1, 1) and ARMA (5, 3)-GARCH (1,
78
1) process. Sunspot cycles (Feb 1954 - Oct 1964) has best fitted models are ARMA (4, 3), AR (2)-
GARCH (1, 1) and ARMA (5, 4)-GARCH (1, 1) process. Cycle 18 and 19 express that best fitted
model is ARMA-GARCH model with values 122.4058 and 87.83657 respectively. Sunspot cycles
(Oct 1964 - May 1976) reveal best fitted models are ARMA (3, 3), AR (2)-GARCH (1, 1) and
ARMA (2, 2)-GARCH (1, 1) process. Cycle 20 and 3 represent that the appropriate model is AR-
GARCH with value 71.04261. Sunspot cycles (May1976 - Mar 1986) express suitable models are
ARMA (3, 3), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles
(Mar1986 - Jun1996) has best fitted models are ARMA (6, 5), AR (2)-GARCH (1, 1) and ARMA
(6, 1)-GARCH (1, 1) process. Sunspot cycles (Jun1996 - Jan 2008) describe appropriate models
are ARMA (5, 4), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles
(Aug1755 - Jan 2008) represent suitable models are ARMA (5, 5), AR (2)-GARCH (1, 1) and
ARMA (2, 2)-GARCH (1, 1) process. Cycles 21, 22, 23 and 24 reveal that the best model is
ARMA-GARCH with value 80.66910, 75.91525, 93.96324 and 205.0083 respectively. Most of
the cycles follow ARAM-GARCH models are appropriate model except cycles 2, 3 and 20 shows
AR-GARCH models and cycles 1, 9, 10 and 17 express are best fitted model are ARMA models.
Figure 1.13 displayed that the compression of MAPE values of sunspot cycles by using ARMA,
AR-GARCH and ARMA-GARCH models.
Figure 1.13: MAPE of ARMA, AR-GARCH and ARMA-GARCH process of sunspot cycles.
79
1.11.4: Theil’s U-Statistics of ARMA, AR-GARCH AND ARMA-GARCH of Sunspot
Cycles
Theil’s U-statistics test is used to analyze the correlation of observations. Strong correlation shows
when U test value near to zero. Table 1.12: depict that sunspot cycles (Aug 1755 - Mar 1766) has
best fitted models are ARMA (6, 5), AR (2)-GARCH (1, 1) and ARMA (6, 5)-GARCH (1, 1)
process. Sunspot cycles (Mar 1766 - Aug 1775) has expressed that the appropriate models are
ARMA (5, 3), AR (2)-GARCH (1, 1) and ARMA (5, 1)-GARCH (1, 1) process. Sunspot cycles
(Aug 1775 - Jun 1784) has suitable fitted models are ARMA (6, 4), AR (2) -GARCH (1, 1) and
ARMA (1, 1) -GARCH (1, 1) process. Sunspot cycles (Jun 1784 - Jun 1798) has best fitted models
are ARMA (5, 6), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles
(Jun 1798 - Sep 1810) shows appropriate models are ARMA (6, 4), AR (2)-GARCH (1, 1) and
ARMA (3, 3)-GARCH (1, 1) models. Sunspot cycles (Sep1810 - Dec 1823) express the suitable
models are ARMA (3, 4), AR (2)-GARCH (1, 1) and ARMA (3, 3)-GARCH (1, 1) process.
Sunspot cycles (Dec1823 - Oct 1833) depict best fitted models are ARMA (5, 3), AR (3)-GARCH
(1, 1) and ARMA (3, 3)-GARCH (1, 1) process. Sunspot cycles (Oct1833 - Sep 1843) shows best
appropriate models are ARMA (4, 6), AR (2)-GARCH (1, 1) and ARMA (3, 2)-GARCH (1, 1)
process. Sunspot cycles (Sep1843 - Mar 1855) has best fitted models are ARMA (4, 3), AR (2)-
GARCH (1, 1) and ARMA (4, 2)-GARCH (1, 1) process. Sunspot cycles (Mar1855 - Feb 1867)
express best fitted models are ARMA (6, 4), AR (2)-GARCH (1, 1) and ARMA (4, 4)-GARCH
(1, 1) process. Sunspot cycles (Feb1867 - Sep 1878) reveals best fitted models are ARMA (6, 4),
AR (2)-GARCH (1, 1) and ARMA (5, 1)-GARCH (1, 1) process. Sunspot cycles (Sep1878 - Jun
1890) has suitable models are ARMA (5, 3), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH
(1, 1) process. Sunspot cycles (Jun1890 - Sep 1902) shows best fitted models are ARMA (4, 3),
AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Sep 1902 - Dec
1913) express appropriate models are ARMA (4, 6), AR (2)-GARCH (1, 1) and ARMA (2, 2)-
GARCH (1, 1) process. Sunspot cycles (Dec1913 - May 1923) describe the best fitted models are
ARMA (5, 3), AR (3)-GARCH (1, 1) and ARMA (3, 3)-GARCH (1, 1) process. Sunspot cycles
(May 1923 - Sep 1933) represent suitable models are ARMA (6, 5), AR (2)-GARCH (1, 1) and
ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (Sep 1933 - Jan 1944) has best fitted models
are ARMA (5, 3), AR (3)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles
(Jan 1944 - Feb 1954) describe the appropriate models are ARMA (5, 3), AR (2)-GARCH (1, 1)
80
and ARMA (5, 3)-GARCH (1, 1) process. Sunspot cycles (Feb 1954 - Oct 1964) has best fitted
models are ARMA (5, 3), AR (2)-GARCH (1, 1) and ARMA (5, 3)-GARCH (1, 1) process.
Sunspot cycles (Oct 1964 - May 1976) reveal best fitted models are ARMA (4, 3), AR (2)-GARCH
(1, 1) and ARMA (2, 2)-GARCH (1, 1) process. Sunspot cycles (May1976 - Mar 1986) express
suitable models are ARMA (5, 3), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1)
process. Sunspot cycles (Mar1986 - Jun1996) has best fitted models are ARMA (4, 5), AR (2)-
GARCH (1, 1) and ARMA (6, 1)-GARCH (1, 1) process. Sunspot cycles (Jun1996 - Jan 2008)
describe appropriate models are ARMA (3, 4), AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH
(1, 1) process. Sunspot cycles (Aug1755 - Jan 2008) represent suitable models are ARMA (6, 4),
AR (2)-GARCH (1, 1) and ARMA (2, 2)-GARCH (1, 1) process. The forecasting evolution of
Theil’s U-statistics test shows that ARMA (p, q) model are best fitted and appropriate model as
compared to AR(p)-GARCH (1, 1) and ARMA (p, q)- GARCH (1, 1) models. Sunspot cycles have
smallest values of Theil’s U-statistics test in ARMA (p, q) process. Figure 1.14 displayed that the
compression of Theil’s U-statistics test values of sunspot cycles by using ARMA, AR-GARCH
and ARMA-GARCH models.
Figure 1.13: Theil’s U-statistics test of ARMA, AR-GARCH and ARMA-GARCH process of
sunspot cycles.
1.12: CONCLUSION
The novelty of this study is to analyze the sunspot cycles using GARCH models. Sunspot cycles
have stationary nature with second difference. Autocorrelation (AC), Partial Autocorrelation
81
(PAC) and Ljung-Box Q-statistics test are used to check the presence of white noise in the time
series data. Unit root test is rejected for each sunspot cycle, which confirms that each cycle has
stationary nature. To test the normality of sunspot cycles skewness and kurtosis were checked and
Jurque-Bera test was performed. This chapter utilizes the stochastic autoregressive and moving
average (ARMA) models to forecast the evolution of sunspot cycles. Least Square Estimation is
used to investigate the ARMA process. Various best fitted ARMA models are developed to
estimate and forecast the sunspot cycles. The selection of appropriate ARMA model is based on
the smallest value of Durbin- Watson statistics test. Durbin- Watson (DW) statistics test value of
each sunspot cycle is less than 2 which shows that sunspot observations are correlated to each
other. The stationary generalized autoregressive conditional heteroskedasticity GARCH (1, 1)
volatility model with specification autoregressive AR (p) process are used to estimate and forecast
the evolution of sunspot cycles. In addition, the stationary generalized autoregressive conditional
heteroskedasticity GARCH (1, 1) volatility model with specification autoregressive and moving
average ARMA (p, q) models are also utilized to estimate and forecasted evolution of the sunspot
cycles. GARCH (1, 1) stationary volatility model appeared to be the best forecasting model as
compared to the other models. GARCH (1, 1) model is heavy tail (fat tailed). Under model
Identification and Estimation, diagnostic checking and Forecasting the ARMA (p, q), AR (p)-
GARCH and ARAM (p, q)-GARCH models are found to be the most appropriate. The Gaussian
quasi maximum likelihood estimation (QMLE) is used to estimate GARCH (1, 1) process for
specification of AR (p) and ARMA (p, q) models. The selection of models in based on least value
of the Durbin- Watson statistic test. To detect the presence of Autoregressive Conditional
Heteroscedastic (ARCH) effect on sunspot cycle data, Lagrange Multiplier test is used. The
selection of the appropriate model is based on residual diagnostic checking such as ARCH LM,
normality test and correlogram squared residuals. Forecasting evolutions are verified by Root
Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error
(MAPE) and Theil’s U-Statistics test (U test). Moreover, Akaike information criterion (AIC),
Bayesian Schwarz information criterion (BIC) and Hannan Quinn information criterion (HQC),
maximum log likelihood estimation values are also calculated. The residuals of the best fitted
model are chosen by diagnostic checking. The forecasting evolution of each sunspot cycle under
the normality test is based on Skewness, Kurtosis and Jurque-Bera statistic tests. GARCH (1, 1) is
leptokurtic (a process having a kurtosis value greater than 3). The adequate ARMA, AR-GARCH
82
and ARMA-GARCH modeling for fractional Brownian motion of sunspot cycles will be useful to
predict the dynamical variables for 24th cycle and next cycles in the future. Finally, the forecasts
for the evolution of sunspot cycles obtained by ARMA, AR-GARCH and ARMA-GARCH models
are compared. RMSE, MAE and U test are utilized to check the appropriateness of various ARMA
models. Only the MAPE exhibited the appropriateness of ARMA-GARCH model.
83
CHAPTER: 2
FRACTAL ANALYSIS OF SUNSPOTS CYCLES AND ENSO CYCLES
AND DETERMINATION OF THEIR RELATIONSHIP
In this chapter sunspots cycles and ENSO cycles are studied and analyzed using fractal techniques.
Further, it is tried to establish a relationship between Sunspot cycles and corresponding ENSO
cycles. Sunspot cycles and ENSO cycle can be studied as time series using different techniques.
In this study, these time series will be analyzed using fractal techniques. As time series these data
sets can be characterized as linear and non-linear time series. Because these time series are
associated with natural phenomena, like other natural phenomena complexity is associated with
these time series. This complexity can be better explored by using fractals because the fractal
dimension of a curve is the space filling property of a curve. Also, fractal dimensions can be used
to study the persistency of the data. To check the persistency (smoothness and complexity) of the
sunspot cycles and ENSO cycles data, fractal dimension and Hurst exponent are calculated for
each sunspot cycle and ENSO cycle. The novelty of this study is to use the fractal instrument to
determine the statistical correlation between the sunspot cycles and the corresponding ENSO cycle
data. It is a new methodology to determine a correlation between two different data sets using
weak and strong fractal dimensions. In fact the method determines a correlation between the
natures of the two data sets regarding their persistency. The other novelty of this study is to analyze
the universal scaling parameters (Scaling exponent, spectral exponent and autocorrelation
coefficient) using fractal dimensions (both the self-similar and self-affine fractal dimensions) for
sunspot cycles. These results can be helpful to understand the relationship between solar and
terrestrial phenomena.
As for as the fractal dimensions are concerned, two main types of fractal dimensions viz. self-
similar and self-affine fractal dimensions are used in this study. In the self-similar, the geometric
object is composed into a union of rescaled copies of itself, which are uniform in all directions (or
isotropic rescaling). Whereas in the self-affine type, the geometric object is described as a union
of anisotropic (directionally non-uniform) rescaled copies of itself. Both the processes are
84
represented by linear transformation. One is self-similar fractal dimension and another is self-
affine fractal dimension. Self-similar fractal dimension FDS in calculating by box counting
technique or Hausdorff – Besicovich method. Whereas Self-affine fractal dimension FDA in
calculating by rescaled range analysis and another method Higuchi’s fractal dimension FDH is also
analyzed.
The common method to calculate the fractal dimension is Hausdorff-Besicovich method. Different
other methods like rescaled range analysis and Higuchi’s method are also used to calculate fractal
dimension. Box dimension or box counting method is more appropriate than other methods
(Michael, 1988). As mentioned earlier the fractal dimension describes the roughness and
smoothness of the data. To study solar activity fractal dimension is used to evaluate the
predictability of the solar cycle (Gayathri, 2010). Yanguang (Yanguang, 2010) has described three
different analyses involved in the fractal study of parameters. These are universal scaling analysis,
spatial correlation analysis and spectral analysis. Some useful details regarding spectral analysis
can be found in (Y. Chen, 2009).
2.1: DATA DESCRIPTION AND METHODOLOGY
The present study describes the Fractal Dimension (FD) approaches for the association of Sunspot
cycles and El Nino- southern oscillation (ENSO) related to local climate variability. The self-
similar fractal dimension (FDS) which is calculated by the Box counting method and self-affine
fractal dimension (FDA) which is applied by rescaled range for calculating the Hurst exponent long
range persistency of time series. Comparison among the FDS and FDA for each mean monthly
sunspot cycle and ENSO Cycles for the period 1st cycle to 23rd cycle and (1866 to 2012) 1st cycle
to 23rd cycle respectively. The present study, investigate that the association of the Mean monthly
Sunspot cycles from 1755 to 2008 which is consisted of 23 cycles whereas overall data cover in
24 cycles. The communication also examines the climatic variability impact of ENSO cycles from
1980 to 2000, in which duration the featured of the ENSO cycle is very active. All results are
explored the ENSO has five El Nino periods (1982-83, 1986-87, 1991-1993, 1994-95, and 1997-
98) and La Nina has three periods (1984-85, 1988-89, 1995-96) that were obtained from World
Meteorological Organization (WMO) and National Oceanic and Atmospheric Administration
(NOAA). The MATLAB 2016 and FRACTALYSE 2.4 software use tool for fractal dimension
and Hurst exponent analysis.
85
2.1.1: Fractal Dimension
The most appropriate method to calculate fractal dimension is Hausdorff - Besicovich box
counting or box dimension (Michael, 1988). It has belonged to the family of self-similarity. It
reveals empirical estimation and mathematical calculation. It can be calculated by taking a ratio
between log changes in object size and log change in measurement scale. This phenomena,
measurement scale approaches to zero. This expression can be mathematically written as
FDs = ln 𝑅
ln 𝑆 (2.1)
Fractal dimension has an ability to connect real world data which can easily measure by
approximated experiments, like coastlines, dust in the air, neuronal networks in the body, the
distribution of frequencies the colors emitted by the sun and during the storm, wrinkle the surface
of the sea are related with the fractal dimensions (Salakhutdinova, 1998).
This relationship can be summarized by (Sugihara, 1990) which reveals correlation as well as the
nature of the data.
Fractal Dimension = no of small pieces
magnification level (2.2)
The relationship between D and H calculate the analysis of the persistency of any time series data.
D + H = 2 (2.3)
2.1.2: Rescaled Range Analysis
Rescaled range analysis method belongs to the family of self-affine fractal dimension. Harold
Edwin Hurst (1880-1973) was introduced the rescaled range analysis (R/S) which is used to
analyze statistical methods for long term natural phenomena (Hurst, H. E. 1951). The purpose of
rescaled ranged analysis to reveal persistency analysis of time series data. Peters and Lo’s (Peters,
E. E. 1991and Lo, A. 1991) describe the dependence of long term persistence of finite non-period
cycle’s indices of stock market. In this technique, the ratio of two main variables, the range of time
series data (the maximum value is subtracted by minimum value) and the standard deviation is
known as Hurst exponent. In case a trend value exists, Hurst exponent can extrapolate a future
value. (Chen et al. 2002 and He et al. 2010) found that time series and Hurst exponent are
correlated. It generates by one dynamic system that is statistically significant degree does not
change when a segment is randomly cut from the correlated signals.
86
The rescaled range R/S is defined as the average of a given sequence of observation (Xt) as
⟨X⟩t = 1
τ∑ Xt
τ−1t=0 (2.4a)
Y = (𝑡, 𝜏) = ∑ [𝑋𝑡 − ⟨𝑋⟩𝑡]𝜏−1𝑡=0 (2.4b)
R (𝜏) = max0≤𝑡≤𝜏−1
𝑌(𝑡, 𝜏) - min0≤𝑡≤𝜏−1
𝑌(𝑡, 𝜏) (2.16)
S (𝜏) = (1
𝜏 ∑ ( 𝑋𝑡 − ⟨𝑋⟩𝑡) 2𝜏−1
𝑡=0 )1
2 (2.4c)
R / S (𝜏) = 𝑅 (𝜏)
𝑆 (𝜏) (2.4d)
We can now apply
R
S∝ τH (2.4e)
Where H is the Hurst exponent.
2.1.3: Persistency
The Hurst exponent is used to calculate persistency analysis (H.E. Hurst, 1951). In the data series,
the geometric (fractal) scaling is calculated by Hurst exponent (Turcotte, D.L., 1997). The Hurst
exponent measures long term memory of time series, which is very useful in forecasting. The
range of the value of Hurst exponent parameter HE lie from 0 to 1 (Hasting et al., 1993; Sun et al.,
2005). Hurst exponent measures the persistency and anti-persistency of regularity or irregularity
(chaos) the time series function respectively. If the value of H approaches to 1 it is indicated that
the time series data is quasi-regular (persistence) and it is approached to zero, so it is shown that
irregular (anti-persistence) (salakhudinova, 1998; Hanslmeier, 1999). The Hurst exponent value is
increased from 0 to 1 then persistency will also increase and similarly behave when a value will
decrease. When the value of HE = 0.5, this known as the Brownian process. If the value lies in the
range from 0 to 0.5, it called random walk (anti-persistence) process. The process stands for
persistent if the value of H lie from 0.5 to 1. In case the value of H approaches to 1 then the trend
of the data strength will also be increased (Hasting et al., 1993; Bundeet al. 1994). This relationship
between fractal dimension and Hurst exponent is described that correlation and the nature of the
data (Thomas. et. al 2008). Calculating Hurst Exponent (HE) = 2 – FD.
87
2.1.4: Higuchi’s Algorithm
In the time series data, Higuchi’s Algorithm is used to calculate the fractal dimension FD.
Higuchi’s method (Higuchi 1988) defines the scaling relationship between the mean normalized
length’s curves of the coarse-grained time series.
Whereas in the irregular time series, the power spectrum analysis is used for analysis fractal
dimension. The Higuchi’s Algorithm calculates the fractal dimension (FD), in the finite time series
data.
X(1), X(2), X(3), …, X(n) (2.5a)
Newly constructed 𝑋𝑘𝑚 is written as
𝑋𝑘𝑚 ; X(m), X(m + k ), X(m + 2k), …, X [𝑚 + (
𝑁−𝑚
𝐾) 𝑘] (2.5b)
With m = 1, 2, 3,…, k and [.] represent the Gauss notion, that is the highest integer. Where m
represents as initial time and k express time interval.
Suppose that, if k = 5 and N = 50
𝑋51: X(1), X(6), X(11), …, X(46) (2.5c)
𝑋52: X(2), X(7), X(12), …, X(47)
𝑋53: X(3), X(8), X(13), …, X(48)
𝑋54: X(4), X(9), X(14), …, X(49)
𝑋55: X(5), X(10), X(15), …, X(50)
Where k set of Lm (k) is calculated as follows
88
Lm(k) = 1
𝑘(∑ (𝑋(𝑚 + 𝑖𝑘) − 𝑋 (𝑚 + (𝑖 − 1)𝑘))
[𝑁−𝑚
𝑘]
𝑖=1) (2.5d)
Where the term 𝑁−1
[𝑁−𝑚
𝑘]𝑘′
indicate that the normalization factor.
The average value ⟨𝐿(𝑘)⟩ indicates the lengths association to the time series. The relationship
exists as follows
⟨𝐿(𝑘)⟩ ∝ 𝑘−𝐷 (2.5e)
Where D is the fractal dimension by Higuchi’s method. The Higuchi’s method is applied over time
series data which are not stationary.
The fractal dimension is a useful tool to analysis the relationship of some tools like spectral
exponent and autocorrelation coefficient by using fractal dimensions and Hurst exponents in both
self-similar and self-affine methods.
2.2: MATHEMATICAL MODELS AND FRACTAL DIMENSION RELATIONS OF
SUNSPOTS CYCLES
In this section demonstrate the brief information about certain qualities of universal scaling law of
self-similar and self-affine fractal dimensions.
2.2.1: Spatial Correlation Dimensions
A fractal is basically scale- free phenomena, whereas fractal dimension is used to be measured
with a characteristic scale. The fractal dimension of sunspots follows three basic concepts as
describe as, Euclidean plane of sunspots has 2-dimension so the Euclidean dimension of the
embedding of sunspots d = 2. The smallest unit of the sunspot is considered as a point, so the
topological dimension of sunspots is considered to be dt = 0 (Mandelbrot, 1983). So the value of
sunspots fractal dimension range from dt = 0 to d = 2. Fractal analysis of a sunspots time series
presented here is based upon the existence of correlation among sunspot cycles. So, this study
associates fractal analysis to correlation analysis. In this sense the generalized fractal dimension is
89
often called correlation dimension (Chen and Jiang, 2010; Grassberger and Procaccia, 1983).
Fractal dimension can characterize non equilibrium system FDS ≥ 0, in which the number of
independent degrees of freedom can be measured. In time series data, the fractal dimension usually
estimate and the relationship between Hurst exponent and fractal dimension are as follow
FDS = 2 – HE (2.6)
In the time series, the fractal dimension FDS is equal to 1.5 shows that unpredictable and no
correlation between amplitude changes between two successive time intervals. The value of the
fractal dimension is approached to 1 indicates that the process becomes more and more predictable.
The fractal dimension has ranged from 1 to 1.5 indicates that the process is persistent. If the fractal
dimension increased from 1.5 to 2 means that the process is anti-persistent. The value of FD = 1.5
indicates that the data is considered to be a normal scalar and is purely random, whereas FDS = 1
reveals that the time series data is smooth curve and nature is purely deterministic (Yousuf H et.al,
2008).
The length of sunspot cycles of self-similar fractal dimension FDS is given as follows.
δ(C) = δ1CFDs - d = δ1C –β (2.7)
Whereas δ1 indicates that the proportionality coefficient and β = d - FDS is the scaling exponent.
The relationship between d and FDS shows that FDS < d (Frankhauser and Sadler, 1998). If the
value of FDS lies between 1 and 2 then scaling exponent range from 0 to 1. If the value is FDS < 1
or FDS > 2 then the value of β > 1 or β < 0 respectively. For solar cycle, the fractal dimension FDS
can be revealed as a one-point correlation dimension which is indicating that zero-order correlation
dimension. Based on the equation (2.6), the density-density correlation function can be constructed
as
ζ (c) = ∫ 𝑓(𝑐)𝛿(𝑥 + 𝑐)𝑑𝑥 ∞
−∞= 2𝛿1
2 ∫ 𝑥𝐹𝐷𝑠−𝑑 (𝑥 + 𝛿) 𝑑𝑥∞
0 (2.8)
Where x define to the distance of the initial point and c to the distance of another point from the
initial point.
90
Let x = ηy, where η is the scale factor. A scaling relation can be express as
ζ (ηr) = 2𝛿1 2 ∫ 𝑥𝐹𝐷𝑠−𝑑 (𝑥 + ηr )𝐹𝐷𝑠−𝑑 𝑑𝑥
∞
0 (2.9)
= 2𝛿1 2 ∫ ηy𝐹𝐷𝑠−𝑑 (ηy + ηr )𝐹𝐷𝑠−𝑑 𝑑ηy
∞
0 (2.10)
= η2(𝐹𝐷𝑠−𝑑)+1 2𝛿1 2 ∫ y𝐹𝐷𝑠−𝑑 (y + r )𝐹𝐷𝑠−𝑑 𝑑y
∞
0 (2.11)
ζ (ηr) = η2𝐻 ζ (r) (2.12)
Where 2H = 2 (FDS - d) + 1, and H proved as a generalized Hurst exponent.
2.2.2: The Wave-Spectrum Relation of Sunspots
This section stresses upon the calculation of spectral exponent and spatial autocorrelation
coefficient (spatial scaling). These two play an important role to study the spatial behavior of data.
The data under consideration comprises of 23 sunspot cycles. To perform spatial scaling the
correlation function associated with the data is changed into an energy spectrum using Fourier
transform (Y. Chen, 2009). In addition to other methods Fourier transform can also be used to
study similarity. In this method relations of fractal parameters are determined by calculating
spectral exponents. For this purpose the following scaling law is used.
f (λρ) = λβf (λ) (2.13)
Where λ denotes the scaling factor, β describe the scaling exponent (β = d - FDS) and ρ is called
the length variable of each cycle.
The Fourier transform is applied in (2.3) the following scaling relation is obtained.
F (λγ) = F [f (λρ)] = λ-(1- β) F [f (ρ)] = λ-(1- β) F (γ) (2.14)
91
Where F has known as Fourier operator, γ represent the wave number, whereas F (γ) is the image
function of f (λ). Finally, the wave spectrum relation can be described from (2.9) as
S (γ) ∝ γ -2(1- β) (2.15)
The numerical relation between fractal dimension and the spectral exponent can be described by
comparison. The equation (2.7) can be correlated to the wave spectrum scaling by taking β = d -
FDS in (2.15), thus
S (γ) ∝ γ -2(1- d + FDs) = γ -2(FDs -1) = γ –α (2.16)
Here,
α = 2 (FDS - 1) (2.17)
Whereas, α is the spectral exponent and consider to be a constant which is describing the dynamical
behavior of the system in the time series fractal dimension. If α = 0 is described as white noise-
like system, and the system is uncorrelated and power spectrum which we have is independent of
the frequency. If α =1 is known as flicker or 1/f noise system which indicate moderately correlated.
If α = 2 is called Brownain noise-like system, which shows strongly correlated. The fractal
dimension DF lies between the range (1< DF < 2) indicate that the structure of nature is chaotic
and irregular then the spectral exponent, having ranged (1< α < 3) (Van Ness, 1968). The idea case
of chaotic physical structure is DF = 5/3.
The prerequisite of (2.17) is 1< FDS < 2 then the spectral exponent α can be revealed to be the
point-point correlation dimension. It is indicated that the relationship between one-point
correlation dimension (FDS) and the point-point correlation dimension (α). The relation between
the fractal dimension (FD) and spectral exponent (α) was introduced (Higuchi, 1988) and describe
by
FD = (5 - α) / 2 (2.18)
92
The parameters to determine the fractal dimension can be calculated by using two methods, self-
similarity and self-affinity consequently two different fractal dimensions are obtained viz. Self-
similar fractal dimension (FDS) and Self-affine fractal dimension (FDA). Here the wave-spectrum
scaling is performed by calculating both the fractal dimensions viz FDS and FDA. Earlier this sort
of analysis was performed by (S. D. Liu et.al. 1992, Benoit B 1999) using FDA. The fractal
dimension is related with the spectral exponent α with Hurst exponent H is known as the
Hausadroff measure, define the following relation (Burlaga L F et.al 1986, Turcotte 1992)
α = 2H + 1 = 5 – 2FDA (2.19)
The Hurst exponent can be calculated by the method of rescaled range analysis (Hurst 1965), which
is used nonlinear random process. HE is describe the power function R(𝜏) / S(𝜏) = (𝜏/2)H (Feder
1988). The Fractal dimension FDS is used to analysis the characters of spatial distribution of
sunspot cycles, whereas FDA is revealed the spatial autocorrelation of sunspot cycles. The local
dimension FDA of self-affine fractal records instead of FDS self-similar fractal trails (Feder 1988,
Takayasu 1990). The relationship between FDS and FDA can be described from (2.12) and (2.14)
FDS = 7
2 - FDA (2.20)
The parameters and equation can be express logically organized into a relation and a system of
fractal parameters (Figure 1).
93
Figure 2.1: A sketch map among the relationship of different fractal parameters.
The spatial activity of sunspots is supposed to be expressed as fractal Brownian motion (fBm) thus
the value of FDA of each cycle of sunspots lies between 1 and 2.
The relation between autocorrelation coefficient (C∇) and H of an incremental series on the
based on fBm can be given as
C∇ = 22H-1 – 1 (2.21)
Where C∇ represents the autocorrelation coefficient, which is based on the multiple-lag 1-
dimension spatial autocorrelation. The value of H = ½ then C∇ = 0 which is mentioned Brownian
motion. If H > ½ then C∇ > 0 representing that positive spatial autocorrelation and H < ½ then C∇
< 0 indicating that negative spatial autocorrelation.
The numerical relationship between different fractal parameter like FDS, FDA, HS, HA, αs, CA∇
and CS∇ can be indicated by using following equations (2.19), (2.20) and (2.21). The result which
is described in table 1 show that all values lies in significant scale. The range FDS from 1 to 1.5
which is indicate that number of spots are correlated to each other and have linear behavior. The
Hurst exponent range from 0 to 0.5 indicate that the persistency of the sunspots. The relation (2.20)
is theoretically valid when the fractal dimension FDS values is range from 1.5 to 2.
94
In this research, we intend to evaluate the complexity of the fractal geometric parameters of fractal
dimension. The scaling analysis and spectral analysis of sunspot cycles from this research is based
on spatial correlation. This study, the correlation function described the 1-dimension space. All
fractal parameters which are calculated by both techniques, self-similar fractal dimension (FDS)
and self-affine fractal dimension (FDA) are lying in significant scale and persistent. The table 2.1.
Indicate that the values of FDS and FDA in each cycle lies from 1 to 1.5 which are shown that each
cycle is persistent, correlated and predictable whereas the value of FDA > FDs. Similarly, self-
similar Hurst exponent HS and self-affine Hurst exponent HA both values are from 0.5 to 1 which
reveals that each cycle is persistent, the relationship between HS is greater than HA. The scaling
exponent β has ranged from 0 to 1 if fractal dimension lies between from 1 to 2. The values of β
lies its own valid scale which describes the dynamical behavior as 1-point correlation dimension.
The table1: also describe the numerical relation between the spectral exponent (α) and
autocorrelation coefficient (C∇) of self-similar and self-affine fractal dimension. If 1< DF < 2 then
the spectral exponent α having a range from 1 to 3. If α = 0 is indicate as white noise-like system
which describe uncorrelated behavior and power spectrum has frequency independent. If α =1 is
called flicker or 1/f noise system which behave moderately correlated. If α = 2 is called Brownain
noise-like system shows strong correlation. The equation (2.18) is used to calculate self-similar
spectral exponent (αS) and self-affine spectral exponent (αA). All values either αS or αA reveals the
system like Brownain noise. The equation (2.17) reveals that if 1< FDS < 2 then the spectral
exponent α describe point-point correlation dimension. It is mentioned that the relationship
between one-point correlation dimension (FDS) and the point-point correlation dimension (α). The
autocorrelation coefficient (C∇) which is described in the multiple-lag 1-dimension spatial
autocorrelation. The autocorrelation coefficient (C∇) is calculated by the relation of Hurst
exponent from the equation (2.21). The value of C∇ = 0 if H = ½ which is described Brownian
motion. If H > ½ then C∇ > 0 indicate positive spatial autocorrelation and if H < ½ then C∇ < 0
represent the negative spatial autocorrelation. Figure 2.1: indicate a sketch map among the
relationship of different fractal parameters, figure 2.2: represent the behavior of each solar cycle
in one domain. Figure 2.3: describe the relationship between the fractal dimensions of self-similar
and self-affine in each cycle.
95
Figure 2.2: Sunspot cycles (1- 24)
Figure 2.3: self-similar and self-affine fractal dimensions of sunspot cycles (1-24)
96
97
98
99
Figure 2.4: shows that the fractal dimension (self-similar and self-affine) of each sunspot cycle
(1-23) and the cycle 24 (complete data from Aug 1755 to Jan 2008)
100
2.3: MATHEMATICAL RELATIONSHIP BETWEEN HIGUCHI FRACTAL
DIMENSION AND SPECTRAL EXPONENT OF SUNSPOTS CYCLES
Higuchi’s fractal dimension (FDH) is usually used for non-linear dynamics. It is an important tool
to reveal relevant information from time series. Higuchi describes the method which provides the
precise estimate of the fractal dimension, even a small number of data. Higuchi method was
developed as an alternative method to spectral analysis. The relation between Higuchi’s fractal
dimension (FDH) and spectral exponent (α) (the slope is plots of spectral analysis). The Higuchi’s
fractal dimension (FDH) is related to the spectral exponent (α) as
α = 5 – 2FDH (2.22)
If FDH is in the range 1< FDH <2 then 1< α <3 (Van Ness, 1968). Higuchi method describes the
concept of two or more regions in which the graph between log Lm (k) and log k divide by
crossover. For instance the value that divides different scaling region associated with different
values of fractal dimension (FDH) (Higuchi 1988, 1990). Whereas FDH = 1.5, the dynamics of the
system are Brownian, FDH < 2 corresponding to pink noise and FDH = 2 represent white noise
(Galvez-coyt et al, 2011). If the value of α is equal to 0 define white noise-like system, and the
following system is uncorrelated and power spectrum which behave like independent of the
frequency. If α equal to 1 is known as flicker or 1/f noise system which describe moderately
correlated. If α = 2 is called Brownain noise-like system, which shows strongly correlated.
Higuchi’s fractal dimension FDH used to describe the behavior of spectral exponent.
The spectral exponent (α) values lie 1.7 < α < 2.3 which indicates that all cycles are strongly
correlated to each other’s. Sunspot cycles 5, 8, 10, 16, 17 and 23 has the spectral exponent (α)
value 2. The sunspot cycles 1, 2, 7, 11 and 13 shows the spectral exponent (α) value 1.8. The
sunspot cycles 9, 15 and 20 reveal the spectral exponent (α) value 1.9. The sunspot cycles 3, 4, 11,
18 and 21 calculate the spectral exponent (α) value 2.1. The sunspot cycle 6 has the value of α =
1.7, cycle 22 has α = 2.2 and cycle 19 indicate α = 2.3.
2.4: FRACTAL DIMENSION AND HRUST EXPONENT OF ENSO CYCLES BY USING
SELF-SIMILAR AND SELF-AFFINE FRACTAL DIMENSION
El Nino- southern oscillation (ENSO) is an irregular periodical fluctuation in wind and sea surface
temperature (SST) of the covering tropical eastern Pacific Ocean. Tropics and subtropics both are
101
affected on ENSO. There are two phases which are classified by temperature cooling phase known
as La Nino whereas warming phase is called El Nino. Southern oscillation is associated with the
sea surface change and atmospheric components. La Nino is associated with low air pressure
surface in the tropical western Pacific, whereas El Nino is associated with high air pressure surface
(Herring D. 2009). ENSO provide a mechanism which has maintained long-term climate stability
by transferring heat with high latitudes from the tropics. El Nino and La Nino have own regional
effect which is fluctuate, like the extremes of sun (Guhathakurta, M., & Phillips, T. 2013). The
Southern Oscillation is calculated by SOI, which is stand for the Southern Oscillation Index. The
SOI measures the variation of surface air pressure between Tahiti (in the Pacific) and Darwin,
Australia (on the Indian Ocean) (Climate glossary, 2009). El Nino periods have negative SOI,
which indicate a lower pressure over Tahiti and higher pressure at Darwin and La Nino periods
have positive SOI, which show higher pressure in Tahiti and lower in Darwin. The ENSO cycle
consists of averagely four years, but as per historical record ENSO cycle has varied between two
and seven years (NOAA).
This research is calculated the complexity of ENSO phenomena by using Self-similar fractal
dimension and self-affine fractal dimension and comparison both techniques along with long term
memory forecasting for data persistency also determined the Hurst exponent for the twenty four
Sunspots Cycle and 23 ENSO cycle with total cycles and active ENSO cycle periods. Each ENSO
cycle duration is 7 years from 1866 to 2012. ENSO cycle 22 cover 147 years data, although we
have selected 23 cycle as the more active duration from 1981 to 2000 respectively. The twenty
four Sunspot cycles and twenty two ENSO cycles along total both cycles with 23 ENSO cycle
have persistent behavior which is calculated from self-similar fractal dimension (FDS) and self-
affine fractal dimension (FDA), Hurst exponents also computed through FD which is depicted by
table 2.3. Results show that self-affine fractal dimension (FDA) and self-similar fractal dimension
(FDS) for twenty four sunspot cycles and twenty two ENSO cycles all along total data and active
ENSO cycles have fractal dimension value less than 1.5, the values of FDA > FDs, which is
indicating that each cycle is becomes predictable, correlated and persistent . Figure 2.5: represent
the behavior of each ENSO cycle in one domain. Figure 2.6: indicate a sketch map among the
relationship of different fractal parameters of ENSO cycles. Figure 2.7: describe the relationship
between the fractal dimensions of self-similar and self-affine in each cycle.
102
Figure 2.5: ENSO cycles (1-23)
Figure 2.6: self-similar and self-affine fractal dimension of ENSO cycle (1-23)
103
104
105
106
Figure 2.6: Shows that the fractal dimension (self-similar and self-affine) of each ENSO cycle (1-
21) and the ENSO cycle 22 (complete data from 1866 to 2012) and ENSO cycle 23 (active region
from 1981 to 2000)
107
2.5: COMPARATIVE ANALYSIS OF SELF- SIMILAR AND SELF-AFFINE FRACTAL
DIMENSION OF SUNSPOTS CYCLE AND ACTIVE ENSO PHENOMENA IN SAME
TIME INTERVAL
The impact of the earth climatic condition of solar activity oscillations is some extent is measurable
only in the long run duration. Solar cycles and ENSO are correlated to each other. Theory about,
the relationship between the sunspots and ENSO phenomena is premature, but now is developed
by collection of evidence that the solar cycle moderates wind field in the stratosphere and
troposphere (Anderson 1992, tnislay et.al, 1991) which provide a potential mechanism for the link
between the solar geometric effect and ENSO phenomena. The change in sunspot number can
reveal the association in which ENSO episode in more active when solar activity is weak
(Anderson 1990). The ENSO events were reduced by long term solar changes (e.g Mander
minimum, medieval grand maximum), similar solar activity is strong then ENSO period is weak.
For instance, the medieval warm period, solar activity is increased (Damen, 1988 and Anderson
1990).
ENSO cycle featured is very active in the duration between 1980’s and 1990’s. In this duration El
Nino has five periods (1982/83, 1986/87, 1991-1993, 1994/95, and 1997/98) and La Nina has three
periods (1984/85, 1988/89, 1995/96). In this duration El Nino has two strongest periods of the
century, which is consisted of 1982/83 and 1997/98 similarly two consecutive periods during
1991-1995 has cold period without intervening. ENSO cycle has varied from one cycle to the next
cycle. The fractal dimension is used in this study to define the complexity of the ENSO cycles.
Here, self-similar and self-affine behavior of same duration of sunspots and ENSO cycles.
Table 2.4 and figure 2.7, indicate our selected same cycle duration both of data also signify the
self-affine fractal dimension (FDA) is greater than self-similar fractal dimension (FDS ) and ENSO
cycles is also less than self-similar fractal dimension (FDs) of Sunspot cycles. The Hurst exponent
for each cycle, calculate with the help of FD to observe that FDA is increasing than, Hurst exponent
of self-affine HEA is decreasing, this relation also provides the smoothness and accurate
persistency of the cycles. Table 2.3 – 2.1 depicts Hurst exponents for smoothness. Figure 2.8 shows
that the self-similar and self-affine fractals of sunspot cycles and ENSO cycles in one domain.
108
Figure 2.7: Comparative analysis of self- affine and self-similar fractal dimension of sunspot
cycles and ENSO cycles (May 1976 to Jan 2008)
Figure 2.8: comparison of self-similar and self-affine fractal dimensions of sunspot and ENSO
cycles.
2.6: CONCULSION
This chapter reveals the persistency analysis of solar activities and ENSO cycles by using fractal
dimension involving Hurst exponent. The fractal dimension (FD) represents the roughness (local
property) of the time series data of solar activities (sunspot activity) and ENSO episode, whereas
Hurst exponent (HE) provides the smoothness (universal property) of the data. This study
109
investigates the relationship between self-similar fractal dimension (FDS) and self-affine fractal
dimension (FDA). FDS calculated by Box counting method and self-affine fractal dimension (FDA)
is calculated by using rescaled range method. The Hurst exponents are calculated by FD which is
calculated by both the self-similar fractal dimension (FDS) and self-affine fractal dimension (FDA).
The spectral exponent is calculated by using equation (2.13) and the autocorrelation coefficient is
calculated by using equation (2.16). Table 2.1: indicates the numerical relationship between Hurst
exponents, spectral exponent and autocorrelation coefficient. Both fractal dimensions (self-similar
and self-affine) are used. Each cycle is found to be persistent and correlated, the values of Hurst
exponent lie between 0.5 and 1 but HES values are greater than HEA. All values of the spectral
exponent (αs and αA) of each sunspot cycles behave like Brownain noise which indicates the long
term dependency. The autocorrelation coefficient is found to be significant using both the fractal
dimensions FDS and FDA. For sunspot cycles the relationship between self-similar and self-affine
fractal dimensions expressed by equation (2.15) failed as this relation is valid only if the fractal
dimension lies between 1.5 and 2. Here each for each sunspots cycles the fractal dimension is less
than 1.5. The spectral exponent value which is calculated by Higuchi’s Fractal dimension (FDH)
also identifies the strong correlation among Sunspots cycles.
The complexity of each cycle of ENSO data, the complete data of all the cycles and the data of the
active ENSO period are analyzed. First, self-similar fractal dimension (FDS) (using Box counting
method) and self-affine fractal dimension (FDA) (by using rescaled range method are calculated.
Then the Hurst exponents are calculated by using both methods FDS and FDA. It is observed that
FDA > FDs in both the cases of sunspot and ENSO cycles. Similarly, self-similar fractal dimension
(FDS) of ENSO cycles is also less than the self-similar fractal dimension (FDs) of Sunspot cycles.
This means that ENSO cycles are more persistent than Sunspot cycles data. Sunspots cycles and
ENSO are correlated with each other. It can be noted easily that if FD increases, then HE decreases.
Each sunspot cycle has greater mean and prolonged tail. The mean-tail assessment confirms the
FD-HE analysis. This study can be useful for further investigation of the impact of Sunspot and
ENSO related local climatic variability.
110
CHAPTER 3
A COMPARATIVE STUDY OF SUNSPOT CYCLES AND ENSO CYCLES
ANALYSING PROBABILITY DISTRIBUTIONS AND HEAVY TAIL
This chapter describes the behavior of solar activity or solar cycle (Sunspot cycle) and the El Nino-
southern oscillation (ENSO cycle) data and analysis the influence of it in the earth’s climate. The
study of probability distributions of climatic parameters usually gives imminent into the physical
processes (Hussain et al., 2012). Sunspots cycles and ENSO cycles follow the probabilistic
distribution. In evaluating the nature of the distribution observed by existing data, we check for
the deviation from normality using Kolmogorov–Smirnov test (KST), Anderson-Darling test
(ADT) and Chi-Square test (CST). The analysis of earth’s climate, solar cycle plays an essential
role (Weiss et al., 2007). In the early 1600s, the variation of dark spots which are appearing on
the surface of the sun, known as sunspots were observed. These variations are based on regular
cycle which has peak approximately 11.1 years. The climate change is usually affected by about
11 years or 18.6 years solar cycle (Seleshi et al. 1994). The cycle of solar activity concurs with a
small oscillation as an output in solar energy. The output energy of sun light to some extent higher
during episodes with large numbers of sunspots. The sun has a great influences the atmosphere
and the climate of the earth (Bal et.al 2010, Siingh et.al 2010 and Spinage 2012). The solar activity
increase and the intensity of galactic cosmic ray flux is decreasing, causing the high rainfall (Sun
et al. 2002 and svensmark 1997). The magnetic field of the earth increases with the solar cycle
intensified. The annual rainfall in 10/11 year cycles is described in many regions monsoonal
rainfall received (Seleshi et al. 1994). In the recent research described that tropospheric and
stratospheric weather system are usually influenced by the solar cycle, which is indicating that the
dependency of climates in solar activity (Almedia et al. 2004, Lakshmi et al. 2003, Kumar et al
2010 and Guhathakurta et al. 2007). The correlation between rainfall and solar cycle (22 years) for
long term time scale is positive and 11 year time scale provide very low or negative correlation
(Hiremath 2006). It is difficult to calculate the direct relationship between the sunspot cycle and
climate, but some occasions of climate relate with sunspot numbers. For instance, in the interval
1645 to 1715, the sunspot number was dramatically lower than previous one. This interval known
111
as the Maunder Minimum. The solar energy reduction in the minimum duration could have a slight
cooling of the Earth. The main task of this chapter to analysis the comparative study of solar cycles
and ENSO cycle in the perspective of probabilistic model and heavy tail analyzes and long range
persistence.
3.1: DATA DESCRIPTION AND METHODOLOGY
The present study investigates the association of the Mean monthly Sunspots cycles and ENSO
cycles for the period from 1755 to 2008 and 1866 to 2012 respectively. The Generalized Pareto
and Generalized extreme value, probability distribution techniques are applied to calculate the
mean and standard deviation and most appropriate distribution for each cycle of sunspots and
ENSO activity on the basis of goodness of fit Kolmogorov–SmirnovD, Anderson-Darling and
Chi-square test. The communication also examines the climatic variability impact of ENSO cycles
from 1980 to 2000, in which duration the featured of the ENSO cycle is very active. All results are
explored the ENSO has five El Nino periods (1982/83, 1986/87, 1991-1993, 1994/95, and
1997/98) and La Nina has three periods (1984/85, 1988/89, 1995/96) that were obtained from
World Meteorological Organization (WMO). This chapter is divided by three main sections,
section one is consists of the best fitted probability distribution with the help of Kolmogorov–
Smirnov test (KST), Anderson-Darling test (ADT) and Chi-Square test (CST). In this section, also
described the probability distribution of fractal dimension of sunspot cycles and ENSO cycles from
the perspective of self-similar and self-affine fractal dimension. Comparative study of the sunspot
cycle and Active region of ENSO cycle also investigate. The second section investigates long
range persistence behavior of the solar cycle (or sunspot cycles) and ENSO cycles and also
compare the same interval of active regions of the ENSO cycle and sunspot cycles. Last section
investigates the heavy tail parameter for long term relation and persistency in the perspective of
Hrust exponent of self-similar and self-affine fractal dimension. The Easy fit version 5.6 and
Matlab 2016 are utilized for the testing of underlying distributions. The techniques which are
described in this chapter will not only be useful in the field of solar phenomena, but also will be
helpful to investigate the solar-terrestrial relations (Hassan and Abbas, 2014).
112
3.2: PROBABILITY DISTRIBUTION APPROACH
In this section, we briefly discuss the probability distribution techniques which are used to check
the best fitted distribution
3.2.1: Goodness of Fit Tests
The goodness of fit test is used to measure random sample with a theoretical probability
distribution function. These tests indicate that which distribution is best for the data. The general
method of goodness of fit based on analysis a test statistic in which some function of the data,
calculating the distance between the hypothesis and the data, and also measuring the probability
of the data.
3.2.2: Kolmogorov-Smirnov Test
If a sample follows from a hypothesized continuous distribution, then Kolmogorov-Smirnov test
is used to decide. The Kolmogorov-Smirnov test is basically based on the empirical cumulative
distribution function (ECDF). The Kolmogorov-Smirnov test is defined as the largest vertical
difference between F (x) and G (x).
D = max|H(x) − G(x)| (3.1)
Where H (x) and G (x) are predetermined cumulative distribution and sample cumulative
distribution for the given sample of size n (Hussain, 2006).
3.2.3: Anderson-Darling Test
Anderson-Darling (ADT) test was intended to detect differences in the tails of the distribution,
especially (Li et al., 2002). The Anderson- Darling test is same as the Kolmogorov-Smirnov test,
but it provides strong information related to the tails of the distribution. This test depends on the
number of intervals. The Anderson-Darling test only applies to the input data sample which is a
drawback of the test (Walck, 2000). The Anderson-Darling statistic (A2) is defined as
A2 = −1
n∑ (2i − 1)[lnαi + ln(1 − α n−i+1)] − n,n
i (3.2)
113
Where 𝛼i represent the theoretical cumulative distribution at the ith largest observation, 𝑛 is the
number of observations (Hussain, 2006).
3.2.4: Chi-Squared
The Chi-Squared test apply to determine where a population sample with a specific distribution.
This test is used for binned data. The nature of binned data depends on the test of statistic. Many
formulas are used to calculate bins (S) which depends on sample size (N).
S = 1 + log2N (3.3)
Grouped data can be used which is divided into equal difference. The statistic of Chi-Squared is
defined as
(3.4)
Where Ei is the expected frequency for bin i and Oi is the observed frequency for bin i, calculated
by Ei = G(x2) - G(x1), where G is the Cumulative Density Function of the probability
distribution being tested, and x1, x2 are the limits for bin i.
3.2.5: Best Fitted Probability Distribution
In many branches of science, experimental analysis may encounter the problems where the
standard probability distributions are useful. These probability distributions are useful to generate
the random numbers (Walck, 2007). In the study of probability distribution the comparative study
sunspots and ENSO cycle are very useful to analyze the change and variation in the earth’s climate.
The advantage probability distribution of random number study for solar and terrestrial parameter
provides the essential knowledge about the physical processes governing in it. The KST
[Kolmogorov-Smirnov D-test] is used to check the deviation of normality to evaluating the nature
of the distribution with the real time data. We have described the suitable and most fitted
distributions for solar cycle and the ENSO cycles and the relation of sunspots and active region of
the ENSO cycle in the same interval. The probability distribution of self-similar and self-affine
fractal dimension also calculated.
114
3.2.5.1: The Generalized Extreme Value (GEV) Distribution
The generalized extreme value (GEV) distribution in probability theory and statistics is a family
of continuous probability distributions which are developed with extreme value theory to combine
the Fréchet, Gumbel and Weibull families also called type I, II and III extreme value distributions.
The location, scale and shape parameters of generalized extreme value are μ R, σ > 0 and k R.
The symmetrically generalized extreme value distributed mathematically it can be expressed as a
probability density function.
f (x) = (1 + 𝑘 (
𝑥− 𝜇
𝜎))−
1
𝑘 𝑖𝑓 𝑘 ≠ 0
𝑒− (𝑥− 𝜇)
𝜎 𝑖𝑓 𝑘 = 0
(3.5)
Mean =
𝜇 + 𝜎(𝑔1−1)
𝑘 𝑖𝑓 𝑘 ≠ 0, 𝑘 < 1
𝜇 + 𝜎𝛾 𝑖𝑓 𝑘 = 0∞ 𝑖𝑓 𝑘 > 1
(3.6)
(3.7)
and γ is Euler’s constant.
Variance =
𝜎2 (𝑔1−𝑔22 )
𝑘2 𝑖𝑓 𝑘 ≠ 0 , 𝑘 <1
2
𝜎2 𝜋2
6 𝑖𝑓 𝑘 = 0
∞ 𝑖𝑓 𝑘 ≥ 1
2
(3.8)
In Eqs. (3.7) and (3.8), E [X] and Var (X) represent mean and variance of samples respectively
(Hussain, 2006).
115
3.2.5.2: Generalized Pareto Distribution
The Generalized Pareto Distribution (GPD) was developed by (Pikands 1975) and further studied
(smit 1984, Castillo 1997 and 2008). The Generalized Pareto Distribution (GPD) plays an essential
role to study extreme events in general.
Parameter of Generalized Pareto distribution is;
𝛼 ∈ (−∞, ∞); Location (real) (3.9)
𝛽 ∈ (0, ∞); Scale (real)
𝛾 ∈ (−∞, ∞); Shape (real)
Support
x ≥ α (𝛾 ≥ 0)
α ≤ x ≤ 𝛼− 𝛽
𝛾 (𝛾 < 0)
f (x) = (1 + 𝑘 (
𝑥− 𝜇
𝜎))−
1
𝑘 𝑖𝑓 𝑘 ≠ 0
𝑒− (𝑥− 𝜇)
𝜎 𝑖𝑓 𝑘 = 0
(3.10)
𝑒− 𝑡 (𝑥) for x ∈ support
Probability density function
f (x) = 1
𝛽 (1 + 𝛾 𝑧)
−(1
𝛾 +1)
where z = 𝑥− 𝛼
𝛽 (3.11)
Cumulative distribution function:
F (x) = 1 – (1 + 𝛾 𝑧)−
1
𝛾 (3.12)
Mean = α + 𝛽
1− 𝛾 (𝛾 < 1) (3.13)
116
Variance = 𝛽2
(1− 𝛾)2(1−2𝛾) (𝛾 <
1
2) (3.14)
The probability distribution of sunspots and ENSO cycles are described as below,
3.3: BEST FITTED PROBABILITY DISTRIBUTIONS OF SUNSPOT CYCLE
The sunspot has an intense magnetic field and cool temperature and the center temperature of a
sunspot is 6400°F (TJO News, 2006). The Sunspots are associated with active regions, which are
areas of locally increased magnetic flux of the Sun (Kevin R. et.al 2014). The sunspot is cooler
and look dark on the sun’s surface is heated to the orange-white stage. The sunspots are a large
number of the blotches on the sun that is often larger in diameter than the earth. The yearly average
of sunspot areas has been recorded since 1700 (Kok K. et. al. 2001). A maximum number of
sunspots are known as solar maximum. Normally sunspots appear in pairs in which one has a
positive magnetic polarity and the other has a negative magnetic polarity and join these two spots
centers by a line which is nearly parallel to the sun’s equator. Sunspots often appear near 30-35
degree north and south of the sun’s hemisphere with higher latitudes. The sunspots life consists of
either days or one week or few weeks.
This portion investigates that sunspot data from 1755 to 2016 (25 cycles) which are consist of 1 to
23 cycles, 24 cycle (duration Jan 2008 to continue) prediction and cycle (1-23 complete data). In
this study, two types of probability distribution have been stressed one is Generalized Pareto
(GPD) Distribution and another is Generalized Extreme value (GEV) Distribution. The deviation
is checking from normality using Kolmogorov–Smirnov test (KST), Anderson-Darling test (ADT)
and Chi-Square test (CST). Both distributions have continuous nature and have heavy tail
behavior. The Generalized Pareto Distribution (GPD) is commonly used for those qualities which
are distributed with long right tail. The table 3.1 represent that the mean, standard deviation and
distributions of each sunspot cycle (from 1 to 23, 24 and complete data 1-23) with parameters of
testing probability distributions. From Table 3.1 we observe that different cycles follow different
distributions. This behavior shows that the distributions become increasingly heavy right tailed.
Figures. 3.1 represent the plot of total sunspots data and plot of each cycle from 1755 to 2016 as
well. Figure 3.2 depicts a complete sketch the probability distribution with different peak of
sunspot data.
117
118
119
120
121
Figure 3.1 These figures shows the fitting of probability distribution of 23 sunspot cycles and 24th
incomplete cycle and 1-23 complete data cycle with different peaks and duration along with total
duration 1755-2016.
Figure 3.2: The probability distributions different peaks plots of 24 sunspot cycles (1755-2016)
This study estimates the probability of the mean monthly Sunspot cycle (1st to 24th cycle is in
continuation and total (1stto 24thcycles). The probability distribution (Tables 3.1 and figure 3.1 and
3.2) for Sunspot data series depicts most of the sunspot cycles are followed Generalized Pareto
distribution and Generalized extreme value distribution. In Table: 3.1 indicates the mean and
122
standard deviation for the duration of all cycles and total cycles of Sunspots parameters of test
probability distributions. The sunspot cycles reveals the Generalized Pareto Distribution
probability apart from cycle 9 which describes the generalized extreme value distribution. Both
distributions indicates the heavy tail on the right side. In this section, the prediction of 24 cycle
also follows the Generalized Pareto Distribution. Table 3.1 depicted. Variations of the mean and
standard deviation are useful to analysis the changeable behaviors of the sunspot cycles start from
1755 to till 2011, sunspot cycle 24th is still processed. The figure 3.2 indicate the probability
distributions different peaks plots of 24 sunspot cycles (1755-2016). The variation in mean and
standard deviation represents the change in the strength of the sunspot cycle. Figure: 3.2 represents
the plot of twenty four with the total sunspot cycle plot of each cycle from 1755 to 2008 Cycles
1st to8th and 10th to 23rd follow the Generalized Pareto distribution and the remaining cycle 9
follow the Generalized Extreme value probability distribution.
3.4: BEST FITTED PROBABILITY DISTRIBUTIONS OF ENSO CYCLE
Enso phenomena effects on climate and weather. The strong ENSO episode brings droughts in the
Southeast Asia countries. In the USA, these events make a reason to bring heavy rain and
increasing the number of hurricanes and typhoons in the eastern pacific. The relationship between
rainfall and ENSO in different world areas vary from area to area (Bhalme et. al. 1984). The impact
of ENSO of Indian Ocean varies from season to season (Singh et. al. 1999). In the early nineties
climatologist have intent to understand the relation of the Indian monsoon with the ENSO is
ordered to estimate the monsoon rainfall over Asia (Wabster et.al 1992). The review of the
historical background the relation of monsoon and ENSO some qualitative studies on the coupling
of ENSO year and rainfall over Pakistan (Arif et al. 1994; Chaudhary, 1998). The variability of
Rainfall of Indian summer monsoon and East Asian summer monsoon and their relationship with
ENSO has been discussed (Walker, 1923, 1924; Rasmusson et. al, 1983; Shukla et. al. 1983;
Mooley et. al. 1987; Yasunari, 1990; Webster et. al. 1992; Lau et. al. 1996; Ju et. al. 1995; Soman
et. al. 1997; Liu et. al. 2001; Lau et. al. 2000).
This portion investigates that ENSO data from 1866 to 2012 (23 cycles) which are consist of 1 to
21 cycle, 22 cycle (duration from 1866 to 2012 complete data) and 23 cycle (duration from 1980
to 2000) active episode of ENSO cycle. In this study, we stress upon various types of the
probability distribution. Generalized Extreme value Distribution (GEV), Generalized Pareto
123
Distribution (GPD) and Logistic Distribution are used to analyze the behavior of ENSO data with
the help of testing Kolmogorov–Smirnov test (KST), Anderson-Darling test (ADT) and Chi-
Square test (CST). These distributions indicates that the ENSO cycles have continuous heavy tail
persistency on the right side. The table 3.2 shows that the mean, standard deviation and
distributions of each ENSO cycle (from 1 to 21, complete data 1-21 and 22) with parameters used
to test probability distributions. From Table 3.2 we analyze the different cycles follow different
distributions. This behavior reveals that the distributions become increasingly heavy right tailed.
Figure. 3.3 represent the plot of total ENSO cycles and the plot of each cycle from 1866 to 2012
as well
124
125
126
Figure 3.3: These figures show the fitting of probability distribution of 21 ENSO cycles, 23th cycle
(1-21) complete data cycle with different peaks and duration along with total duration 1755-2016.
22th cycle (1981-2000) active region of ENSO cycle.
127
Figure 3.4: The probability distributions different peaks plots of 23 ENSO cycles (1866 to 2012)
The probability distributions of ENSO data (from 1866 to 2012) are formed (figure 3.3). In this
study, we analyze the probability of the mean monthly ENSO cycles (1st to 23rdwhile 22 cycle
consists of complete data duration of 1866 to 2012 and 22 cycle duration most active durations
from 1981 to 2000 respectively). The probability distribution (Tables 3.2 and figure 3.3 and 3.4)
ENSO episode series show followed the generalized extreme value distribution, generalized Pareto
distribution and only cycle 6thfollowed the logistic distribution. The ENSO cycles show the
generalized extreme value distribution, the ENSO cycle 13th is revealed Generalized Pareto
distribution which is shown heavy tail similarly cycles 6 seem to Generalized Pareto distribution.
Cycles 22 which is covered for 112 year cycle and cycle 23th is consisting of a 20 year data which
completes the durationfrom1981 to 2000 in which ENSO activity are stronger similar to follow
the generalized extreme value distribution.
In particular, the changes in the distribution of Generalized Extreme Value and Generalized Pareto.
These distributions are becoming an increasingly right side heavy tailed ENSO cycles (figure 3.3
– 3.4). We are able to calculate the distribution parameters mean and standard deviations to
confirm the climatic variability. Table: 3.2 demonstrates the mean and standard deviation for the
duration of all cycles and total cycles ENSO and active region of the ENSO cycle with parameters
of test probability distributions. The mean and standard deviation come out to be for ENSO cycles
128
is ((0.49464 ±1.021), (0.52708 ±1.097) and (0.17989 ±1.107) respectively (table 3.2). The result
(Tables 3.1- 3.2) indicate that mean variation of Sunspots cycles maximum as compared to ENSO
cycles. As figure 3.4 depicts the plot of twenty two and total ENSO cycles plot from 1866 to 2012
cycles, 1st to 5th, 7th to 11th, 13th to 21st follow Generalized Extreme value distribution while cycle
6thand 12thlogistic and generalized Pareto probability distribution show a tail prolonged.
3.5: COMPARATIVE ANALYSIS OF BEST FITTED DISTRIBUTION OF ENSO
ACTIVE REGION AND SUNSPOT CYCLES IN SAME DURATION
It confirms that the period 1980-2000 of ENSO cycles were very active. El Nino was active for
the periods 1982-83, 1986-87, 1991-1993, 1994-95, and 1997-98 these periods include two
strongest periods of the century viz. 1982-83 and 1997-98. When the ENSO active period was
going on, meanwhile the sunspot had cycle 21 (May1976 - Mar 1986), cycle 22 (Mar1986 -
Jun1996) and cycle 23 (Jun1996 - Jan 2008). In this study we intend to analysis the complexity of
the ENSO cycle in the same duration of the sunspot cycle.
In Table 3.3 explore the most active ENSO cycles (May1976 - Mar 1986), (Mar1986 - Jun1996)
and (Jun1996 - Jan 2008) which are mean and standard deviation come out to be three active ENSO
cycles are ((0.49464 ±1.021), (0.52708±1.097) and (0.17989 ±1.107) have positive correlation
and less standard error for these active ENSO cycles, the table results indicates that the Generalized
Extreme value probability distribution become increasingly heavy tailed behavior. In the same
duration sunspots followed the Generalized Pareto distribution (GPD). The nature of the
Generalized Pareto distribution (GPD) reveals that the behavior of time series data is a heavy tail
behavior, strong correlated and long range persistence. Figure 3.5 depicted that same duration of
sunspots and ENSO time series cycles (May1976 - Mar 1986), (Mar1986 - Jun1996) and (Jun1996-
Jan 2008) and probability distributions in one domain.
129
Figure 3.5: same duration of sunspot and ENSO cycles and Probability Distribution of same cycles
of sunspot and ENSO cycles.
130
3.6: PROBABILITY DISTRIBUTION OF FRACTAL DIMENSION OF SUNSPOT
CYCLES AND ENSO CYCLES
This section explores the probability distribution of self-similar and self-affine fractal dimension
of sunspot cycles (1-23 and total 1-23) and ENSO cycles (1-22 and total 1-21) with the help of
checking Kolmogorov–Smirnov test (KST), Anderson-Darling test (ADT) and Chi-Square test
(CST). The table 3.4 indicates the best fit probability distribution of self-similar and self-affine
fractal dimension and the parameter of the mean and standard deviation are also calculated in these
distribution of sunspots and ENSO cycles. The self-similar and self-affine fractal dimension are
also followed Generalized Extreme Value distribution (GEV) and Generalized Pareto distribution
(GPD).
Table 3.4. Shows that the self-similar fractal dimension (FDS) of sunspot cycle followed the
Generalized Extreme Value distribution (GEV) with (1.1928 ± 0.0818). The self-affine fractal
dimension (FDA) of sunspot cycle indicate Generalized Extreme Value distribution (GEV) with
(1.3606± 0.102). Similarly the self-affine fractal dimension (FDA) of the ENSO cycle followed
the Generalized Extreme Value distribution (GEV) with (1.117± 0.0911). Whereas the self-
similar fractal dimension (FDS) of the ENSO cycle followed the Generalized Pareto distribution
(GPD). These distributions are becoming an increasingly right side heavy tailed and persistence.
In the end a relation between probability distribution and fractal dimension persistency is presented
in table 3.4. All the sunspots and ENSO cycles probability distributions indicate persistency. The
research also concludes that twenty four cycles of sunspots will more prolong if it has greater
means and the Gen. Extreme Distribution tail prolongs. Figure 3.6 displayed that the probability
distributions of self-similar (FDS) and self-affine (FDA) fractal dimensions of sunspot cycles.
Figure 3.7 depicted that the probability distributions of self-similar (FDS) and self-affine (FDA)
fractal dimensions of ENSO cycles.
131
Figure 3.6: Probability distributions of self-similar (FDS) and self-affine (FDA) fractal dimensions
of sunspot cycles
Figure 3.7: Probability distributions of self-similar (FDS) and self-affine (FDA) fractal dimensions
of ENSO cycles
3.7: SOLAR ACTIVITY CYCLES AND ENSO ACTIVITY CYCLES IN THE
PERSPECTIVE OF HEAVY TAILS PARAMETER
The heavy tail parameter strongly based on the self-similarity parameter HE which is also called
as Hurst exponent. In the persistent noise, the heavy tail parameter β will be range from 0 to 2.
Whereas, the long-range dependence parameter d is valid in the range from 0 to 1- 1
𝛽. This condition
is applicable if HE < 1. In the stationary time series, the above condition is mandatory. For the
132
finite variance case HE = d + 1
2 , whereas for infinite variance case HE = d +
1
𝛽. The long-range
dependence degree is given by (Feldman and Taqqu, 1998).
HE = (3−𝛽 )
2 (3.15)
The heavy tail parameter (β) range depends on the differencing parameter (d). Case 1, if d > 0,
then the heavy tail parameter (β) will be greater than 1. The limitation of β for any noise is range
from 0 to 2 (Gourieroux and Jasiaky, 1998). Case 2, if d < ½, then it represents the power series
expansion for all the sunspot cycles. Whereas, β < 2, then the tail is asymptotically equivalent to
Pareto law (Sun et al., 2005). For all sunspots cycle β has ranged from 1 to 2. This study develops
the heavy tail parameter (β) calculate by using Hurst exponent in both method self-similar (HES)
which is calculated by Hausdorff – Besicovich method (Box counting Technique) and self-affine
(HEA) obtained by rescaled range analysis technique. dS and dA represented as differencing
parameter of self-similarity and self-affinity respectively. βS is known as self-similar heavy tail
parameter, whereas βA is called as self-affine heavy tail parameter Table 3.5 shows the fractional
differencing parameter and heavy tails parameter of sunspot cycles (1-24) in the perspective of
Hurst exponent.
Table 3.5 indicates that the heavy tail parameter values within range from 1 to 2 which is shows
that the dynamic is more regular. The differencing parameter value of all 24 sunspot cycle (24
cycle represents that complete Aug 1755 –Jan 2008 data) is less than 0.5. The heavy tail parameter
(β) is less than 2 indicate that asymptotically equivalent to Pareto law. The heavy tail parameter
(β) shows that the strength of the dynamics is regular and periodic for all the solar cycles. The
table 3.5 depicted that the solar cycle 5 has smallest value (1.198) of the heavy tails parameter (βS)
whereas, the solar cycle 3 has largest value (1.656) of the heavy tails parameter (βS). Whereas, the
table 3.5 shown that the solar cycle 21 has smallest value (1.374) of the heavy tails parameter (βA)
whereas, the solar cycle 3 and the solar cycle 17 have largest value (2.000) of the heavy tails
parameter (βS). For each sunspot cycle heavy tails are profound. The heavy tail parameter (β) and
differencing parameter (d = HE-0.5) is obtained from Hurst parameter. The heavy tail parameter
(β) value towards 2 depicted that the strength of heavy tail decreases. The novelty of the study
concludes that the heavy tail parameter of self-similar (βS) is more long time persistent and
correlated as compared to self-affine heavy tails parameter (βA).
133
Best fitted probability distribution of maximum ENSO cycles followed Generalized Extreme value
distribution (GVT) and despite of the ENSO cycle shows Generalized Pareto distribution (GPD).
These distributions have heavy tail behavior. For all ENSO cycles β has ranged from 1 to 2. This
study explores the fractional differencing parameter and heavy tails parameter of the ENSO cycles
(1-23) in the perspective of Hurst exponent of self-similar (HES) and self-affine (HEA). Table 3.6
depicts the tail analysis of ENSO cycles.
Table 3.6 develops the heavy tail parameter values (β) within the range from 1 to 2 which are show
that the dynamical theory are more regular and persistent. The differencing parameter value of all
23 ENSO cycle (22 cycle indicates the duration of an active region of the ENSO cycle, whereas,
23 cycle are complete duration 1866-2012) is less than 0.5. The value of the heavy tail parameter
(β) is less than 2 represent the asymptotically equivalent to Pareto law. The heavy tail parameter
(β) reveals that the strength of the dynamics is regular and periodic for all the ENSO cycles. The
table 3.6 explored that the ENSO cycles 22 and 23 have the smallest value (1.000) of the heavy
tails parameter (βS) whereas, the ENSO cycle 1 has largest value (1.118) of the heavy tails
parameter (βS). Whereas, the table 3.6 shown that the ENSO cycle 11 has smallest value (1.002)
of the heavy tails parameter (βA) whereas, the ENSO cycle 3 has largest value (1.720) of the heavy
tails parameter (βA). For each ENSO cycle heavy tails are profound. The novelty of the study
concludes that the heavy tail parameter of self-similar (βS) is more long time persistent and
correlated as compared to self-affine heavy tails parameter (βA). The heavy tail parameter (βS) of
the ENSO cycle values near to 1 which is indicate that ENSO cycle data behave heavy tail increase
as compared to sunspots time series data. Figure 3.8 (a and b) self-similar and self-affine tail
parameter of Sunspot Cycles respectively. Similarly, figure 3.9 (a and b) self-similar and self-
affine tail parameter of ENSO Cycles respectively.
134
Figure 3.8 (a): Self-similar Tail Parameter of Sunspot Cycles
Figure 3.8 (b): Self-Affine Tail Parameter of Sunspot Cycles
135
Figure 3.9 (a): Self-similar Tail Parameter of ENSO Cycles
Figure 3.9 (b): Self-Affine Tail Parameter of ENSO Cycles
3.8: SOLAR ACTIVITY CYCLES AND ENSO ACTIVITY CYCLE IN THE
PERSPECTIVE OF LONG RANGE CORRELATION:
In the statistical modeling of time series, considered two types of correlation. One is a short range
correlation (persistency) (Priestlay 1981, Box et al. 1999) and another is a long range correlation
(persistency) (Beran 1994, Taqqu and Samorodintsky 1992). Short range correlation (persistency)
is considered by a decay in the autocorrelation function that is bounded by a decreasing of an
exponential for large lags, whereas, long range correlation (persistency) in time series (explicit
subclass often referred to as 1/f noise or functional noise) in which any given value is related by
136
all pervious values of the time series. Long range correlation (persistency) is considered by a
decreasing of the Power Law (asymptotic and exact) of all the persistency between values as a
function temporal distance (or lags) between them. The long range correlation can be defined as,
maximum values are correlated to each other at strong, long lags in time (Beran 1994, Taqqu and
Samorodintsky 1992). In the variation of spatially and temporally aggregated variable can be
recorded by the long range persistence (Beran 1994)
Long-range persistence technique follows a Power-Law scaling of the autocorrelation function
[C (ζ) = 1
𝜎𝑥2(𝑁−𝛾)
∑ (𝑥𝑘 − �̅�)(𝑥𝑘+ζ − �̅�)𝑁𝑘=1 ]
Such that
|𝐶(ζ)|~ ζ−(1−𝛾) , ζ→ ∞ , -1 < 𝛾 < 1 (3.16)
Holds for large time lags ζ. Where 𝛾 is known as the strength of long-range persistency. If 𝛾 = 0
depict no long-range persistence between values. If 𝛾 > 0 long-range persistent and 𝛾 < 0 long-
range anti-persistent. In the study, we intents to calculate the strength of long-range persistence
with the help of self-similar and self-affine Hurst exponent. Hurst exponent self-similar (HES)
which is calculated by Hausdorff – Besicovich method (Box counting Technique) and self-affine
(HEA) obtained by rescaled range analysis technique. The self-affine (HEA) is correlated to the
strength of long-range correlation 𝛾 as
𝛾𝐴 = 2HEA – 1 -1 < 𝛾𝐴< 1 (3.17)
(Malamud and Turcotte, 1999). The self-similar (HES) is associated with the strength of long-range
correlation 𝛾 as
𝛾𝑆= 2HES + 1 1 < 𝛾𝑆 < 3 (3.18)
(Burrough 1981 and 1983, Mark and Aronson 1984) . If 𝛾𝑆= 2 depict no long-range persistence
between values. If 𝛾𝑆 > 2 long-range persistent and 𝛾𝑆< 2 long-range anti-persistent.
The novelty of this study to analysis the long range correlation of sunspot cycles (1-24) in the
perspective of self-similar and self-affine Hurst exponent and also compare both of them. The table
3.7 depicts the strength of long term persistence of each solar activity (sunspot cycles).
The table 3.7 explores the analysis of the strength of long-range correlation of the sunspot cycles.
The self-similar strength of long-range correlation (𝛾𝑆) range from 1 to 3, whereas, the self-affine
strength of long-range correlation (𝛾𝐴) range from -1 to 1. The value of 𝛾𝑆 > 2 shows that long-
range persistent whereas, 𝛾𝑆 < 2 indicate long-range anti-persistent. On the contrary, the value of
137
𝛾𝐴 > 0 shows that long-range persistent whereas, 𝛾𝐴 < 0 indicate long-range anti-persistent. The
table 3.7 shows that each sunspot cycle has 𝛾𝑆 > 2 which are indicate that every sunspot cycle is
strongly correlated to previous one. Similarly, each sunspot cycle has 𝛾𝐴 > 0 which are indicate
that every sunspot cycle is strongly correlated to previous one, despite of sunspot cycle 3 which is
show that the value of 𝛾𝐴 = 0. In this case, 𝛾𝐴 = 0 shows that no long-range persistence between
values. The conclude of this study that self-similar 𝛾𝑆 has strong strength long-range correlation
as compared to self-affine 𝛾𝐴.
In this study, we also analyze that the strength of long-range correlation of ENSO cycles (1-23).
The table 3.8 explores the correlation of each ENSO cycle in the perspective of self-similar 𝛾𝑆 and
self-affine 𝛾𝐴.
The table 3.8 depicts the analysis of the strength of long-range correlation of the ENSO cycles.
The self-similar strength of long-range correlation (𝛾𝑆) range from 1 to 3, whereas, the self-affine
strength of long-range correlation (𝛾𝐴) range from -1 to 1. The value of 𝛾𝑆 > 2 shows that long-
range persistent whereas, 𝛾𝑆 < 2 indicate long-range anti-persistent. On the contrary, the value of
𝛾𝐴 > 0 shows that long-range persistent whereas, 𝛾𝐴 < 0 indicate long-range anti-persistent. The
table 3.8 shows that each ENSO cycle has 𝛾𝑆 > 2 which are indicate that every ENSO cycle is
strongly correlated to previous one. Similarly, each ENSO cycle has 𝛾𝐴 > 0 which are indicate that
every ENSO cycle is strongly correlated to previous one. The conclusion of this study that self-
similar 𝛾𝑆 and self-affine 𝛾𝐴 both have long-range correlation persistent behavior which is
representing that the value of each cycle is correlated to preceding one. Figure 3.10 (a and b) self-
similar and self-affine Strength of long range-correlation of Sunspot Cycles respectively.
Similarly, figure 3.11 (a and b) self-similar and self-affine Strength of long range-correlation of
ENSO Cycles respectively.
138
Figure 3.10 (a): Self-similar Strength of long range-correlation of Sunspot Cycles
Figure 3.10 (b): Self-Affine Strength of long range-correlation of Sunspot Cycles
139
Figure 3.11 (a): Self-similar Strength of long range-correlation of ENSO Cycles
Figure 3.11 (b): Self-Affine Strength of long range-correlation of ENSO Cycles
3.9: UNITE ROOT STOCHASTIC PROCESS OF TAIL PARAMETER AND LONG
RANGE CORRELATION OF SUNSPOT AND ENSO CYCLES
The random walk model
Yt = ρYt-1 + ut −1 ≤ ρ ≤ 1 (3.19)
In this model Yt belongs to the time series and ut represent the white noise error while ρ is the
coefficient. The equation (6) is used to regress the Yt on the lagged value Yt-1estimate for unit root
test. If the predictable value ρ identical to 1 statistically. The series comprises unit root and null
hypothesis rejected (McCarthy, 2015). Transform the equation 6 by subtraction Yt-1 both sides
140
Yt−Yt -1 = ρYt -1− Yt -1 + ut
Yt−Yt -1 = Yt -1(ρ− 1) + ut
∆ Yt = ɸYt -1 + ut (3.20)
Where ∆ and ɸ represent the operation of the first difference. The null hypothesis (ɸ = 0) for time
series non stationary condition represents the unit root (da silve Lopes, 2006).
The Augmented Dickey Fuller (ADF) test to solve the autocorrelation problem. This test is
implemented by augmenting three following equations, the dependent variable’s (∆Yt) adding
lagged values and Yt represents a random walk with
∆ Yt = 𝛾1+ 𝛾2t + ɸ Yt -1 + ut (3.21)
Where 𝛾2t and 𝛾1represents the intercept and trend respectively.
∆ Yt = ɸYt -1 + ut (3.22)
It shows neither intercept nor trend.
∆ Yt = 𝛾1+ ɸ Yt -1 + ut (3.23)
It shows intercept (𝛾1) only (Kwiatkowski et al., 1992; McCarthy, 2015).
We assume the null hypothesis H0: variables have a unit root (non-stationary data). The condition
where the H0 is rejected if p-value is less than 5% or the critical value of the absolute value of
Augmented Dickey Fuller (ADF) test is greater than at 1% and 5% significance level.
The Augmented Dickey Fuller (ADF) test for stationary explored that the tail parameter and the
strength of long range-correlation of self-similar and self-affine sunspot (1755 to 2008) and ENSO
cycles (1866 to 2012) time series data. In case the data is not stationary than behave like a unit
root. Table 3 (a), 3 (b), 4 (a) and 4 (b) depicted that the tail parameter of self-similar (βS) and self-
affine (βA) of sunspots and ENSO cycles has stationary nature and unit root test rejected. Similarly,
table 5(a), 5(b), 6(a) and 6(b) analyzed that self-similar strength of long-range correlation (𝛾𝑆) and
self-affine (𝛾𝐴) of sunspot and ENSO cycles has stationary nature and unit root test rejected. The
reason to reject H0 is p-value is less than 5% or the critical value of the absolute value of
Augmented Dickey Fuller (ADF) test is greater than at 1% and 5% significance level. These
conditions are satisfied in both cases.
141
3.10: CONCLUSION
Different methods have been used to develop the certainty of significant relations among the
Sunspot cycles and some of the terrestrial climate parameters such as temperature, rainfall and
ENSO etc. This study explores the dependence of ENSO cycles on Mean Monthly Sunspots
Cycles. Sunspot cycles range from 1755 to 2008 whereas, ENSO cycles range from 1866 to 2012.
To find the above mentioned dependence probability distribution approach is utilized. In this
regards the appropriateness of distributions is investigated with the help of Kolmogorov–Smirnov
D, Anderson-Darling and Chi-square tests. It is found that most of the sunspot cycle follows
Generalized Pareto Distribution (GPD) whereas, Generalized Extreme Value Distribution (GEV)
were found appropriate for ENSO cycles. This study confirmed that during the period 1980-2000
ENSO cycles were very active. Simultaneously, El Nino was active for the periods 1982-83, 1986-
87, 1991-1993, 1994-95, and 1997-98 these periods include two strongest periods of the century
viz. 1982-83 and 1997-98. Two consecutive periods 1991-1993 and 1994-1995 were cold periods.
Sunspots cycles and ENSO cycles both were found to be persistent. This research is a part of a
larger research project investigating the correlation of Sunspot cycles and ENSO cycles and the
influence of ENSO cycles on variations of the local climatic parameters which in term depend on
solar activity changes. In the time series data, sunspot and ENSO cycles have stressed the
Generalized Pareto Distribution (GPD) and Generalized Extreme Value Distribution (GEV).
These distributions have a heavy tail on the right side.
In the next section of this chapter described the analysis of heavy tail parameter for further analyze
the data behavior. All the solar cycles (1-24) has a stationary nature as the differencing parameter
(0 < d < 0.5) in both perspective self-similar (dS) and self-affine (dA) which represent that the
dynamic is more regular. The heavy tail parameter βS as well as βA exploring that asymptotically
equivalent to Pareto law which is showing that the strength of the dynamics is regular and periodic
for all the solar cycles. For each sunspot cycle heavy tails are profound. The heavy tail parameter
(β) and differencing parameter (d = HE-0.5) are obtained from the Hurst parameter (0.5 < HE < 1)
persistent. The heavy tail parameter (β) value towards 2 depicted that the strength of heavy tail
decreases. Similarly, every persistent data contains the heavy tail since for d > 0 the HT parameter
β > 1. Similarly, All the ENSO cycles (1-23) also has a stationary nature as the differencing
parameter (0 < d < 0.5) in self-similar (dS) and self-affine (dA) respectively. The heavy tail
parameters (βS and βA) of ENSO cycles are depicted asymptotically equivalent to Pareto law. The
142
comparative study of solar cycle and the ENSO cycle conclude that the heavy tail parameter (βS)
of the ENSO cycles, values approximately 1 which are explored that ENSO cycle data behave
heavy tail increase as compared to sunspots time series data. This study concludes that ENSO
cycles have a more heavy tail as compare to sunspot cycles.
In the time series data, the tail parameter helps to analysis the persistency and long term
dependency. Statistical modeling based on two types of correlation short range correlation and
long range correlation. All sunspot cycles explored the strength of long-range correlation (𝛾). The
strength of self-similar long-range correlation (1 < 𝛾𝑆 < 3) and the self-affine strength of long-
range correlation (-1<𝛾𝐴 < 1) is persistent in the perspective of 0.5 < HES < 1 and 0.5 < HEA < 1.
The novelty of this study shows that every value of sunspot cycles is strongly correlated to
preceding ones in both manner self-similar as well as self-affine. Similarly, each ENSO cycle
shows that each value is strongly correlated to preceding ones in both manner self-similar (𝛾𝑆)
and self-affine (𝛾𝐴). In all aspects Self-similar techniques are more appropriate as compared to
self-affine.
The unit root test is used for non-stationary data. H0 is rejected when the p-value is less than 5%
or the critical value of the absolute value of Augmented Dickey Fuller (ADF) test is greater than
at 1% and 5% significance level. This study analyzed that the heavy tail parameter (βS and βA) of
Sunspot and ENSO cycles are stationary. Similarly long-range correlation (𝛾𝑆 𝑎𝑛𝑑 𝛾𝐴) of sunspot
and ENSO cycles are also stationary in time series data.
143
CHAPTER 4
A STUDY OF ACTIVE REGIONS OF ROTATING SUNSPOTS: IMAGE
PROCESSING APPROACH
This chapter demonstrates the image processing of active regions of solar cycles. Image analysis
is a widely used in image processing technique to analysis the image features such as smoothness,
texture, roughness, area, solidity and persistency. The novelty of this chapter intends to enhance
the study of solar active region by using fractal dimensions, mathematical morphology and Image
segmentation of rotating sunspots using Genetic active contour.
24 solar cycle had been starting from January 08, 2008 with the observation of high latitude of
sunspot pair. Although 24 cycle is considered to be a weak cycle because of below average
intensity. It has maximum 90 sunspots number, but it has some large sunspots. Solar maximum
exists in May 2014.There are no sunspots in the duration of 2008 and 2009. The situation is very
unusual which were not happening for almost a century. In 24 solar cycle, northern hemisphere
has following sunspots of active regions with negative polarity field while southern hemisphere
has leading sunspots with positive polarity. In the 24 cycle, AR12192 of October 18, 2014 hosts
the largest sunspot. According to NASA scientist C. Alex Young and Dean Pasnell demonstrate
that AR12192 has ranked 33rd largest of 32908 active region since 1874. Sunspot AR12192 as
same size as Jupiter. Figure 4.1 depicted that the active region AR 12192.
144
Figure 4.1: Active region 12192 AR
This study intends to enhance the study of image analysis and mathematical morphology of the
sunspots. The morphology of sunspot groups not only is a predictor of their own future evolution,
but also of the other associated explosive events like solar flares and coronal mass ejections
occurring higher in the solar atmosphere (Stenning D et al 2010). Further, fractal dimensions, the
space filling properly of the images will be calculated to study the images of sunspots and sunspot
groups. Fractals are developed by Mandelbort (1983). Fractal techniques have been used to analyze
the complexity and quantification of irregular shapes of the geometric structure of the active
region. The fractal dimension is an ideal tool for measuring texture, roughness and persistency.
Images define in two dimension function (x, y). Whereas x and y are spatial coordinates. Fractal
features is a new approach to describe segmentation of the texture gray-scale images (kaspair et.
al 2001).
The theory of mathematical morphology consists of functional and morphological transformation.
Most of the operations of mathematical morphology (MM) are reversible and linear which are
directly related to the shapes. The theory was introduced with binary images by Matheron (1975)
and Serra (1982). Mathematical morphology (MM) is the theory and technique for the analysis and
processing of geometrical structures and for describing shapes using set theory, lattice theory,
topology and random functions.
145
Rotating sunspots are usually observed in active region, which is correlated with some solar
eruption events like High flare activity and coronal mass ejection (CME). Observations indicate
that some sunspots rotate about their own vertical axis. Rotating sunspots has both direction
northern hemisphere (shows anticlockwise) and southern hemisphere (shows clockwise). The
magnetic helicity injected due to the rotational motion of sunspots. The magnetic helicity
associated with hemisphere pattern. The positive and negative sign of magnetic helicity represents
that solar magnetic field in the northern and southern respectively hemisphere.
4.1: DATA DESCRIPTION AND METHODOLOGY
The present study investigates the image processing of solar active region by using fractal
dimensions and mathematical morphology (MM). Fractal dimension is an essential parameter of
fractal geometry that define the significant application in many fields including image processing.
The novelty of this study to compare both fractal dimensions methods Hausdorff - Besicovich box
dimension and correlation dimension belong to self-similar family. The largest active region
AR12192 of solar cycle 24 are used to evaluate fractal dimensions and Mathematical morphology
(MM). Moreover, this study is an attempt to provide a new way of contouring the rotating sunspots
by means of a genetic algorithm which is based on the actual observation. The proposed scheme
is supported by Temperature, Energy, magnetic Helicity and Tangential velocity, and the features
of the sun, which are essential in the formation of sunspots. In this study the active region AR9114
and AR10696 as a case study to analysis the active contour of rotating sunspot by using image
segmentation. For this purpose, we analyze 4 active contours of each active region with variations
of temperature with constant other parameters. MATLAB 2016 is used to calculate transformation
of mathematical morphology. FRACTALYSE 2.4 software is used for the computation fractal
dimension and Hurst exponents and universal scaling parameters.
4.1.1: Fractal Dimensions
Self-similar fractal dimension is defined as the geometric object is composed into a union of
rescaled copies of itself with uniform in all directions or rescaling isotropic. Hausdorff -
Besicovich box dimension or simply box dimension and correlation fractal dimension are belong
to the self-similar fractal dimension family. The plane of Active region has Euclidean dimension
d = 2. The smallest unit of the active region (AR) is considered as a point so the topological
146
dimension of active region is considered to be dt = 0. So the fractal dimension of active region
(AR) range from dt = 0 to d = 2.
The Hausdorff - Besicovich box dimension or simply box counting dimension methods are widely
used to calculate the fractal dimensions. It can be calculated by taking a ratio between log changes
in object size and log change in measurement scale (Paul 2005) (FD = 𝑙𝑜𝑔𝑁𝑡
𝑙𝑜𝑔(1
𝑡) ).In the box counting
method, the smallest number of squares of size t is known as reference element which are covered
all black pixels contained in a counting window. Where t is the roughly counting of black pixels
which are existing in the window. Nt is the union of reference element or the size of counting
window. In the infinite time, the algorithm of the box counting method converges to the minimum.
Therefore, the results are only an approximation of the adequate and best coverage. This method
is the generalized form of grid method.
Correlation Dimension was developed by Grassberger and Procaccia (1983) are used to calculate
fractal dimension measurements of a reconstructed attractor. Correlation dimension is closely
related to Hausdorff - Besicovich box dimension. The correlation integral C (I) is defined as
C ( I ) = 2
(𝑛+1)(𝑛) ∑ ∑ ℵ(𝐼 − ‖𝑋(𝑖) − 𝑋(𝑗)‖)𝑛−𝑖
𝑗=𝑖𝑛𝑖=1 (4.1)
Where n = N – (En - 1)η is the embedded number points is a En – space. N represent the length of
the data series and I be the radius of the sphere centered object X (i) or its size of the box. Whereas,
ℵ(𝑥) is the Heaviside step function. The function is defined as
ℵ(𝑥) = 1 𝑥 ≥ 00 𝑥 < 0
(4.2)
If C (I) scale behave like C (I) = ID, where D represent that the correlation dimension of time series
data (x).
Fractal dimension D and Hurst exponent H have relation 𝐹D + HE = 2. Hurst exponents lie
between 0 and 1. If the value of H approaches to 1 it is shown that the time series data is quasi-
regular (persistent). If it approaches to zero represent that irregular (anti-persistent)
(salakhudinova, 1998; Hanslmeier, 1999). Brownian process shows if H = 0.5. The value ranges
between 0 and 0.5, reveal a random walk (anti-persistent) process. The values of H range from 0.5
to 1 shows persistent positively correlated whereas anti persistent negatively correlated. Universal
scaling laws are also analysis in this study. The scaling exponent is described as β = d – FD.
Whereas, β range forms 0 to 1and d > FD. The spectral exponent α = 5- 2FD or α = 2HE + 1. The
relation between the fractal dimension (FD) and spectral exponent (α) was introduced by (Higuchi,
147
1988). α = 0 describes a white noise-like system shows that uncorrelated and the power spectrum
is independent of the frequency. α =1 is represented as flicker or 1/f noise system which express
as a moderate correlated. α = 2 is called Brownain noise-like system, which shows a strong
correlation. The relation between spatial autocorrelation coefficient (C∇) and HE based on fractal
brownain motion (fBm) can be given as follows C∇ = 22HE-1 – 1. Where C∇ represents the spatial
autocorrelation coefficient. Spatial autocorrelation is used to measure the correlation of spots with
itself through active region. It should be noted that the value of HE = ½ means that C∇ = 0 which
is considered to be the Brownian motion. If HE > ½ then C∇ > 0 representing that positive spatial
autocorrelation exists. HE < ½ implies that C∇ < 0 which indicates that the spatial autocorrelation
is negative.
4.1.2: Heavy Tail Analysis
The heavy tail parameter strongly based on the Hurst exponent (HE). In the persistent noise, the
heavy tail parameter β will be range from 0 to 2. Whereas, the long-range dependence parameter
d is valid in the range from 0 to 1- 1
𝛽. This condition is applicable if HE < 1. In the stationary time
series, the above condition is mandatory. For the finite variance case HE = d + 1
2 , whereas for
infinite variance case HE = d + 1
𝛽. The long-range dependence degree is given by (Feldman and
Taqqu, 1998).
HE = (3−𝛽 )
2 (4.3)
The heavy tail parameter (β) range depends on the differencing parameter (d). Case 1, if d > 0, then the
heavy tail parameter (β) will be greater than 1. The limitation of β for any noise is range from 0 to 2
(Gourieroux and Jasiaky, 1998). Case 2, if d < ½, then it represents the power series expansion for all the
sunspot cycles. Whereas, β < 2, then the tail is asymptotically equivalent to Pareto law (Sun et al., 2005).
For all AR 12192 β has ranged from 1 to 2. This study develops the heavy tail parameter (β) calculate by
using Hurst exponent in both method Hausdorff – Besicovich method or Box counting Technique (HEB)
and correlation dimension (HEC). dB and dC represented as a differencing parameter of Box counting
technique and correlation dimension respectively. βB is known as heavy tail parameter of box counting,
whereas βC is called as correlation dimension heavy tail parameter.
148
4.2: FRACTAL DIMENSIONS, WAVE SPECTRUM AND HEAVY TAIL ANALYSIS OF
AR12192
The novelty of this research is to analyze the complexity and roughness of largest AR 12192 of 24
cycle by fractal dimensions. This calculation is based on scaling analysis and spectral analysis of
AR 12192. Fractal parameters are calculated using box counting (FDB) and correlation dimension
techniques (FDC) lying in a significant range.
Table 4.1: universal scaling parameters and heavy tail analysis of AR12192
Hausdorff - Besicovich box dimension
FDB HEB αB C∇B dB βB
1.336 0.664 2.328 0.2553 0.164 1.672
Correlation Dimension
FDC HEC αC C∇C dC βC
1.15 0.85 2.7 0.6245 0.35 1.30
Table 4.1 demonstrates that the values of FDB and FDC in AR12192 range from 1 to 1.5 which
indicate that AR is persistent, correlated and predictable. The value of FDB was found to be greater
than FDC in AR12192. Similarly, Hurst exponents HEB and HEC both values are from 0.5 to 1
which reveals the persistency. The relationship between HEB is greater than HEC. Since, the fractal
dimension lies between 1 and 2, the scaling exponent β ranges from 0 to 1. This range for β assures
it is significant. Table 1 describes the numerical relation between the spectral exponent (α) and
autocorrelation coefficient (C∇) which are calculated by using both the box counting method and
the correlation dimension. If 1< DF < 2 then the spectral exponent α has a range from 1 to 3. α =
0 is indicates a white noise-like system which describes uncorrelated behavior. α =1 is called
flicker which represents a moderately correlated behavior. α = 2 is called Brownain noise-like
system and shows strong correlation. Spectral exponent (αB is box counting method and αC is
correlation dimension) reveals that the cycles behave like Brownain noise. The autocorrelation
coefficient (C∇) describes the multiple-lag 1-dimension spatial autocorrelation of AR12192. The
autocorrelation coefficient (C∇) is calculated by the relation of Hurst exponent. The value of C∇
= 0 if HE = ½ which describes Brownian motion. If HE > ½ then C∇ > 0 indicates positive spatial
149
autocorrelation and if HE < ½ then C∇ < 0 represents the negative spatial autocorrelation. The
autocorrelation coefficients (C𝐵∇ and CC∇) show the positive spatial autocorrelation in AR12192.
This study estimates the long-range dependence parameters dB and dC. The heavy tail parameter
(βB and βC) ranges are influenced by the differencing parameter (dB and dC). Table 4.1 shows the
fractional differencing parameter and heavy tails parameter of AR12192 in the perspective of Hurst
exponent (HE). Table4.1 depicts that the heavy tail parameter values within range from 1 to 2
which is shows that the dynamic is more regular. The differencing parameter value of AR 12192
is less than 0.5. The heavy tail parameter (β) is less than 2 indicate that asymptotically equivalent
to Pareto law (nature of heavy tail any side). The heavy tail parameter (β) shows that the strength
of the dynamics is regular and periodic for AR 12192. The heavy tail parameter (β) value towards
2 depicted that the strength of heavy tail decreases. The novelty of the study concludes that the
heavy tail parameter of βB is more long time persistent and correlated as compared to heavy tails
parameter βC. Figure 4.2 indicates a sketch map of the numerical relationship of box counting and
the correlation dimension of AR12192.
Figure 4.2: indicates a sketch map of the numerical relationship of AR12192
4.3: MATHEMATICAL MORPHOLOGY (MM)
Image processing deals with the manipulation and analysis of the available image information. In
image processing, mathematical morphology is used to investigate the interaction between an
image and a certain chosen structuring element using the basic operations of erosion and dilation.
Mathematical morphology stands somewhat apart from traditional linear image processing, since
the basic operations of morphology are non-linear in nature. It over all deals with image
enhancement algorithms for reconstructing the relative order, geometry, topology, patterns and
150
dynamics. The three stages involved in the study of images are image processing, image analysis
and image understanding.
The theory of mathematical morphology consists of functional and morphological transformation.
Most of the operations of mathematical morphology (MM) are reversible and linear which are
directly related to the shapes. The theory was introduced with binary images by Matheron (1975)
and Serra (1982). Mathematical morphology (MM) is the theory and technique for the analysis and
processing of geometrical structures and for describing shapes using set theory, lattice theory,
topology and random functions. It is used to analyze the evolution of images. The main concept
of mathematical morphology (MM) which is copied by the following quotation from (J Serra 1982)
“The perception of the Geometric structure, or texture which is not purely
objective. It does not exist in the phenomenon itself, nor in the observer, but
somewhere in between the two. The mathematical morphology (MM)
quantifies this intuition by developed the idea of structuring elements.
Chosen by the morphology, they interact with the object under study.
Modifying its shape and reducing it to a sort of caricature which is more
expensive than the actual initial phenomenon….”
The transformation of morphological of an image (subset ℤn and Rn) is described as prescribed
manner intersection and union. Constitutes of translation vector is known as structuring element.
The geometric information regarding objects by sequential application of morphological
transformation involving to select structuring elements (the number of possibilities is unlimited).
The morphological transformation has irreversibility: less information provides transform image
as compared to real one. In the discrete case, the transformation works as a cellular automata or
cellular logic. The basic mathematical morphological tools are dilation, erosion, opening and
closing.
4.3.1: Dilation and Erosion, Closing and Opening
The basic mathematical morphological tools are dilation, erosion, opening and closing. In this
study, E be the Euclidean space Rn or discrete space ℤn. Whereas, E is a commutative group. ɸ (E)
is a space which is consists of all subsets of E. A subset S of E is the binary image. Moreover, S is
called the object and ɸ (E) is known as object space. Let S ⊂ E and t 𝜖 E the St is known as
translated by S along h: St = {s + t: s 𝜖 S}. If S, R ⊂ E then S hits R, S R. If S ∩ R ≠ ∅. The
151
dilation of a set S by structuring element A is demonstrated by S ⊕ A = {t 𝜖 E: At S}. Whereas,
the erosion of S by A is described as S ⊖A = {t 𝜖 E: At ⊂ S}. Erosion as an impact of shrinking
bright region in the output image. So the eroded image is the subset of the input image. Dilation
has an expanding bright region in the output image. Another two important features, invariant
translation transformation ɸ (E) are the closing and the opening. The closing of a set S by
structuring element A is demonstrated by SA = (S⊕A) ⊖A. Whereas, the opening of S by A is
described as SA = (S ⊖A) ⊕A. The opening and closing has a relation with S such as SA ⊂ S ⊂
SA.
4.4: MATHEMATICAL MORPHOLOGY (MM) OF AR12192
The active region AR12192 shows the distribution complexity of structure with different levels of
intensity and no regular pattern. So, the other novelty to analyze the boundary edges of AR12192.
For this purpose the mathematical morphology basic principal such as erosion, dilation, closing
and opening are applying of AR12192. Figure 4.3 (a, b, c, d) demonstrated that mathematical
morphological operations such as erosion, dilation, closing and opening of AR12192. Figure 4.2
(a) express erosion of AR12192 which shows the subset of actual sunspot. Whereas, Figure 4.3 (b)
expresses dilation of AR12192 which shows the expand edge of actual sunspot. The two important
features such as opening and closing are depicted in figures 4.3 (c and d) respectively show that
the relation which are described that actual sunspot is the subset of closing. Whereas, opening
operation is the sunset of actual sunspot. The purpose of mathematical morphology is the edge
detection of irregular and complex distribution of objects.
152
Figure 4. 3(a): Erosion of Active region AR12192 Figure 4.3(b): Dilation of Active region AR12192
Figure 4.3(c): Opening the Active region AR12192 Figure 4.3(d): Closing the active region AR12192
4.5: IMAGE SEGMENTATION OF ROTATING SUNSPOTS USING GENETIC ACTIVE
CONTOUR
Sunspots are associated with high active power regions, which increased magnetic flux on the sun
locally (Kevin et. Al, 2014). It has been observed by many authors that there is a rotational motion
of and around the sunspots and due to this feature it is easier to identify and observe the rotating
behavior sunspots. The magnetic polarity of sunspots being changed with each solar cycle
(approximation 11 years). The phenomenon of rotating sunspots was introduced by J. Evershed in
153
1910. The direction of rotating sunspots, rotation in the northern hemisphere indicates anti-clock
wise whereas the rotation of southern hemisphere shows clock wise (Hale, 1927; Bao et. al 2002).
The magnetic helicity are usually associated with hemisphere pattern. The negative and positive
sign of magnetic helicity represents that solar magnetic field in the northern and southern
respectively hemisphere (Pevtsov, 2008). The magnetic helicity has important property is that
conversed in high magnetic Reynolds number plasma, which is also presence even in the deceptive
process (Berger et. al 1984). Magnetic helicity is exists in the solar convection zone, which is
exists is below to the photosphere. The rotating sunspots indicate that different angular and
tangential velocity with radius. The velocity of rotating is calculated by the twist and the
occurrence rate of the twisted flux-tube from lower than the photosphere (Brown et. al 2003). They
are considered to influence many terrestrial environmental and climatic phenomena (TJO News,
2006). Determination of the area of rotating sunspots is important because of their role in the study
of the evolution of rotating sunspots and their effect with solar irradiance (Barangi et. al 2001).
The rotational motion of sunspots is round, its own axis in the plane of the Photosphere. This
rotational motion of the sunspots, which are not usually of a regular circular shape, but frequently
elongated (Knoska., S. 1975).
Segmentation of images is considered as one of the most important and earliest stages of image
processing and serves as a vital role in quantitative and qualitative analysis of sunspots. This study
is intended to provide the image segmentation of rotating sunspots together with a useful technique
of genetic algorithm. The proposed methodology involves the active contouring which has been
one of the widely used methodologies for image segmentation due to the rapid rotational motion
of sunspots. The generation of sunspots depends upon the two components (1) the accumulated
magnetic Helicity (2) Homologous flares that occur in the immediate periphery of the rotating
sunspots. It has been observed by many authors that the evaluation of sunspots reflects the transport
of magnetic energy and the magnetic Helicity from the sub atmosphere to Corona. In case of
unipolar patches the proposed method considers the magnetic Helicity along with the rotating
sunspots. Magnetic Helicity can be viewed as a quantitative measure of global chiral properties of
magnetic field in the atmosphere. For the approximation of sunspots as unipolar patches the
magnetic Helicity injected by the rotational motion is.
𝑑𝐻
𝑑𝑡= −
1
2𝜋∅2(𝜔+ + 𝜔−) (4.4)
154
Where 𝜔+and 𝜔−are the angular speed of the positive and negative polarities, respectively, and ∅
is the magnetic flux in equation (4.4).
We are considering only the magnetic Helicity injected by the rotational motion of sunspots and
due to their two homologous flares that occurred in the immediate periphery of the rotating
sunspots and showed an inverse S shape after the ribbon extension. Due to the rotational motion
of sunspots the accumulated magnetic Helicity can be obtained by the equation (4.5),
∆𝐻= −1
2𝜋𝜔 ∙ ∆𝑡 ∙ ∅2 (4.5)
Another associated parameter in the formation of sunspots is energy loss, due to which there is a
sudden change in the temperature of the active region which ranges from 4200 to approximately
4600. For the implementation of genetic algorithm the above mentioned parameters will be highly
appropriate. As described earlier, we have made our calculations on the bases of actual
observations in the active regions. As described by (Attila, 2010) there is heavy loss of energy due
to the magnetic flux and the tangential velocity of rotational sunspots. Since we are dealing with
the rotating sunspots, therefore the loss of energy will be considered as one of the most important
parameters of the rotating sunspots in this study. According to (Attila, 2010) the loss of energy is
ranges from (5.7 1.9) 1033 Ergs and during the active regions of sunspots this loss ranges from
(2.6 0.11) 1033 to (2.6 0.11) 1034ergs.
4.5.1: Description of Algorithm
Homogenous flares are occurring to indicate the periphery of the rotating sunspots. By considering
the rotational motion and calculating the magnetic Helicity are certainly meant that the magnetic
Helicity is injected due to the rotational motion of Sun spots. By considering the contouring point
of each circle we calculate the fitness values of overall contour points to remedy the problem of
local minima. The region is called active region. We can consider only the magnetic Helicity
injected by the rotating sun spots and due to this two homologous flares that occurred in the
indicate periphery of the rotating sun spots and showed an inverse S shape after the ribbon
extension.
The aim of our study is providing a versatile, intelligent tracking algorithm for the identification
and tracking of sunspots together with the motivation of study solar evaluation of sunspots that
make a valuable contribution towards the advance studies of the rotation of sunspots.
155
4.5.2: Genetic Algorithm with Active Contours
Genetic Algorithm is a heuristic approach that has captured the attention of many of many
researches, although this algorithm does not provide the surety for optimal solution but with the
help of these algorithms. We may find the near optional solution which may close to optimal
solution in any case. By using the genetic algorithm we obtained the best, most optimal solution
among all the other existing solution. Genetic algorithm starts its operation from the initial
population also called the initial set of chromosomes. These chromosomes are encoded by either
a real number or binary number. Each chromosome is a set of real coded genes or binary coded
gene depends upon the type of the problem. Firstly, we decided the size of the population and the
nature chromosomes, which it is evaluated by using the fitness function further. The fitness
function is the most important factor of the genetic algorithm also known as the bone of GA. The
fitness function is based upon the factor that affect the initially generated chromosomes and the
fitness function will decided which chromosomes is better individual and which is not able to go
for the further produce. After the selection the better individuals and selected chromosomes will
go under the procedure of crossover which provided the most fitted child chromosomes as a better
individual. This process is further enhanced with mutation where the local optima converge to the
global optima. The stopping criteria of genetic algorithm will decide when this procedure will stop
producing further initial population according to the pre decided rule. Figure 4.4 and 4.5 show the
complete picture of genetic algorithm.
Figure 4.4: show that the Active contour of rotating Sunspots
4.5.3: Initial Population
156
For the initialization process, here we are applying the active contouring of rotational sun spots as
given in (Kevin et. al, 2014) with a little modification. This algorithm will apply in several steps.
The process will start under these assumptions that the rotation spot is confined within the two
circles, having the minimum radius and maximum radius. The maximum circle with a maximum
radius will cover the complete sunspot of any shape, therefore the region under consideration will
be studied as a concentric circle and the center of this concentric circle will be considered as a
center of the spot of any shape. Each of them will act like a contour as show in the figure 4.3. Each
of these contours contains points from the sun spots edge. Here we can try to separate these points
from each contour and connect them in such away so that we can able to reveal the Sun spots edge.
Initially we generate the population as an initial population to generate these circles with the
inclusion of two genes for every chromosome. The first gene indicates the radius of the circle and
the second gene will specify the radius of each point. After the formation of the contours we
indicate the number of equidistant points on every one. If N denotes the number of points on the
contour, the distance of every point on the contour, the distance of every point from the adjacent
one would be 360/N. The above procedure will help in determining the contour points and now
the next step is to evaluate these contour points for the further selection.
4.5.4: Fitness Function
After the formation of the entire contour and the determination of contour points, each of these
contour points are going to be analysis for the selection of the best one for the further procedure.
As described earlier the evaluation must be done through the fitness function:
Due to the rotational motion of sunspots the Helicity is induced and due to this the internal energy
is released of that portion which directly affect the temperature as well. For calculating the fitness
function we are considering following important features of sunspots:
∅ = Tangential velocity or angular velocity of the selected contour point.
H= Helicity of the initial chromosomes
T= Temperature of the selected contour point
E= energy losses during the time period
Emax = Maximum energy losses during the time period
∅𝑚𝑎𝑥 = Maximum Tangential velocity or angular velocity during the time period
Hmax = Maximum Helicity of the initial chromosomes during the time period
157
The Average of temperature will be considered in the calculation since due to the rotational motion
the size the sunspot may decrease or increase that the temperature may not be constant. Therefore
the average of temperature will be considered for the calculation in each contour.
Tavr = 𝑡1+ 𝑡2+𝑡3+𝑡4
4 (4.6)
The idea is to calculate the fitness value of each point one by one from each contour. We calculate
the energy for selected point and selection for next procedure will be done by using the predefined
criteria. After the selection of points from each contour, further procedure of crossover will be
applied.
The above features are mainly involved in any rotated sunspots therefore, our fitness function will
definitely base on these parameters. According to the above features the suggested fitness function
is defined as,
𝐹𝑣𝑎𝑙𝑢𝑒𝑠 = 𝐻∅
𝐸𝑇𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
(4.7)
If the above value is nearer to 0, then that individual is supposed to be more fitted chromosomes
and will be able to go for the further procedure of crossover. After calculating the overall fitness
function from an equation (4.7), we select the contour point that possesses best value according
to our selection criteria as the best solution.
4.5.5: Crossover
At this step the selected contour points from each circle are going to place into the reduction pool
and the members of this pool are paired up and reproduce themselves in order to have best child
chromosomes. We call this procedure, a crossover operation which creates better individual from
the selected parents with the help of the fitness function.
To implement the crossover operation, we will generate the string of each selected contour point
belongs to specific circle which we initially generated in the first step of genetic algorithm. In this
way we consider two selected strings at a time to produce two fitted child chromosomes. Here we
use the two point crossover procedure to generate our children and after that we apply the process
of mutation. Figure 4.5 shows the procedure of crossover for the proposed algorithm.
4.5.6: Mutation
158
In order to prevent from the convergence and to have better generation, we apply the process of
mutation. This will help us to obtain the global optimal from the local optima. For the mutation
procedure, we have to set some stopping criteria, usually the process of mutation will automatically
stop when there is no further improvement is possible. In this algorithm the mutations are done by
simply changing the bit of a string and replace it by the bit of the second string. The resultant
string will be further checked by the fitness function until no more improvement is possible.
Parent 1:
P11 P12 P13 P14 P15 P16 P17
Parent 2:
P21 P22 P23 P24 P25 P26 P27
Child 1:
P21 P12 P13 P24 P25 P26 P27
Child 2:
P21 P22 P13 P14 P15 P16 P17
Figure 4.5: Showing the crossover procedure of two parenet choromosomes
159
Figure 4.6: is showing the crossover operation
4.6: ACTIVE REGION AR9114 and AR10696
In this study, we consider the rotating sunspots of 23 cycle of four active region AR9114 and
AR10696 as a case study. In 23 cycle, identified 186 rotating sunspots in 153 active regions
observed by SOHO / Mirchclson Dophler image (MDI) and TRACE. The active region AR9114
and AR10696 were injected negative total helicity and located in the northern hemisphere.
AR9114 is located rotation in the northern hemisphere. The leading polarity has a strong magnetic
field. Whereas, the following negative polarity has the appearance of discontinuous. The maximum
tangential velocity of AR9114 is 4O /h (0.13 km /s). The negative helicity dH-/dt is about -5 1040
Mx2 / h. Whereas, the positive helicity dH+ /dt is about -2.4 1041 Mx2 / h. The total negative
helicity injection dH-/dt is about -3.4 1043 Mx2 / h from the photosphere to the upper solar
atmosphere and the total positive helicity injection dH+/dt is approximately -7 1043 Mx2 / h in
this duration when the leading rotating sunspot has maximum tangential velocity. AR10696 is a
160
strong, active region (AR) which consist of rotating sunspots. The maximum tangential velocity
of AR10696 is 3O /h (0.098 km /s). The negative helicity dH-/dt is greater than -8 1041 Mx2 / h.
Whereas, the positive helicity dH+ /dt is less than -2 1041 Mx2 / h. The total result is negative
helicity injection dH-/dt is about -6.3 1043 Mx2 / h and total result is dH+/dt is approximately -
5.5 1043 Mx2 / h. The novelty of this study for analyzing the edge detection of rotating sunspot
with the same energy released and temperature of each contour. In this study, the random values
of helicity and tangential velocity are taken to leading and following sunspot which is near to
average values. The temperature and energy released in each active region are considering to be
constant. This study examines the first contour of each active region of rotating sunspot.
In this study, four contours are analyzed with variations of temperature and the rest of parameter
is constant. The temperature of rotating sunspot of very contour to contour which contour near to
the center has minimum temperate as compare to others. Genetic algorithms selects that sufficient
value which is nearer to zero, as per the predefined rules, different points will be selected from
each circle. The crossover will further filtrate the circles, points and mutation will provide the
global optima. The calculations are based on most of the actual observations as given in (Zhu ),
the considered energy loss is supposed to be (2.6 0.11) 1033engrin in our calculations and
temperature effect will be from 4200K to 4600K. The active region AR9114 is located in the
northern hemisphere, with strong magnetic field. AR9114 considering the values of parameters
with negative helicity as follows,
-Hmax = -3.4x1043, ∅𝑚𝑎𝑥= 4, Emax = 0.24x1033 and Tavg = 4350 so −𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔 = -1.30 x 1012
Similarly, AR9114 with positive helicity the maximum values are
+Hmax = -7 x1043, ∅𝑚𝑎𝑥= 4, Emax = 0.24x1033 and Tavg = 4350 so +𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔 = -2.68 x 1012.
The table 4.2 (a, b, c and d) depicted that according to fitness criteria each value of AR9114 with
negative helicity is nearly to zero which shows that each point exactly locate in the contour.
Whereas, the values are far from 0 shows that the observation does not exist in the active contour.
The table 4.3 (a, b, c and d) describes that the behavior of AR9114 with positive helicity. Every
value is fulfilling the fitness criteria has value less than 1. The first observation of each contour
exactly locates in contour with the closest value approaches to zero. This study concludes that the
161
image segmentation using genetic active contour of rotating AR9114 with negative helicity is more
appropriate than the results with positive helicity.
Similarly, the active region AR10696 is located in northern hemisphere, with strong magnetic
field. The maximum parameter values of AR10696 with negative helicity describe as,
-Hmax = -6.3x1043, ∅𝑚𝑎𝑥= 3, Emax = 0.24x1033 and Tavg = 4350 so −𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔 = -1.80 x 1012
And AR10696 with positive helicity the maximum values are
+Hmax = -5.5x1043, ∅𝑚𝑎𝑥= 3, Emax = 0.24x1033 and Tavg = 4350 so +𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔 = -1.58 x 1012
The table 4.4 (a, b, c and d) depicted that according to fitness criteria each value of AR10696 with
negative helicity is nearly to zero which shows that each point exactly locate in the active contour.
The table 4. (a, b, c and d) describes that the behavior of AR10696 with positive helicity and
maximum tangential velocity. Every value is fulfilling the fitness criteria has a value less than 1.
First observation of each contour exactly locates in contour with closest value approaches to zero.
Image segmentation using genetic active contour of rotating AR10696 with negative helicity is
found to be more suitable than the results with positive helicity. From the observation of the values
related to sunspots, we may conclude that the values which are greater than one is outside the
contour and contour point with value one will exists on the edge of sunspots. In this way a new
contour will be generated from the above calculation as shown in figure 4.7. By taking the above
mentioned values, promising results came out from these observations.
162
Figure 4.7: The steps of Active contours of AR9114
4.7: CONCULSION
This chapter reveals the persistency of AR12192 by using self-similar fractal dimensions involving
Hurst exponent. The fractal dimension (FD) represents the complexity and roughness of the active
region AR12192, whereas Hurst exponent (HE) provides the smoothness of the active region. This
study investigates the relationship between self-similar fractal dimensions FDB and FDC. FDB is
calculated by the Box counting method and FDC is calculated by using the Correlation Dimension
method. The Hurst exponents are calculated by FD which is calculated by both techniques. The
spectral exponent and the autocorrelation coefficient are calculated by both techniques. The active
region AR12192 has a stationarity nature as the differencing parameter follows the inequality 0 <
d < 0.5 in both perspective Box counting (dB) and correlation dimension (dC). This in turn shows
that underlying dynamic is more regular. The heavy tail parameter (β) is less than 2 which confirms
the equivalence of the asymptotic nature of heavy tail (one sided) and the Pareto law which
confirms that the underlying dynamics is strong and regular. For the he active region AR12192
heavy tails are profound. The heavy tail parameter (β) and differencing parameter (d = HE-0.5)
are obtained from the Hurst parameter (0.5 < HE < 1) showing persistency. The heavy tail
parameter (β) tending towards 2 depicts that strength of heavy tail is decreasing. If the data is
persistent, then heavy tails exist. This implies that d > 0 and HT parameter β > 1 heavy tails exist.
Table 1: demonstrates the numerical relationship between Hurst exponents, spectral exponents and
autocorrelation coefficients and differencing parameters and the outcome of the heavy tail analysis
of AR12192. The active region AR12192 is found to be persistent, correlated and heavy tailed.
The spectral exponent (αB and αC) of AR12192 behaves like Brownain noise which indicates the
long term dependency. The autocorrelation coefficient is found to be significant using both the
163
fractal dimensions FDB and FDC. Mathematical morphological operations such as erosion, dilation,
closing and opening are also analysis for AR12192.
The proposed study has provided image segmentation of rotating sunspots with the help of genetic
algorithm. We have used the criteria of selection of best contour points from the proposed fitness
criteria and then improvements have been made by the process of crossover and mutation which
are the most important features of Genetic algorithm (GA). Although GA is an approximated
approach, but it may provide very accurate results by making a suitable selection of crossover and
mutation. In the proposed algorithm we have applied crossover in a different way between the
contour points of each cycle the proposed idea is unique and producing promising results. AR9114
and AR10696 of solar cycles 23rd are used as a case study to calculate the image segmentation of
active contour.
164
CHAPTER 5
CONCLUSIONS AND FUTURE WORK
This chapter comprises the results obtained from this dissertation. According to the chapter
sequential, principal results are mentioned. Suggestions and future work are exhibited at the end
of this chapter.
5.1: SUMMARY AND OUTLOOK
Sunspot cycles have influenced on earth climates. In the time series, sunspot cycles in second
difference have stationary nature. Autocorrelation (AC), Partial Autocorrelation (PAC) and Ljung-
Box Q-statistics test are used for checking white noise in solar cycles. Moreover, the unit root test
with Augmented Dickey Fuller (ADF) test has applied for verification of stationary. For a selection
of appropriate models, diagnostic checking is used. For checking normality of sunspot cycles test
of normality are used. Test of normality based on skewness, kurtosis and Jurque-Bera test. This
chapter is utilizing the stochastic autoregressive and moving average (ARMA) modeling, AR-
GARCH (1, 1) process and ARMA-GARCH (1, 1) models and forecast evolution of sunspot
cycles. Least Square Estimation is used for ARMA process. Various best fitted ARMA models
estimate and forecast for each sunspot cycle. Least square method is used to calculate ARAM
models and quasi maximum likelihood estimation (QMLE) are used to calculate AR-GARCH and
ARMA-GARCH models. The selection of ARMA, AR-GARCH and ARMA-GARCH models are
focused on smallest value of Durbin-Watson statistics test. Durbin-Watson (DW) statistics test
value of each sunspot cycle is less than 2 which shows that sunspot observations are correlated to
each other. GARCH (1, 1) stationary volatility model has the best forecasting model as compared
with other models. Diagnostic checking is used to identify and estimation of most appropriate
models and confirmation is found by forecasting evolution. The Gaussian quasi maximum
likelihood estimation (QMLE) is used to calculate AR-GARCH and ARMA-GARCH models.
ARCH effect is found by checking Lagrange Multiplier test, correlogram squared residuals and
test of normality. Forecasting evolutions are verified by Root Mean Squared Error (RMSE), Mean
165
Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and Theil’s U-Statistics test (U
test). AIC, SIC and HQC, maximum log likelihood estimation are also calculated. GARCH (1, 1)
is leptokurtic in both perspective AR (p) and ARMA (p, q) for sunspot cycles. The adequate
ARMA, AR-GARCH and ARMA-GARCH modeling for fractional Brownian motion of sunspot
cycles will be useful to predict the dynamical variables for 24th cycle and next coming cycles in
the future. The comparison of forecasting evolution of ARAM, AR-GARCH and ARMA-GARCH
process are also discussed. RMSE, MAE and U test are described that ARMA model is an
appropriate model for sunspot cycles. MAPE exhibits that ARAM-GARCH model is the most
appropriate model of sunspot cycles.
The persistency analysis of solar activities and ENSO cycles by using fractal dimension and Hurst
exponent. The fractal dimension (FD) represents the roughness and complexity of the time series
data of solar activities (sunspot activity) and ENSO cycles. Whereas Hurst exponent (HE) provides
the smoothness of the data. This study investigates the relationship between self-similar fractal
dimension (FDS) and self-affine fractal dimension (FDA). FDS calculated by Box counting method
and self-affine fractal dimension (FDA) is calculated by using rescaled range method. The Hurst
exponents also calculated by FD in both methods, self-similar fractal dimension (FDS) and self-
affine fractal dimension (FDA). The spectral exponent is calculated by using equation (2.13) and
the autocorrelation coefficient to calculate by using equation (2.16). The table 2.1: indicate the
numerical relationship between Hurst exponents, spectral exponent and autocorrelation coefficient
of both fractal dimensions self-similar and self-affine. In the sunspots cycles indicate that each
cycle is persistent and correlated, the values of Hurst exponent lies from 0.5 to 1 but HES values
are greater than HEA. All values of the spectral exponent (αs and αA) of each sunspot cycles behave
like Brownain noise which indicates the long term dependency. The autocorrelation coefficient is
also lies in a significant range of both fractal dimensions. The relation of self-similar and self-
affine fractal dimensions which is described in equation (2.15) in failed in each sunspot cycles
because this relation is valid if fractal dimension is lies from 1.5 to 2 but each sunspots cycles
fractal dimension lies from less than 1.5. The spectral exponent value which is calculated by
Higuchi’s Fractal dimension (FDH) also identify the strong correlation among Sunspots Cycles.
We analyzed the complexity of each cycle of ENSO data along total cycles and active ENSO
period, then compared them by estimating self-similar fractal dimension (FDS) by using Box
counting method and self-affine fractal dimension (FDA) by rescaled range method, and Hurst
166
exponents also calculated through FD which is show that self-similar fractal dimension (FDS) for
ENSO cycles and Total data are less than those for self-affine fractal dimension (FDA). Similarly,
self-similar fractal dimension (FDS) of ENSO cycles is also less than self-similar fractal dimension
(FDs) of Sunspot cycles. This means that ENSO cycles are more persistent than Sunspot cycles
data. Sunspots cycles and ENSO are correlated to each other. It is also verifying the fact if FD
increases, then H decreases. The cycle will prolong sunspot activity has greater means and the tail
prolongs. In the end a relation between probability distribution and Fractal Dimension establishes
the persistency approach for both of the data sets. The mean- tail assessment confirm the FD-HE
analysis. This study can be useful for further investigation of the impact of Sunspot and ENSO
related local climatic variability.
Different methods have been used to develop the certainty of significant relations among the
Sunspot cycles and some of the terrestrial climate parameters such as temperature, rainfall and
ENSO etc. This study explores the dependence of ENSO cycles on Mean Monthly Sunspots
Cycles. Sunspot cycles range from 1755 to 2008 whereas, ENSO cycles range from 1866 to 2012.
To find the above mentioned dependence probability distribution approach is utilized. In this
regards the appropriateness of distributions is investigated with the help of Kolmogorov–Smirnov
D, Anderson-Darling and Chi-square tests. It is found that most of the sunspot cycle follows
Generalized Pareto Distribution (GPD) whereas, Generalized Extreme Value Distribution (GEV)
were found appropriate for ENSO cycles. This study confirmed that during the period 1980-2000
ENSO cycles were very active. Simultaneously, El Nino was active for the periods 1982-83, 1986-
87, 1991-1993, 1994-95, and 1997-98 these periods include two strongest periods of the century
viz. 1982-83 and 1997-98. Two consecutive periods 1991-1993 and 1994-1995 were cold periods.
Sunspots cycles and ENSO cycles both were found to be persistent. This research is a part of a
larger research project investigating the correlation of Sunspot cycles and ENSO cycles and the
influence of ENSO cycles on variations of the local climatic parameters which in term depend on
solar activity changes. In the time series data, sunspots and ENSO cycles have stressed the
Generalized Pareto Distribution (GPD) and Generalized Extreme Value Distribution (GEV).
These distributions have a heavy tail on the right side.
In the next section of this chapter described the analysis of heavy tail parameter for further analyze
the data behavior. All the solar cycles (1-24) has a stationary nature as the differencing parameter
(0 < d < 0.5) in both perspective self-similar (dS) and self-affine (dA) which represent that the
167
dynamic is more regular. The heavy tail parameter βS as well as βA exploring that asymptotically
equivalent to Pareto law which is showing that the strength of the dynamics is regular and periodic
for all the solar cycles. For each sunspot cycle heavy tails are profound. The heavy tail parameter
(β) and differencing parameter (d = HE-0.5) are obtained from the Hurst parameter (0.5 < HE < 1)
persistent. The heavy tail parameter (β) value towards 2 depicted that the strength of heavy tail
decreases. Similarly, every persistent data contains the heavy tail since for d > 0 the HT parameter
β > 1. Similarly, All the ENSO cycles (1-23) also has a stationary nature as the differencing
parameter (0 < d < 0.5) in self-similar (dS) and self-affine (dA) respectively. The heavy tail
parameters ( βS and βA) of ENSO cycles are depicted asymptotically equivalent to Pareto law. The
comparative study of solar cycle and the ENSO cycle conclude that the heavy tail parameter (βS)
of the ENSO cycles, values approximately 1 which are explored that ENSO cycle data behave
heavy tail increase as compared to sunspots time series data. This study concludes that ENSO
cycles have more heavy tail as compare to sunspot cycles.
In the time series data, the tail parameter helps to analysis the persistency and long term
dependency. Statistical modeling based on two types of correlation short range correlation and
long range correlation. All sunspot cycles explored the strength of long-range correlation (𝛾). The
strength of self-similar long-range correlation (1 < 𝛾𝑆 < 3) and the self-affine strength of long-
range correlation (-1<𝛾𝐴 < 1) is persistent in the perspective of 0.5 < HES < 1 and 0.5 < HEA < 1.
The novelty of this study shows that every value of sunspot cycles is strongly correlated to
preceding ones in both manner self-similar as well as self-affine. Similarly, each ENSO cycle
shows that each value is strongly correlated to preceding ones in both manner self-similar (𝛾𝑆)
and self-affine (𝛾𝐴). In all aspects Self-similar technique is more appropriate as compared to self-
affine.
The unit root test is used for non-stationary data. H0 is rejected when p-value is less than 5% or
the critical value of absolute value of Augmented Dickey Fuller (ADF) test is greater than at 1%
and 5% significance level. This study analyzed that the heavy tail parameter (βS and βA) of Sunspot
and ENSO cycles are stationary. Similarly long-range correlation (𝛾𝑆 𝑎𝑛𝑑 𝛾𝐴) of Sunspot and
ENSO cycles are also stationary in time series data.
The persistency of AR12192 is determined by using self-similar fractal dimensions (Box counting
FDB and Correlation Dimension FDC) involving Hurst exponent. The fractal dimension (FD)
expresses the complexity and roughness of the active region AR12192, whereas Hurst exponent
168
(HE) provides the smoothness of the active region. The Hurst exponents are calculated by FD
which is calculated by both techniques. The spectral exponent and the autocorrelation coefficient
are also calculated by both techniques. The active region AR12192 has a stationarity nature as the
differencing parameter follows the inequality 0 < d < 0.5 in both perspective Box counting (dB)
and correlation dimension (dC). The heavy tail parameter (β) is less than 2 which confirms the
equivalence of the asymptotic nature of heavy tail (one sided) and the Pareto law which confirms
that the underlying dynamics is strong and regular. For the he active region AR12192 heavy tails
are profound. The heavy tail parameter (β) and differencing parameter (d = HE-0.5) are obtained
from the Hurst parameter (0.5 < HE < 1) showing persistency. The heavy tail parameter (β) tending
towards 2 depicts that strength of heavy tail is decreasing. If the image is persistent, then heavy
tails exist. This implies that d > 0 and heavy tail parameter β > 1 heavy tails exist. The active region
AR12192 is found to be persistent, correlated and heavy tailed. The spectral exponent (αB and αC)
of AR12192 behaves like Brownain noise which indicates the long term dependency. The
autocorrelation coefficient is found to be significant using both the fractal dimensions FDB and
FDC. Mathematical morphological operations such as erosion, dilation, closing and opening are
also analysis for AR12192.
The novelty of this study has delivered image segmentation of rotating sunspots with genetic
algorithm. The criteria of selection of best contour points from the proposed fitness criteria and
then improvements have been prepared by the process of crossover and mutation which are the
most important features of Genetic algorithm (GA). Although GA is an approximated approach,
but it may provide very accurate results by making a suitable selection of crossover and mutation.
In the proposed algorithm we have applied crossover in a different way between the contour points
of each cycle the proposed idea is unique and producing promising results. AR9114 and AR10696
of solar cycles 23rd are used as a case study to calculate the image segmentation of active contour.
5.2: PRINCIPAL RESULTS
The principal results of this dissertation are obtained and discussed which is very useful for the
world of scientific community is demonstrated as follow:
169
1. Describe the stationarity of the time series sunspot cycles (1-24) at second difference with
the help of Autocorrelation (AC), Partial Autocorrelation (PAC) and Ljung-Box Q-
statistics test.
2. Stationarity is confirmed by applying unit root test with Augmented Dickey Fuller (ADF)
test for sunspot cycles.
3. AIC, SIC, HQC and maximum likelihood are estimated for ARMA, AR-GARCH and
ARMA-GARCH models
4. Various ARMA (p, q) models are selected for sunspot cycles. Selection is based on the
minimum value of Durbin-Watson statistics test.
5. Selection for AR-GARCH and ARMA-GARCH models are based on the minimum value
of Durbin-Watson statistics test.
6. AR (p)-GARCH (1, 1) model is estimated for sunspot cycles.
7. ARMA (p, q)-GARCH (1, 1) model is estimated for sunspot cycles.
8. ARCH effect is diagnostic by Lagrange multiplayer, correlogram squared residuals and
test of normality are used to verify ARCH effect in time series data.
9. The least square estimator is used for estimation ARMA models and the Gaussian quasi
maximum likelihood estimator is used to estimate AR-GARCH and ARMA-GARCH
models.
10. Forecasting evolution is used to estimate residual diagnostics checking techniques such as
RMSE, MAE, MAPE and U test for ARAM, AR-GARCH and ARMA-GARCH models.
11. Comparison of residuals diagnostics checking are used for ARAM, AR-GARCH and
ARMA-GARCH models.
12. Fractal dimension and Hurst exponent (self-similar and self-affine) are estimated for
sunspot and ENSO cycles along with the total time series to understand the long term
behavior.
13. Dynamics of sunspot cycles and ENSO cycle’s active episode are compared with the help
of fractal dimension (self-similar and self-affine).
14. Universal scaling parameters like scaling exponent, spectral exponent and autocorrelation
coefficient are estimated by self-similar and self-affine fractal dimension for sunspot
cycles.
170
15. Higuchi’s Fractal dimension (FDH) are used to estimate the spectral exponent for sunspot
cycles.
16. The adequate probability distributions for sunspot and ENSO cycles associated are
obtained and selected with the help of Kolmogorov-Smirnov test.
17. Dynamics of sunspot and ENSO active period compare the adequate probability
distributions with the help of Kolmogorov-Smirnov test.
18. The adequate probability distributions for fractal dimension of sunspot and ENSO cycles
associated are obtained and selected with the help of Kolmogorov-Smirnov test.
19. The heavy tails and fractional differencing parameters (self-similar and self-affine) for
sunspot cycles and ENSO cycles are estimated to understand the peak and strength of each
cycle.
20. Self-similar and self-affine Long-Range correlation (persistency) for sunspot and ENSO
cycles are estimated.
21. Stationarity of heavy tail analysis and Long-Range correlation are confirm by applying unit
root test with Augmented Dickey Fuller (ADF) test for sunspot cycles and ENSO cycles.
22. The complexity and roughness of largest active region AR12192 of 24 solar cycle is
calculated by using fractal dimensions (Box counting FDB and correlation dimension FDC).
23. Persistency and wave spectrum or universal scaling parameters such as scaling exponent,
spectral exponent and autocorrelation coefficient are estimated for AR12192.
24. The principles of Mathematical morphology such as dilation, erosion, opening and closing
are analyzed for AR12192.
25. Image Segmentation of rotating sunspots by using Genetic active contour for AR9114 and
AR10696 of solar cycle 23rd.
5.3: Future Outlook
On the behalf of the principles results obtained in this dissertation some future works and
suggestions are focused. Direction of future studies are mentioned below:
• Forecasting evolution of ENSO cycles are established.
• The effective relationship between ENSO and rainfall and sunspot and rainfall, which are
not discussed in this thesis will be studied.
171
• The spectrum scaling parameter of ENSO cycles and rainfall (self-similar and self-affine)
will be analyzed.
• Time series analysis of ENSO cycles. ARMA (p, q), AR-GARCH and ARMA-GARCH
models are determined for ENSO cycles.
172
REFERENCES:
Akgiray V. (1989). Conditional Heteroskedasticiy in Time Series of Stock Returns: Evidance and
Forecast. Journal of Business, 62, 55-80.
Almeida A, Guse A, Mello MGS, Martin IM, Pugacheva G, Pankov VM, Spjeldvik WN, Schuch
NJ, (2004). Rainfall cycles with bidecadal periods in Brazilian region”, Geofisica
Internacional, Vol. 43, pp. 271-279.
Andersen T. G. and T. Bollerslev (1998). Answering the skeptics: yes, standard volatility models
do provide Accurate Forecasts. International Economics Review 39: 885-905.
Anderson R.y.(1992). Possible connection between surface winds, solar activity and the earth’s
magnetic field. Nature 358, 51-53
Anderson R.y. (1990). Solar cycle of modulation of ENSO: a mechanism for specific and globel
climate change. In betan court, J.L and Mackey, A.M.(eds), proceeding Six annual pacific
climate (PACLIM) workshop march 5-8, 1989, California Department of water resources,
interagency Eclolgical studies program technical report 23, 27-81.
Anderson JD, Lieberman M, Stewart RN (1979). Ethylene production by apple protoplasts. Plant
Physiol63: 931-935.
Annette witt, Bruce D. and Malamud (2013). Quantification of Long-Range Persistent in
Geophysical Time series: Conventional and Benchmark- Based Improvement Techniques Sun
Geophys. Dol 10.1007/s 10712-012-9217-8.
Aquilar O. and M. west (2000). Bayesian Dynamic Factor Models and Portfolio Allocation.
Journal of Business and Economic Statistics. 18(3).
173
Arif, M., H. Askari and A. Rashid, (1993). Influence of El Nino on Pakistan Rainfall, Research
Study Report, Pakistan Meteorological Department, Karachi. Pp. 25.
Babcock H. W. (1961). The Topology of the Sun’s Magnetic Field and the 22-years cycle.
Astrophysical Journal, 133, 572-578.
Bal S, Bose M,“A. (2010). Climatological study of the relations among solar activity, galactic
cosmic ray and precipitation on various regions over the globe”, J. Earth Syst. Sci., Vol. 119,
pp. 201–209.
Baranyi T. Gyori L. Ludmany A and Coffey H E (2001). Comparison of Sunspots Area Data Base.
Mon. Not R. Astron Soc. 323, 223-230.
B. B. Mandelbrot (1983). The Fractal Geometry of Nature, W. H. Freeman and Company, New
York, NY, USA.
Benoit B. Mandelbrot (1999). Multifractals and 1/f Noise: Wild Self-Affinity in Physics (1963–
1976), Springer, New York, NY, USA.
Beran J (1994) Statistics for Long-memory Processes. Chapman and Hall/CRC, New York.
Bhalme, H.N. and S.K. Jadhav, (1984). The southern oscillation and its relation to the monsoon
rainfall, Journal of Climatology, 4, 5: 509-20.
Burlaga L F and Klein L W (1986). Fractal structure of the interplanetary magnetic field J.
Geophys. Res. 91A1 347-350.
Bunde, A., &Havlin, S. (1994). A brief introduction to fractal geometry (pp. 1-26).Springer Berlin
Heidelberg.
Bollerslev T. (1986). Generalized Autoregressive Conditional Heteroskedasticiy. Journal of
Econometrics. 31: 307-327.
174
Box GEP, Jenkins GM, Reinsel GC (1994). Time Series Analysis: Forecasting and Control. 3rd
edn. Prentice Hall, Englewood Cliffs.
Box GEP, Pierce DA (1970). Distribution of residual autocorrelation in autoregressive integration
moving average time series model. J Am stat Assoc 65: 1509-1526
Box GEP, COX DR (1964) An analysis of transformation. J R stat Soc series B stat Methodol 26:
211-252
Burrough PA (1981). Fractal Dimension of Landscape and Other Environmental Data. Nature 294:
240-242
Burrough PA (1983). Multi-scale Sources of Spatial Investigation in Soil. I. the application of
fractal Concepts to nested level of Soil variation. J soil sci 34: 577-597
Castillo E and A.S. Hadi (1997). Fitting the Generalized Pareto Distribution to Data. JASA 92
(440): 1609-1620
Castillo E and Daoubid J (2008). Estimate of the Generalized Pareto Distribution. Statistics and
Probability Letters, in press.
Chaudhary, Q.Z., A.B. Farooqi and G. Rasul, (1998). Tendency of Pakistan's Current Climate and
Climate Change Scenarios: Climate Change Impact Assessment and Adaptation Strategies in
Pakistan, UNEP. Pp. 31
Chen Z, Ivanov P, Hu K and Stanley H E (2002) Phys. Rev. E 65041107.Climate glossary-
southern Oscillation Index (SOI), Bureau of meteorology (Australia) 2002-04-03. Retrieved
2009-12-31.
Cole, J. E., Dunbar, R. B., McClanahan, T. R., & Muthiga, N. A. (2000).Tropical Pacific Forcing
of Decadal SST Variability in the Western Indian Ocean Over the Past Two
Centuries. Science, 287(5453), 617-619.
175
Cryer J. D. and Chan K.S. (2008). Time Series Analysis with Application is R. Springer, 2nd
Edition U.S.A.
Danish H, Shaheen A, M.R.K Ansari, Bulbul J (2014). Solar Flares Data Analysis on application
of Probability Distributions and Fractal Dimensions and a comparative analysis of North-
South Hemispheric Solar Flares Data Behavior. Proceedings of the Pakistan Academy of
Sciences 51 (4): 345-353.
Danish H, Shaheen A, M.R.K Ansari, Bulbul J (2014).The Study of Sunspots and K-Index Data
in the Perspective of Probability Distributions. International Journal of Physical and Social
Sciences Volume 4, Issue 1.ISSN: 2249-5894.
Demon P. E., (1988), Production and decay of radio carbon and its modulation by geomagnetic
field-solar activity changes with possible implication for global environment. In Stephenson,
F. R. and wolfendala , A. W. (eds) secular, solar an geomagnetic variation in the last 10000
years. Dordrecht, Kluwer, 267-285
E. Forg´acs-Dajka, B. Major, and T. Borkovits (2014). Long-term variation in distribution of
sunspot groups. Astronomy and Astrophysics, v.424, p.311-315
Enfield, D. B., & Cid S, L. (1991). Low-frequency changes in El Nino-southern oscillation. Journal
of Climate, 4(12), 1137-1146.
Feldman, R., &Taqqu, M. (Eds.). (1998). A practical guide to heavy tails: statistical techniques
and applications. Springer Science & Business Media.
Frankhauser P. (1998). The fractal approach: A new tool for the spatial analysis of urban
agglomerations. Population: An English Selection, 10(1): 205-240.
176
Gourieroux, C., &Jasiaky, J. (1998). Truncated maximum likelihood, goodness of fit tests and tail
analysis (No. 1998, 36). Discussion Papers, Interdisciplinary Research Project 373:
Quantification and Simulation of Economic Processes.
Guhathakurta, M., & Phillips, T. (2013). The Solar Cycle Turned Sideways. Space Weather, 11(5),
212-213.
Hale G. E., Ellerman F., Nicholson S. B. and Joy A. H. (1919). The Magnetic Polarity of Sunspots.
The Astrophysical Journal, 49, 153-178.
Hanslmeier, A., Denkmayr, K., & Weiss, P. (1999).Long term prediction of solar activity using
the combined method. Solar Physics, 184(1), 213-218.
Hastings, H. M., & Sugihara, G. (1993). Fractals .A user's guide for the natural sciences. Oxford
Science Publications, Oxford, New York: Oxford University Press,| c1993, 1.
H E Hurst, R P Black and Y M Simaika (1965). Long-term Storage: An experiment Study.
Constable, London, UK.
H.E. Hurst (1951). Transactions of the American Society of Civil Engineers. 116-770
Herring D. (2009). what is El nino? NASA’s Earth Observatory. Accessed October 22.
He W P, Deng B S, Wu Q, Zhang W and Cheng H Y (2010). Acta Phys. Sin. 598264 (in Chinese)
Higuchi T (1988). Approach to an irregular time series on basis of the fractal theory. physica D 31
277-283.
Hiremath KM (2006). Influences of solar activity on the rainfall over India”, ILWS Workshop,
2006.
177
H. Takayasu (1990). Fractals in the Physical Sciences, Nonlinear Science: Theory and
Applications. Manchester University Press, Manchester, UK.
Hussain, M. A. (2006). Mathematical Aspects of The Impact of Urban Greenhouse Gas Emissions
on Global Warming (Doctoral dissertation, Federal Urdu University of Arts, Science and
Technology).
Hurst, H. E. (1951). Long term storage capacity of reservoirs. Trans. Am. Soc. Eng. 116: 770–799.
Huang Z., Yong F., Wen-Yuan W., Yi Li and Jiang Liu. (2014). Low-dimensional chaos and
fractal properties of long-term sunspot activity, Research in Astronomy and Astrophysics,
Volume 14, Number 1.
J. Feder (1988), Fractal. Plenum Press, New York, NY USA.
John H. Cochrane. (1997). Time Series of Microeconomics and Finances. Graduate school of
Business, University of Chicago, spring.
K. w. Hipal, A. I. Mcleod. (1994). Time Series Modeling of Water Resources and Environmental
System. Amsterdam Elsevier.
Kevin R. M, Jimmy J. L., Véronique D., Fraser W. and Alfred O. H. III (2014). Image Patch
Analysis And Clustering Of Sunspots: A Dimensionality Reduction Approach. arXiv:
1406.6390v1 [cs.CV].
Kim S. N. Shepard and S. Chib (1998). Stochastic Volatility: Likelihood Inference and
Comparison with ARCH models. Review of Economic Studies. 65: 293-361.
Kok K., Teo, Lipo W., and Zhiping L (2001). Wavelet Packet Multi-layer Perceptron for Chaotic
Time Series Prediction: Effects of Weight Initialization. V.N. Alexandrov et al. (Eds.): ICCS,
LNCS 2074, pp. 310–317
Kumar SP, Roshin RP, Narvekar J, Dinesh Kumar PK, Vivekanandan E, (2010). Signatures of
global warming and regional climate shift in the Arabian sea, in climate change and aquatic
178
ecosystems. Eds. By: Joseph, A.; Nandan, S.B.; Agustine, A”, Climate Change and Aquatic
Ecosystems- (2010), Cochin University of science and technology; India, pp. 55-62, 2010.
Lakshmi SS, Tiwari RK, Somvanshi VK, (2003). Prediction of Indian rainfall index (IRF) using
the ENSO variability and sunspot cycles-An artificial neural network approach, J. Ind.
Geophys, Union, Vol. 7, pp. 173-181.
Li, K. J., Liang, H. F., Yun, H. S., &Gu, X. M. (2002). Statistical behavior of sunspot groups on
the solar disk. Solar Physics, 205(2), 361-370.
Lo, A. (1991). Long-Term Memory in Stock Market Prices. Econometrica. 59: 1279–1313.
doi: 10.2307/2938368
Mark AD and Aronson PB (1984). Scale Dependent Fractal Dimension of Topographical Surface:
an empirical Investigation with Application in Geomorphology and Computer mapping. Math
Geol 16: 671-683
Malamud BD, Turcotte DL (1999) Self-affine Time Series I: measure of weak and strong
persistenc. J stat Plan Infulance 80: 173-196
Michael F. Barnsley (1988). Fractals everywhere. Academic Press, ISBN 978-0-12-079062-3, pp.
I-XII, 1-394.
Mielke PW (1973). Another family of distributions for describing and analyzing precipitation data.
Journal of Applied Meteorology, 12, 275–280.
Mielke PW, Johnson ES (1974). Some generalized beta distributions of the second kind having
desirable application features in hydrology and meteorology. Water Resources Research, 10,
pp: 223–226.
179
Muraközy J., Baranyi T., Ludmány A. (2013). Sunspot Group Development in High
Temporal Resolution. Solar Physics (accepted), doi: 10.1007/s11207-013-0416-3.
Muraközy J., Ludmány A. (2012). Phase lags of solar hemispheric cycle. Monthly Notices of the
Royal Astronomical Society 419 (4), 3624-3630 doi:10.1111/j.1365-2966.2011.20011.x, ADS:
2012MNRAS.419.3624M.
Muraközy J., Baranyi T., Ludmány A. (2012). Development and morphology of leading-following
parts of sunspot groups. Central European Astrophysical Bulletin 36
(1), 1-8 ADS: 2012CEAB...36....1M.
Paul S Addison (2005), Fractals and Chaos. IOP Publishing.
Peter Grassberger and Itamer Procaccia (1983).Measuring the Strangeness Attractors. Physic D:
nonlinear phenomena, 9(1-2), 189-208.
Peter Grassberger and Itamer Procaccia (1983). Characterizing Attractors. Physical review letters.
50 (5), 346-349.
Peter Grassberger and Itamer Procaccia (1983). Generalized Dimensions of Strange Attractors.
Physics Letters, A.97 (6), 227-230.
Peters, E. E. (1991). Chaos and order in the capital markets. John Wiley and Sons. ISBN 0-471-
53372-6
Pickands J. (1975) Statistical Inference using Extreme Order Statistics, the Annals of statistics 3:
119-131
Priestly MB (1981) Spectral Analysis and Time series. Academic Press London
Reale M. (1998). A Graphically Modelling Approach to Time Series. Lancaster University.
180
R. Gayathri and R. Samuel Selvaraj (2010). Predictability of solar activity using fractal analysis,
J. Ind. Geophys. Union Vol.14, No.2, pp.89-92.
Ruzmaikin, A. (1999). Can El Nińo Amplify the Solar Forcing of Climate?. Geophysical research
letters, 26(15), 2255-2258.
Salakhutdinova, I. I. (1998). A fractal structure of the time series of global indices of solar activity.
Solar Physics, 181(1), 221-235.
Schwabe H. (1844). Sonnen-Beobachtungen im Jahre 1843. Astronomische Nachrichten 21 (495),
254-256.
S. D. Liu and S. K. Liu (1992). An Introduction to Fractals and Fractal Dimension, China
Meteorological Press, Beijing, China.
Seleshi Y, Demaree GR, Delleur JW,“Sunspot numbers as a possible indicator of annual rainfall
at ADDIS ABABA, ETHIOPIA”, Int. J. Climatol, 1994, Vol. 14, pp. 911-923.
Siingh D, Singh, RP, Singh Ashok Kumar, Kulkarni M.N., Gautam AS, Singh Abhay
Kumar,“Solar activity, lightning and climate”, Surv. Geophys. Vol. 32, pp. 659-703, 2011.
Solar Active Regions. Encyclopedia of Astronomy and Astrophysics
Spinage CA,“The changing climate of Africa Part1: Introduction and Eastern Africa”, African
Ecology- Benchmarks and Historical perspectives, Spinger Geography, Chapter 2, 2012.
Stenning D, Kashyap V, Lee T. C. M., van Dyk D. A. and Young C. A. (2010). Morphological
Image Analysis and Its Application to Sunspot Classification. Springer.
181
Sumathi, R. and Samuel Selvaraj (2011). FRACTAL DIMENSIONAL ANALYSIS OF
SUNSPOT NUMBERS. International Journal of Current Research Vol. 3, Issue, 11, pp.055-
056.
Sugihara, G., and May, R. M. (1993).Application of fractals in ecology, Trends Ecol. Evol., 5, 79-
86
Sun, W., Rachev, S., Fabozzi, F., &Kalev, P. (2005). Long-range dependence and heavy tailedness
in modelling trade duration. Working Paper, University of Karlsruhe.
Svensmark H, Friis-christensen E, (1997). Variations of cosmic ray flux and global cloud
coverage-a missing link in solar-climate relationships, J. Atmos. Solar Terrestrial Phys., Vol.
59, pp.1225-1232.
T. Raicharoen, C. Lursinsap, P. Sanguanbhoki. (2003). Application of Critical Vector Machine to
Time Series Predication. Circuits and System, ISCAS’03 proceeding of the 2003 International
Symposium on Volume 5, 25-28, May 2003, Pages v-741 – v-744.
Taqqu MS, Samorodnitsky G (1992). Linear Model with Long-Range Dependence and finite and
infinite variance In: New Direction in Time Series analysis. Part II, IMA volume in
Mathematics and its Application 46, Springer, pp 325-340.
Tinsley, B., and G. Deen, (1991), Apparent tropospheric response to MeV-GeV particle flux
variations: A connection via electrofreezing of supercooled water in high-level clouds? J.
Geophys. Res.,96 (D12), 22283–22296, DOI: 10.1029/91JD02473.
Thomas, John H., & Nigel Oscar Weiss (2008). Sunspots and Starspots. Cambridge University
Press,Vol. 46.
Thomas J. H. and Weiss N. O. (2008). Sunspots and Star spots. Cambridge university press.
182
TJO News (2006). The Theodor Jacobsen Observatory. Newsletter
Turcotte, D.L. (1997). Fractals and Chaos in Geology and Geophysics, 2nd Edition. Cambridge
University Press, Cambridge, 398pp
Turcotte D L (1992). Fractal and Chaos in geology and Geophysics. Cambridge University Press.
Webster, P.J. and Yang, S.(1992). Monsoon and ENSO: Selectively interactive systems.
Q.J.R.M.S., 118, 877-926.
Walck, C. (2007). Handbook on statistical distributions for experimentalists.
Weiss, N. O. (2006). Sunspot structure and dynamics. Space Science Reviews, 124(1-4), 13-22.
Weiss, N. O. (2007). "Sunspot structure and dynamics." Solar Dynamics and Its Effects on the
Heliosphere and Earth.Springer New York.13-22.
Yanguang Chen (2010).Exploring the Fractal Parameters of Urban Growth and Form with Wave-
Spectrum Analysis. Hindawi Publishing Corporation Discrete Dynamics in Nature and
Society Volume, Article ID 974917, 20 pages doi:10.1155/2010/974917
Y. Chen (2009).Urban gravity model based on cross-correlation function and Fourier analyses of
spatiotemporal process. Chaos, Solitons & Fractals, vol. 41, no. 2, pp. 603–614.
Yousuf H. Shaikh, A. R. Khan and M. I. Iqbal (2008). Sunspots Data Analysis Using Time Series,
Fraclats, world. Scientific publishing company, vol.16, no 3, 259-265.
183
Table 1.1: ACF, PACF, and Ljung-Box Q-statistics test of sunspot cycles at second differences
Sunspot Cycle (Aug 1755 - Mar 1766)
Lag ACF PACF Q-Stat Prob
1 -0.412 -0.412 23.900 0.000
2 -0.313 -0.581 37.813 0.000
3 0.223 -0.399 44.942 0.000
4 0.047 -0.396 45.265 0.000
5 -0.010 -0.247 45.280 0.000
6 -0.019 -0.009 45.331 0.000
7 -0.180 -0.161 50.103 0.000
8 0.129 -0.212 52.567 0.000
9 0.084 -0.126 53.613 0.000
10 -0.199 -0.379 59.601 0.000
11 0.243 0.004 68.547 0.000
12 -0.199 -0.218 74.626 0.000
13 0.010 -0.101 74.641 0.000
14 0.139 -0.151 77.662 0.000
15 -0.030 -0.005 77.808 0.000
16 -0.130 0.068 79.474 0.000
17 0.042 0.053 79.761 0.000
18 -0.051 -0.094 80.184 0.000
19 0.207 0.167 87.112 0.000
20 -0.202 -0.114 93.768 0.000
21 -0.074 -0.095 94.678 0.000
22 0.275 -0.034 107.27 0.000
23 -0.126 0.056 109.93 0.000
24 -0.107 -0.001 111.87 0.000
25 0.093 -0.097 113.35 0.000
26 0.056 -0.017 113.89 0.000
27 -0.066 -0.010 114.64 0.000
28 0.028 0.049 114.78 0.000
29 -0.155 -0.129 119.05 0.000
30 0.249 -0.039 130.11 0.000
31 -0.057 0.110 130.69 0.000
32 -0.158 0.124 135.22 0.000
33 0.078 0.042 136.35 0.000
34 0.086 0.106 137.72 0.000
35 -0.082 0.053 138.99 0.000
36 0.007 -0.024 139.00 0.000
Sunspot Cycle (Aug 1755 - Mar 1766)
Lag ACF PACF Q-Stat Prob
1 -0.495 -0.495 28.166 0.000
2 -0.112 -0.473 29.629 0.000
3 0.117 -0.237 33.280 0.000
4 -0.078 -0.215 33.997 0.000
5 -0.054 -0.263 34.350 0.000
6 0.116 -0.154 35.961 0.000
7 0.003 -0.019 35.962 0.000
8 -0.143 -0.140 38.462 0.000
9 0.070 -0.188 39.069 0.000
10 0.024 -0.215 39.141 0.000
184
11 0.133 0.157 41.380 0.000
12 -0.252 -0.057 49.459 0.000
13 0.090 -0.112 50.499 0.000
14 0.078 -0.060 51.299 0.000
15 -0.048 0.083 51.606 0.000
16 0.010 0.061 51.619 0.000
17 0.070 0.112 52.282 0.000
18 -0.162 -0.065 55.845 0.000
19 0.076 0.003 56.636 0.000
20 0.133 0.117 59.091 0.000
21 -0.168 -0.002 63.040 0.000
22 0.024 0.100 63.118 0.000
23 0.027 -0.072 63.222 0.000
24 0.047 0.054 63.542 0.000
25 -0.068 -0.028 64.216 0.000
26 0.065 -0.040 64.837 0.000
27 -0.006 0.094 64.844 0.000
28 -0.154 -0.091 68.431 0.000
29 0.163 -0.008 72.508 0.000
30 0.020 -0.030 72.573 0.000
31 -0.061 0.070 73.153 0.000
32 -0.075 0.006 74.057 0.000
33 0.129 0.056 76.743 0.000
34 -0..064 0.005 77.413 0.000
35 -0.042 -0.074 77.710 0.000
36 0.124 0.036 80.301 0.000
Sunspot Cycle (Aug 1775 - Jun 1784)
Lag ACF PACF Q-Stat Prob
1 -0.610 -0.610 40.144 0.000
2 0.167 -0.326 43.198 0.000
3 -0.210 -0.487 48.057 0.000
4 0.172 -0.485 51.353 0.000
5 0.036 -0.359 51.497 0.000
6 0.020 -0.122 51.545 0.000
7 -0.170 -0.151 54.843 0.000
8 0.072 -0.174 55.447 0.000
9 0.055 -0.062 55.804 0.000
10 0.003 -0.021 55.805 0.000
11 -0.056 -0.059 56.177 0.000
12 -0.041 -0.059 56.376 0.000
13 0.130 -0.0159 58.456 0.000
14 -0.110 -0.102 59.951 0.000
15 0.114 0.027 61.568 0.000
16 -0.134 0.146 63.824 0.000
17 0.042 0.085 64.051 0.000
18 0.051 0.030 64.080 0.000
19 0.093 0.122 65.218 0.000
20 -0.182 -0.028 69.574 0.000
21 0.137 -0.018 72.083 0.000
22 -0.106 -0.036 73.614 0.000
23 0.107 -0.074 75.177 0.000
24 -0.052 0.126 75.555 0.000
185
25 0.042 0.009 75.806 0.000
26 -0.096 0.622 77.125 0.000
27 0.043 -0.139 77.391 0.000
28 0.112 0.089 79.207 0.000
29 -0.173 -0.018 83.625 0.000
30 0.136 0.048 86.413 0.000
31 -0.131 0.027 89.008 0.000
32 0.099 -0.041 90.516 0.000
33 -0.009 -0.023 90.528 0.000
34 0.017 0.027 90.579 0.000
35 -0.116 0.002 92.732 0.000
36 0.093 -0.102 94.126 0.000
Sunspot Cycle (Jun 1784 - Jun 1798)
Lag ACF PACF Q-Stat Prob
1 -0.591 -0.519 59.370 0.000
2 -0.037 0.593 59.604 0.000
3 0.249 0.325 70.232 0.000
4 -0.131 -0.196 73.196 0.000
5 -0.084 -0.303 74.434 0.000
6 0.169 -0.250 79.433 0.000
7 -0.606 -0.157 80.064 0.000
8 -0.094 -0.227 81.628 0.000
9 0.179 -0.056 87.339 0.000
10 -0.197 -0.264 94.304 0.000
11 0.154 -0.134 98.616 0.000
12 -0.045 -0.105 98.982 0.000
13 -0.069 -0.165 99.982 0.000
14 0.133 0.004 103.09 0.000
15 -0.082 0.093 104.33 0.000
16 -0.112 -0.142 106.66 0.000
17 0.229 -0.085 116.49 0.000
18 -0.101 -0.029 118.43 0.000
19 -0.095 -0.005 120.15 0.000
20 0.178 0.064 126.25 0.000
21 -0.123 0.046 129.19 0.000
22 -0.022 0.022 129.19 0.000
23 0.133 0.097 132.74 0.000
24 -0.098 0.140 134.64 0.000
25 -0.065 -0.036 135.47 0.000
26 0.168 0.102 141.11 0.000
27 -0.092 -0.023 142.82 0.000
28 -0.037 -0.036 143.10 0.000
29 0.118 0.058 145.93 0.000
30 -0.128 0.005 149.32 0.000
31 0.069 0.006 150.29 0.000
32 -0.002 -0.024 150.29 0.000
33 -0.029 0.032 150.47 0.000
34 -0.025 0.047 150.61 0.000
35 0.006 0.034 150.61 0.000
36 -0.063 -0.059 151.48 0.000
Sunspot Cycle (Jun 1798 - Sep 1810)
186
Lag ACF PACF Q-Stat Prob
1 -0.596 -0.596 53.017 0.000
2 0.133 -0.346 55.662 0.000
3 -0.096 -0.346 57.064 0.000
4 0.052 -0.322 57.469 0.000
5 0.017 0.259 57.514 0.000
6 0.019 0.167 57.567 0.000
7 -0.016 -0.092 57.607 0.000
8 -0.075 -0.204 58.496 0.000
9 0.097 -0.154 59.579 0.000
10 -0.020 -0.085 60.040 0.000
11 -0.057 -0.021 60.565 0.000
12 0.057 -0.214 61.081 0.000
13 0.008 -0.141 61.092 0.000
14 -0.024 0.131 61.188 0.000
15 0.013 -0.101 61.215 0.000
16 -0.014 -0.073 61.245 0.000
17 0.015 0.004 61.281 0.000
18 -0.028 -0.010 61.409 0.000
19 0.032 -0.000 61.581 0.000
20 -0.056 -0.073 62.123 0.000
21 0.059 -0.081 62.729 0.000
22 -0.000 -0.049 62.729 0.000
23 0.010 0.060 62.746 0.000
24 -0.115 -0.109 65.079 0.000
25 0.120 -0.131 67.639 0.000
26 0.025 0.039 67.754 0.000
27 -0.069 0.061 68.612 0.000
28 0.000 0.036 68.612 0.000
29 -0.039 -0.058 68.886 0.000
30 0.063 -0.081 69.623 0.000
31 0.029 -0.023 69.780 0.000
32 -0.063 -0.018 70.539 0.000
33 0.026 -0.023 70.672 0.000
34 -0.041 -0.051 70.999 0.000
35 0.070 -0.057 71.941 0.000
36 -0.020 0.013 72.024 0.000
Sunspot Cycle (Sep1810 - Dec 1823)
Lag ACF PACF Q-Stat Prob
1 -0.159 -0.159 55.885 0.000
2 0.070 -0.425 56.668 0.000
3 -0.043 -0.440 56.972 0.000
4 0.120 0.298 59.324 0.000
5 -0.137 -0.425 62.448 0.000
6 0.262 -0.001 73.830 0.000
7 -0.336 -0.152 92.771 0.000
8 0.215 -0.043 100.77 0.000
9 -0.070 -0.002 101.42 0.000
10 0.033 -0.016 101.58 0.000
11 -0.055 0.044 107.43 0.000
12 0.081 -0.034 103.23 0.000
13 -0.072 0.083 104.12 0.000
187
14 -0.024 -0.178 104.23 0.000
15 0.134 0.044 107.43 0.000
16 -0.109 0.022 109.55 0.000
17 0.036 0.125 109.78 0.000
18 -0.069 -0.027 110.64 0.000
19 0.109 0.025 112.81 0.000
20 -0.060 0.064 113.47 0.000
21 0.045 0.002 113.85 0.000
22 -0.133 -0.082 117.21 0.000
23 0.178 -0.046 123.03 0.000
24 -0.088 0.076 124.50 0.000
25 0.005 -0.021 124.51 0.000
26 -0.050 -0.043 124.99 0.000
27 0.099 -0.054 126.88 0.000
28 -0.081 -0.092 128.17 0.000
29 0.083 0.051 129.53 0.000
30 -0.034 0.142 129.76 0.000
31 -0.082 0.039 131.09 0.000
32 0.092 0.046 132.79 0.000
33 -0.052 0.000 132.92 0.000
34 -0.013 -0.071 132.96 0.000
35 0.072 0.048 134.03 0.000
36 -0.080 0.048 135.36 0.000
Sunspot Cycle (Dec1823 - Oct 1833)
Lag ACF PACF Q-Stat Prob
1 -0.558 -0.558 37.357 0.000
2 0.018 -0.426 37.396 0.000
3 0.026 -0.350 37.479 0.000
4 -0.035 -0.390 37.633 0.000
5 0.170 -0.113 41.211 0.000
6 -0.211 -0.217 46.819 0.000
7 0.090 -0.215 47.847 0.000
8 0.050 -0.099 48.171 0.000
9 -0.101 -0.169 49.481 0.000
10 0.041 -0.259 49.699 0.000
11 -0.016 -0.334 49.734 0.000
12 0.159 -0.065 53.071 0.000
13 -0.224 -0.181 59.759 0.000
14 0.100 -0.106 61.103 0.000
15 0.028 0.018 61.211 0.000
16 -0.113 0.159 62.957 0.000
17 0.132 -0.167 65.693 0.000
18 -0.046 0.016 65.693 0.000
19 -0.081 0.209 66.613 0.000
20 0.180 0.028 71.262 0.000
21 -0.201 0.053 77.132 0.000
22 0.103 -0.069 78.694 0.000
23 -0.002 -0.114 78.694 0.000
24 0.002 -0.035 78.695 0.000
25 -0.002 -0.098 78.695 0.000
26 -0.015 -0.003 78.730 0.000
27 -0.023 0.014 78.811 0.000
188
28 0.083 0.157 79.879 0.000
29 -0.110 -0.005 81.806 0.000
30 0.074 0.051 82.692 0.000
31 -0.033 0.004 82.866 0.000
32 0.049 -0.042 82.256 0.000
33 -0.084 0.022 84.426 0.000
34 0.060 -0.005 85.035 0.000
35 0.028 -0.046 85.169 0.000
36 -0.083 -0.107 86.353 0.000
Sunspot Cycle (Oct1833 - Sep 1843)
Lag ACF PACF Q-Stat Prob
1 -0.645 -0.645 50.398 0.000
2 0.285 -0.226 60.284 0.000
3 -0.220 -0.255 66.261 0.000
4 0.079 -0.291 67.037 0.000
5 -0.112 0.433 68.604 0.000
6 0.188 0.259 74.491 0.000
7 -0.028 -0.061 73.169 0.000
8 -0.101 -0.259 74.491 0.000
9 0.077 -0.308 75.268 0.000
10 0.016 -0.044 75.301 0.000
11 -0.095 -0.065 76.485 0.000
12 0.083 -0.120 77.398 0.000
13 0.017 0.072 77.438 0.000
14 -0.199 -0.249 82.831 0.000
15 0.284 -0.086 93.900 0.000
16 -0.165 0.029 97.961 0.000
17 0.104 0.020 99.204 0.000
18 -0.117 0.022 101.51 0.000
19 0.038 -0.166 101.35 0.000
20 -0.029 0.089 101.48 0.000
21 0.130 0.227 103.93 0.000
22 -0.103 0.072 105.49 0.000
23 0.034 0.086 105.60 0.000
24 0.107 -0.011 107.37 0.000
25 0.087 -0.179 108.52 0.000
26 0.052 0.149 108.95 0.000
27 -0.058 -0.035 109.47 0.000
28 0.105 0.099 111.20 0.000
29 -0.232 0.069 119.47 0.000
30 0.158 -0.142 123.76 0.000
31 -0.109 -0.108 125.69 0.000
32 0.177 0.109 130.83 0.000
33 -0.065 -0.032 131.53 0.000
34 -0.056 -0.047 132.05 0.000
35 0.065 0.033 032.78 0.000
36 -0.131 -0.115 135.76 0.000
Sunspot Cycle (Sep1843 - Mar 1855)
Lag ACF PACF Q-Stat Prob
1 -0.595 -0.595 49.611 0.000
2 0.039 0.448 49.827 0.000
189
3 0.117 0.282 51.779 0.000
4 -0.101 -0.265 53.241 0.000
5 0.041 -0.231 53.482 0.000
6 0.040 -0.117 53.718 0.000
7 -0.062 -0.088 54.276 0.000
8 -0.032 -0.215 54.426 0.000
9 0.137 0.069 57.209 0.000
10 -0.168 -0.190 61.433 0.000
11 0.104 -0.187 63.079 0.000
12 0.001 -0.165 63.080 0.000
13 -0.048 -0.153 63.429 0.000
14 0.072 -0.044 64.221 0.000
15 -0.122 -0.210 66.536 0.000
16 0.112 -0.194 68.154 0.000
17 0.030 -0.002 58.658 0.000
18 -0.117 0.000 70.832 0.000
19 0.029 -0.056 70.965 0.000
20 0.068 0.066 72.161 0.000
21 -0.138 -0.017 75.266 0.000
22 0.087 -0.065 76.527 0.000
23 0.107 0.255 78.433 0.000
24 -0.337 -0.079 97.563 0.000
25 0.335 0.050 116.61 0.000
26 -0.097 0.026 118.24 0.000
27 -0.085 0.023 119.48 0.000
28 0.098 0.034 121.15 0.000
29 -0.023 0.023 121.24 0.000
30 -0.010 0.095 121.26 0.000
31 -0.060 -0.139 121.91 0.000
32 0.140 -0.043 125.47 0.000
33 -0.134 -0.018 128.76 0.000
34 0.110 0.015 130.99 0.000
35 -0.132 -0.122 134.22 0.000
36 0.108 -0.040 136.41 0.000
Sunspot Cycle (Mar1855 - Feb 1867)
Lag ACF PACF Q-Stat Prob
1 -0.625 -0.625 56.633 0.000
2 0.127 0.432 59.004 0.000
3 -0.02 -0.359 59.077 0.000
4 -0.012 -0.379 59.097 0.000
5 0.089 -0.239 60.285 0.000
6 -0.070 -0.167 61.022 0.000
7 0.029 -0.167 61.110 0.000
8 -0.063 -0.172 61.722 0.000
9 0.047 -0.026 62.068 0.000
10 0.119 0.023 64.270 0.000
11 -0.0223 -0.075 72.043 0.000
12 0.116 -0.164 74.175 0.000
13 0.055 0.036 74.650 0.000
14 -0.128 -0.048 77.268 0.000
15 0.179 0.182 82.411 0.000
16 -0.266 -0.059 93.854 0.000
190
17 0.263 0.078 105.19 0.000
18 -0.156 0.067 109.20 0.000
19 0.049 -0.035 109.60 0.000
20 0.013 -0.043 109.63 0.000
21 0.000 0.175 109.63 0.000
22 -0.089 -0.100 110.97 0.000
23 0.127 -0.086 113.72 0.000
24 -0.043 0.029 114.03 0.000
25 -0.059 -0.133 114.65 0.000
26 0.102 0.059 116.49 0.000
27 -0.051 0.094 116.95 0.000
28 -0.025 0.123 117.06 0.000
29 -0.022 0.021 117.15 0.000
30 0.148 0.114 121.15 0.000
31 -0.190 0.093 127.82 0.000
32 0.085 -0.062 129.15 0.000
33 0.058 0.006 129.78 0.000
34 -0.120 -0.109 132.50 0.000
35 0.083 0.130 133.83 0.000
36 0.030 -0.076 134.01 0.000
Sunspot Cycle (Feb1867 - Sep 1878)
Lag ACF PACF Q-Stat Prob
1 -0.464 -0.464 30.353 0.000
2 -0.195 -0.522 35.743 0.000
3 0.226 0.270 43.066 0.000
4 -0.084 -0.0296 44.086 0.000
5 -0.051 -0.328 44.463 0.000
6 0.133 -0.218 47.063 0.000
7 -0.021 -0.098 47.127 0.000
8 -0.142 -0.210 50.105 0.000
9 -0.142 -0.210 56.574 0.000
10 -0.213 -0.207 63.429 0.000
11 0.051 0.234 63.823 0.000
12 0.253 0.078 73.634 0.000
13 -0.338 -0.163 91.255 0.000
14 0.143 0.016 94.430 0.000
15 0.057 -0.008 99.442 0.000
16 -0.144 -0.093 98.204 0.000
17 0.088 -0.008 99.442 0.000
18 0.073 -0.018 100.30 0.000
19 -0.147 -0.055 103.83 0.000
20 0.062 0.017 104.64 0.000
21 0.065 -0.039 105.16 0.000
22 -0.215 -0.206 112.88 0.000
23 0.257 -0.093 123.94 0.000
24 0.003 0.011 123.94 0.000
25 -0.239 0.016 133.69 0.000
26 0.133 -0.080 136.72 0.000
27 0.082 0.062 137.88 0.000
28 -0.174 -0.083 143.20 0.000
29 0.095 -0.117 144.81 0.000
30 0.115 0.051 147.19 0.000
191
31 -0.200 0.025 154.41 0.000
32 0.066 -0.031 155.20 0.000
33 0.089 0.152 156.67 0.000
34 -0.168 0.008 161.90 0.000
35 0.126 -0.074 164.87 0.000
36 0.036 0.050 165.11 0.000
Sunspot Cycle (Sep1878 - Jun 1890)
Lag ACF PACF Q-Stat Prob
1 -0.592 -0.592 50.136 0.000
2 0.047 -0.467 50.126 0.000
3 0.107 -0.250 52.126 0.000
4 -0.154 -0.348 55.610 0.000
5 0.168 -0.183 59.763 0.000
6 -0.145 -0.282 62.866 0.000
7 0.162 -0.041 66.801 0.000
8 -0.221 -0.309 74.147 0.000
9 0.224 -0.096 81.747 0.000
10 -0.121 -0.170 83.975 0.000
11 0.043 0.029 -84.257 0.000
12 -0.037 -0.130 84.473 0.000
13 -0.035 -0.140 -84.664 0.000
14 0.112 -0.207 86.640 0.000
15 -0.056 -0.017 87.142 0.000
16 0.012 -0.078 87.164 0.000
17 -0.036 0.031 87.370 0.000
18 0.014 -0.095 87.403 0.000
19 -0.023 -0.126 87.490 0.000
20 0.116 -0.017 89.727 0.000
21 -0.115 0.060 91.926 0.000
22 0.027 0.101 92.050 0.000
23 -0.048 -0.078 92.441 0.000
24 0.106 0.018 94.363 0.000
25 -0.051 -0.019 94.933 0.000
26 -0.044 -0.042 95.264 0.000
27 0.109 0.008 97.370 0.000
28 -0.126 0.010 100.20 0.000
29 0.121 -0.001 102.81 0.000
30 -0.086 -0.016 104.16 0.000
31 0.034 -0.034 104.36 0.000
32 -0.008 0.053 104.37 0.000
33 -0.024 0.023 104.38 0.000
34 0.084 0.038 105.82 0.000
35 -0.075 0.054 106.88 0.000
36 0.020 0.077 106.96 0.000
Sunspot Cycle (Jun1890 - Sep 1902)
Lag ACF PACF Q-Stat Prob
1 -0.680 -0.680 68.957 0.000
2 0.231 -0.432 76.952 0.000
3 -0.128 -0.450 79.428 0.000
4 0.179 -0.217 84.319 0.000
5 -0.212 -0.318 91.207 0.000
192
6 0.214 -0.153 98.278 0.000
7 -0.171 -0.137 102.83 0.000
8 0.098 -0.138 104.34 0.000
9 -0.056 -0.121 104.84 0.000
10 0.051 0.127 104.84 0.000
11 -0.073 -0.192 106.10 0.000
12 0.084 -0.212 107.24 0.000
13 -0.027 0.100 107.36 0.000
14 -0.029 -0.091 107.50 0.000
15 0.034 -0.011 107.68 0.000
16 -0.001 0.092 107.69 0.000
17 -0.078 -0.062 108.71 0.000
18 0.128 -0.014 111.47 0.000
19 -0.061 0.091 112.10 0.000
20 -0.045 0.040 112.46 0.000
21 0.066 0.077 113.22 0.000
22 -0.057 -0.065 113.79 0.000
23 0.060 -0.054 114.42 0.000
24 0.006 0.102 114.43 0.000
25 -0.109 -0.081 116.55 0.000
26 0.140 0.019 120.10 0.000
27 -0.111 -0.083 122.32 0.000
28 0.180 0.046 124.44 0.000
29 -0.117 0.076 126.95 0.000
30 0.079 0.013 128.12 0.000
31 -0.032 0.062 128.32 0.000
32 0.016 0.015 128.36 0.000
33 -0.015 0.077 128.41 0.000
34 0.015 0.059 128.45 0.000
35 -0.029 0.041 128.61 0.000
36 0.038 0.006 128.89 0.000
Sunspot Cycle (Sep 1902 - Dec 1913)
Lag ACF PACF Q-Stat Prob
1 -0.622 -0.622 52.968 0.000
2 0.240 -0.239 60.926 0.000
3 -0.255 0.377 69.945 0.000
4 0.175 -0.306 74.219 0.000
5 -0.002 -0.098 74.220 0.000
6 -0.145 -0.559 77.220 0.000
7 0.145 -0.319 80.249 0.000
8 -0.069 -0.305 80.940 0.000
9 0.115 -0.279 82.887 0.000
10 -0.098 -0.219 84.281 0.000
11 0.050 -0.097 84.645 0.000
12 -0.035 0.030 84.827 0.000
13 -0.083 -0.108 85.870 0.000
14 0.096 -0.133 87.266 0.000
15 -0.002 -0.042 87.266 0.000
16 0.041 0.028 87.526 0.000
17 -0.107 -0.052 89.309 0.000
18 0.036 0.241 89.514 0.000
19 0.154 0.038 93.261 0.000
193
20 -0.238 -0.131 102.31 0.000
21 0.208 0.141 109.27 0.000
22 -0.204 0.204 118.66 0.000
23 0.197 0.089 125.05 0.000
24 -0.151 -0.108 128.85 0.000
25 0.182 -0.051 134.39 0.000
26 -0.056 0.048 134.92 0.000
27 -0.057 0.044 135.47 0.000
28 0.021 0.008 135.54 0.000
29 -0.023 -0.006 135.63 0.000
30 0.068 -0.038 136.44 0.000
31 -0.115 -0.018 138.79 0.000
32 0.081 0.095 139.69 0.000
33 -0.016 0.041 140.01 0.000
34 0.083 0.063 141.27 0.000
35 -0.097 -0.004 143.00 0.000
36 -0.002 -0.058 143.00 0.000
Sunspot Cycle (Dec1913 - May 1923)
Lag ACF PACF Q-Stat Prob
1 -0.459 -0.459 24.279 0.000
2 -0.207 -0.530 29.258 0.000
3 0.108 -0.496 30.624 0.000
4 0.264 -0.098 38.840 0.000
5 -0.249 -0.104 46.214 0.000
6 -0.103 0.246 47.500 0.000
7 0.223 -0.161 53.551 0.000
8 -0.053 -0.276 53.900 0.000
9 -0.045 -0.178 54.149 0.000
10 -0.002 -0.123 54.149 0.000
11 0.107 0.003 55.590 0.000
12 -0.129 -0.052 57.723 0.000
13 -0.001 -0.146 57.723 0.000
14 0.087 -0.142 58.699 0.000
15 -0.007 -0.082 58.705 0.000
16 -0.085 -0.067 59.660 0.000
17 0.032 -0.088 59.801 0.000
18 0.090 -0.064 60.899 0.000
19 -0.050 0.023 61.243 0.000
20 -0.092 -0.022 62.416 0.000
21 0.058 -0.080 62.894 0.000
22 0.072 -0.072 63.633 0.000
23 -0.075 -0.077 64.440 0.000
24 -0.020 -0.050 64.499 0.000
25 0.018 -0.175 64.544 0.000
26 0.082 -0.154 65.548 0.000
27 -0.004 0.091 65.550 0.000
28 -0.175 -0.063 70.207 0.000
29 0.107 -0.135 71.980 0.000
30 0.076 -0.153 72.886 0.000
31 0.018 0.152 72.938 0.000
32 -0.277 -0.045 85.659 0.000
33 0.286 0.115 98.385 0.000
194
34 -0.036 0.086 98.601 0.000
35 -0.131 -0.036 101.47 0.000
36 0.038 0.024 101.71 0.000
Sunspot Cycle (May 1923 - Sep 1933)
Lag ACF PACF Q-Stat Prob
1 -0.576 -0.576 41.859 0.000
2 0.137 -0.293 44.233 0.000
3 -0.138 -0.336 46.673 0.000
4 0.034 -0.372 46.822 0.000
5 0.055 -0.302 47.215 0.000
6 0.077 -0.085 47.985 0.000
7 -0.112 -0.091 49.650 0.000
8 0.028 -0.066 49.755 0.000
9 -0.051 -0.096 50.104 0.000
10 0.013 -0.202 50.129 0.000
11 0.105 -0.096 51.654 0.000
12 -0.042 -0.003 51.900 0.000
13 -0.067 -0.069 52.533 0.000
14 -0.036 -0.221 52.713 0.000
15 0.070 -0.237 53.411 0.000
16 0.030 -0.241 53.538 0.000
17 0.092 -0.017 54.762 0.000
18 -0.200 -0.082 60.622 0.000
19 0.073 -0.081 61.406 0.000
20 0.008 0.027 61.415 0.000
21 0.041 0.118 61.673 0.000
22 -0.069 -0.038 62.400 0.000
23 0.119 0.136 64.586 0.000
24 -0.233 -0.059 73.107 0.000
25 0.247 0.110 82.577 0.000
26 -0.193 -0.012 88.478 0.000
27 0.144 -0.055 91.806 0.000
28 -0.018 0.054 91.858 0.000
29 -0.102 -0.013 93.556 0.000
30 0.028 -0.065 93.868 0.000
31 0.069 -0.039 94.475 0.000
32 0.006 0.040 94.462 0.000
33 -0.053 -0.056 94.955 0.000
34 0.016 -0.058 95.001 0.000
35 -0.016 0.032 95.046 0.000
36 0.038 0.019 95.302 0.000
Sunspot Cycle (Sep 1933 - Jan 1944)
Lag ACF PACF Q-Stat Prob
1 -0.592 -0.592 44.118 0.000
2 0.094 -0.394 45.281 0.000
3 0.005 -0.274 45.284 0.000
4 -0.035 -0.266 45.443 0.000
5 -0.013 -0.324 45.467 0.000
6 0.115 -0.157 47.190 0.000
7 -0.000 0.103 47.190 0.000
8 -0.271 -0.307 57.014 0.000
195
9 0.284 -0.260 67.921 0.000
10 -0.053 -0.063 68.306 0.000
11 -0.062 -0.074 68.835 0.000
12 0.042 -0.155 69.078 0.000
13 -0.012 -0.144 69.098 0.000
14 -0.025 -0.032 69.188 0.000
15 0.017 -0.099 69.229 0.000
16 0.116 -0.006 71.153 0.000
17 -0.208 -0.026 77.402 0.000
18 0.154 0.126 80.867 0.000
19 -0.121 -0.091 83.047 0.000
20 0.136 0.037 85.811 0.000
21 -0.073 0.149 86.624 0.000
22 -0.076 -0.161 87.495 0.000
23 0.213 0.142 94.453 0.000
24 -0.307 -0.123 109.12 0.000
25 0.345 0.116 127.75 0.000
26 -0.249 -0.011 137.60 0.000
27 0.101 -0.055 139.23 0.000
28 -0.043 0.070 139.53 0.000
29 0.068 0.065 140.28 0.000
30 -0.081 -0.067 141.36 0.000
31 -0.035 -0.213 141.56 0.000
32 0.200 0.008 148.34 0.000
33 -0.193 0.117 154.71 0.000
34 0.087 -0.007 156.01 0.000
35 -0.013 0.020 156.04 0.000
36 -0.044 0.062 156.38 0.000
Sunspot Cycle (Jan 1944 - Feb 1954)
Lag ACF PACF Q-Stat Prob
1 -0.547 -0.547 36.806 0.000
2 0.067 -0.332 37.354 0.000
3 -0.129 -0.420 39.432 0.000
4 0.206 -0.181 44.789 0.000
5 -0.141 -0.211 47.333 0.000
6 0.103 -0.066 48.685 0.000
7 -0.153 -0.212 51.702 0.000
8 0.115 -0.218 53.420 0.000
9 -0.041 -0.277 53.645 0.000
10 0.068 -0.181 54.259 0.000
11 -0.021 -0.030 54.321 0.000
12 -0.072 -0.127 55.031 0.000
13 0.044 -0.112 55.293 0.000
14 0.020 -0.119 55.349 0.000
15 -0.010 -0.119 55.363 0.000
16 -0.017 -0.094 55.402 0.000
17 0.016 -0.060 55.440 0.000
18 -0.054 -0.156 55.682 0.000
19 0.149 0.031 59.061 0.000
20 -0.129 0.034 61.500 0.000
21 -0.011 -0.061 61.518 0.000
22 0.031 -0.039 61.660 0.000
196
23 0.061 0.019 62.231 0.000
24 -0.061 0.057 62.527 0.000
25 -0.069 -0.140 63.527 0.000
26 0.114 -0.058 65.563 0.000
27 -0.005 -0.029 65.567 0.000
28 0.016 0.106 65.609 0.000
29 -0.103 0.076 67.302 0.000
30 0.014 0.114 69.115 0.000
31 0.104 0.114 69.115 0.000
32 -0.112 -0.068 71.194 0.000
33 0.086 0.047 72.452 0.000
34 -0.071 0.029 73.309 0.000
35 -0.008 -0.110 73.319 0.000
36 0.091 -0.002 74.762 0.000
Sunspot Cycle (Feb 1954 - Oct 1964)
Lag ACF PACF Q-Stat Prob
1 -0.981 -0.981 30.139 0.000
2 -0.258 -0.637 38.845 0.000
3 0.398 0.245 59.778 0.000
4 -0.203 -0.371 65.275 0.000
5 0.046 -0.118 65.557 0.000
6 0.005 -0.232 65.560 0.000
7 0.003 -0.037 65.561 0.000
8 -0.030 -0.187 65.686 0.000
9 -0.008 -0.163 65.695 0.000
10 0.122 0.043 67.767 0.000
11 -0.212 -0.264 74.140 0.000
12 0.133 -0.210 76.668 0.000
13 0.142 0.008 79.583 0.000
14 -0.302 -0.062 92.839 0.000
15 0.138 -0.089 95.264 0.000
16 0.105 -0.057 97.264 0.000
17 -0.099 0.140 98.722 0.000
18 -0.063 0.015 99.310 0.000
19 0.025 -0.142 99.406 0.000
20 0.154 0.000 103.06 0.000
21 -0.199 -0.119 109.16 0.000
22 0.092 -0.023 110.84 0.000
23 0.071 0.028 111.28 0.000
24 -0.195 0.006 117.35 0.000
25 0.169 -0.066 121.94 0.000
26 0.023 0.088 122.03 0.000
27 -0.191 -0.010 128.02 0.000
28 0.145 -0.012 131.51 0.000
29 0.030 -0.029 131.66 0.000
30 -0.082 0.130 132.81 0.000
31 -0.056 -0.010 133.35 0.000
32 0.113 0.094 135.55 0.000
33 -0.021 0.267 135.82 0.000
34 -0.022 -0.022 135.71 0.000
35 0.061 0.007 136.37 0.000
36 -0.109 -0.009 138.53 0.000
197
Sunspot Cycle (Oct 1964 - May 1976)
Lag ACF PACF Q-Stat Prob
1 -0.601 -0.601 59.950 0.000
2 0.066 -0.463 51.562 0.000
3 0.079 -0.282 52.453 0.000
4 -0.125 -0.364 54.700 0.000
5 0.180 0.146 59.399 0.000
6 -0.192 -0.270 64.793 0.000
7 0.183 -0.069 69.719 0.000
8 -0.142 -0.140 72.719 0.000
9 0.019 -0.201 72.774 0.000
10 0.108 -0.116 74.545 0.000
11 -0.114 -0.071 76.514 0.000
12 0.044 -0.115 76.814 0.000
13 -0.018 -0.115 76.887 0.000
14 0.052 -0.026 77.282 0.000
15 -0.082 -0.106 78.330 0.000
16 0.073 -0.032 79.166 0.000
17 -0.017 -0.007 79.213 0.000
18 -0.073 -0.106 80.069 0.000
19 0.131 -0.005 82.867 0.000
20 -0.131 -0.100 85.696 0.000
21 0.112 -0.031 87.755 0.000
22 -0.078 -0.047 88.777 0.000
23 0.046 0.024 89.138 0.000
24 -0.029 -0.049 89.277 0.000
25 -0.077 -0.234 90.299 0.000
26 0.247 -0.011 100.79 0.000
27 -0.267 -0.087 113.04 0.000
28 0.116 -0.130 115.38 0.000
29 0.067 0.060 116.18 0.000
30 -0.169 -0.026 121.30 0.000
31 0.170 0.050 126.51 0.000
32 -0.118 0.095 129.05 0.000
33 0.078 0.128 130.18 0.000
34 -0.111 -0.014 132.48 0.000
35 0.091 -0.020 134.02 0.000
36 0.056 0.034 134.62 0.000
Sunspot Cycle (May1976 - Mar 1986)
Lag ACF PACF Q-Stat Prob
1 -0.546 -0.546 35.741 0.000
2 0.048 -0.355 36.024 0.000
3 -0.108 0.433 37.441 0.000
4 0.148 0.305 40.141 0.000
5 -0.002 0.171 40.142 0.000
6 -0.085 0.230 41.042 0.000
7 0.053 -0.168 41.392 0.000
8 -0.012 -0.145 41.410 0.000
9 -0.045 -0.287 41.666 0.000
10 0.099 -0.192 42.954 0.000
11 -0.092 -0.283 44.066 0.000
198
12 0.156 -0.029 47.309 0.000
13 -0.159 0.084 50.711 0.000
14 -0.045 -0.109 50.989 0.000
15 0.115 -0.091 52.789 0.000
16 -0.031 -0.205 52.920 0.000
17 0.135 0.075 55.457 0.000
18 -0.253 0.034 64.487 0.000
19 0.127 0.020 66.789 0.000
20 0.006 0.034 66.794 0.000
21 0.008 -0.019 66.804 0.000
22 -0.012 -0.044 66.825 0.000
23 -0.026 -0.011 66.925 0.000
24 0.014 -0.050 66.954 0.000
25 0.013 0.005 66.978 0.000
26 -0.050 -0.086 67.360 0.000
27 0.162 0.073 71.427 0.000
28 -0.213 -0.046 78.519 0.000
29 0.124 -0.059 80.936 0.000
30 -0.105 -0.108 82.703 0.000
31 0.148 -0.026 86.254 0.000
32 -0.037 0.079 86.474 0.000
33 -0.113 -0.046 88.596 0.000
34 0.112 -0.003 90.710 0.000
35 -0.038 0.083 90.953 0.000
36 -0.007 -0.001 90.962 0.000
Sunspot Cycle (Mar1986 - Jun1996)
Lag ACF PACF Q-Stat Prob
1 -0.639 -0.639 51.124 0.000
2 0.139 -0.456 53.567 0.000
3 0.001 0.343 53.567 0.000
4 0.018 0.222 53.609 0.000
5 -0.115 -0.390 55.314 0.000
6 0.193 0.250 60.166 0.000
7 -0.132 -0.211 62.472 0.000
8 0.070 -0.082 63.128 0.000
9 -0.047 -0.011 63.420 0.000
10 -0.058 -0.215 63.863 0.000
11 0.099 -0.254 65.199 0.000
12 0.088 0.120 66.270 0.000
13 -0.285 -0.096 77.526 0.000
14 0.252 -0.142 86.455 0.000
15 -0.054 0.036 86.872 0.000
16 -0.110 0.007 88.610 0.000
17 0.117 0.028 90.597 0.000
18 0.000 0.149 90.597 0.000
19 -0.145 -0.173 93.699 0.000
20 0.208 -0.052 100.13 0.000
21 -0.139 0.010 103.03 0.000
22 0.071 0.126 103.80 0.000
23 -0.085 -0.108 104.91 0.000
24 0.087 -0.072 106.07 0.000
25 -0.090 -0.035 107.32 0.000
199
26 0.160 0.032 111.38 0.000
27 -0.215 -0.145 118.72 0.000
28 0.213 0.035 125.99 0.000
29 -0.187 -0.112 131.66 0.000
30 0.116 -0.009 133.86 0.000
31 -0.056 -0.010 134.39 0.000
32 0.120 0.040 136.79 0.000
33 -0.190 0.046 142.90 0.000
34 0.128 0.043 145.73 0.000
35 -0.057 0.002 146.29 0.000
36 0.044 -0.022 146.62 0.000
Sunspot Cycle (Jun1996 - Jan 2008)
Lag ACF PACF Q-Stat Prob
1 -0.142 -0.142 23.900 0.000
2 -0.313 -0.581 37.813 0.000
3 0.223 0.399 44.942 0.000
4 0.047 -0.396 45.265 0.000
5 -0.010 -0.247 45.280 0.000
6 0.019 0.009 45.331 0.000
7 -0.180 -0.161 50.103 0.000
8 0.129 -0.121 52.567 0.000
9 0.084 -0.126 53.613 0.000
10 -0.199 -0.379 59.601 0.000
11 0.243 0.004 68.547 0.000
12 -0.199 -0.218 74.626 0.000
13 0.010 -0.101 74.641 0.000
14 0.139 -0.151 77.662 0.000
15 -0.030 -0.005 77.808 0.000
16 -0.103 0.068 79.474 0.000
17 0.042 0.053 79.761 0.000
18 -0.051 -0.090 80.184 0.000
19 0.207 0.167 87.112 0.000
20 -0.202 -0.114 93.768 0.000
21 -0.074 -0.095 94.678 0.000
22 0.275 -0.034 107.27 0.000
23 -0.126 0.056 109.93 0.000
24 -0.107 -0.001 111.87 0.000
25 0.093 -0.097 113.35 0.000
26 0.056 -0.017 113.89 0.000
27 -0.066 -0.010 114.64 0.000
28 0.028 0.049 114.70 0.000
29 -0.155 -0.129 119.05 0.000
30 0.249 -0.039 130.11 0.000
31 -0.057 0.110 130.69 0.000
32 -0.158 0.124 135.22 0.000
33 0.078 0.042 136.65 0.000
34 0.086 0.106 137.72 0.000
35 -0.082 0.050 138.99 0.000
36 0.007 -0.024 139.00 0.000
Sunspot Cycle (Aug1755 - Jan 2008)
Lag ACF PACF Q-Stat Prob
200
1 -0.570 -0.570 986.15 0.000
2 0.033 0.433 989.03 0.000
3 0.029 -0.349 992.03 0.000
4 0.010 -0.284 992.26 0.000
5 -0.012 -0.260 992.80 0.000
6 0.030 -0.188 995.54 0.000
7 -0.017 -0.136 996.41 0.000
8 -0.042 -0.196 100.17 0.000
9 0.062 -0.162 1013.6 0.000
10 -0.028 -0.161 1016.0 0.000
11 0.007 -0.144 1016.1 0.000
12 0.013 -0.088 1016.6 0.000
13 -0.037 -0.106 1020.7 0.000
14 0.014 -0.130 1021.2 0.000
15 0.034 -0.082 1024.8 0.000
16 -0.046 -0.111 1031.3 0.000
17 0.049 -0.030 1038.6 0.000
18 -0.054 -0.052 1047.3 0.000
19 0.025 -0.054 1049.2 0.000
20 0.023 0.006 1050.8 0.000
21 -0.035 0.003 1054.4 0.000
22 -0.010 -0.047 1054.7 0.000
23 0.064 0.040 1067.2 0.000
24 -0.084 -0.031 1088.9 0.000
25 0.045 -0.055 1095.1 0.000
26 0.022 -0.026 1096.6 0.000
27 -0.028 -0.007 1099.1 0.000
28 0.003 0.024 1099.1 0.000
29 -0.013 -0.014 1099.7 0.000
30 0.031 0.019 1102.5 0.000
31 -0.037 -0.022 1106.8 0.000
32 0.037 -0.025 1111.0 0.000
33 -0.020 -0.013 1112.2 0.000
34 0.008 0.004 1112.4 0.000
35 -0.006 0.009 1112.5 0.000
36 0.001 0.015 1112.5 0.000
Table 1.2: unit root test of sunspot Cycles
Null Hypothesis H0: Sunspots Cycle (Aug 1755 - Mar 1766) has a unit root
Exogenous: Constant
Lag Length: 12 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -6.181642 0.0000
Test Critical vales 1% level
5% level
10% level
-3.489117
-2.887190
-2.50525
201
Null Hypothesis H0: Sunspots Cycle (Mar 1766 - Aug 1775) has a unit root
Exogenous: Constant
Lag Length: 12 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -6.091491 0.0000
Test Critical vales 1% level
5% level
10% level
-3.497727
-2.890926
-2.582514
Null Hypothesis H0: Sunspots Cycle (Aug 1775 - Jun 1784) has a unit root
Exogenous: Constant
Lag Length: 12 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -6.893381 0.0000
Test Critical vales 1% level
5% level
10% level
-3.499167
-2.891550
-2.582846
Null Hypothesis H0: Sunspots Cycle (Jun 1784 - Jun 1798) has a unit root
Exogenous: Constant
Lag Length: 12 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -7.579517 0.0000
Test Critical vales 1% level
5% level
10% level
-3.473096
-2.880211
-2.576805
Null Hypothesis H0: Sunspots Cycle (Jun 1798 - Sep 1810) has a unit root
Exogenous: Constant
Lag Length: 13 (Automatics-based on AIC, MaxLag = 13)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -6.823644 0.0000
Test Critical vales 1% level
5% level
10% level
-3.480425
-2.883408
-2.578510
Null Hypothesis H0: Sunspots Cycle (Sep1810 - Dec 1823) has a unit root
Exogenous: Constant
Lag Length: 13 (Automatics-based on AIC, MaxLag = 13)
202
t-statistics prob*
Augmented Dickey Fuller (ADF) test -6.875331 0.0000
Test Critical vales 1% level
5% level
10% level
-3.476143
-2.881541
-2.577514
Null Hypothesis H0: Sunspots Cycle (Dec1823 - Oct 1833) has a unit root
Exogenous: Constant
Lag Length: 12 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -6.695780 0.0000
Test Critical vales 1% level
5% level
10% level
-3.494378
-2.889474
-2.581741
Null Hypothesis H0: Sunspots Cycle (Oct1833 - Sep 1843) has a unit root
Exogenous: Constant
Lag Length: 8 (Automatics-based on AIC, MaxLag = 8)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -8.558407 0.0000
Test Critical vales 1% level
5% level
10% level
-3.491345
-2.888157
-2.581041
Null Hypothesis H0: Sunspots Cycle (Sep1843 - Mar 1855) has a unit root
Exogenous: Constant
Lag Length: 12 (Automatics-based on AIC, MaxLag = 13)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -6.745618 0.0000
Test Critical vales 1% level
5% level
10% level
-3.483751
-2.884856
-2.579282
Null Hypothesis H0: Sunspots Cycle (Mar1855 - Feb 1867) has a unit root
Exogenous: Constant
Lag Length: 13 (Automatics-based on AIC, MaxLag = 13)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -6.933791 0.0000
203
Test Critical vales 1% level
5% level
10% level
-3.482035
-2.884109
-2.578884
Null Hypothesis H0: Sunspots Cycle (Feb1867 - Sep 1878) has a unit root
Exogenous: Constant
Lag Length: 12 (Automatics-based on AIC, MaxLag = 13)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -5.950719 0.0000
Test Critical vales 1% level
5% level
10% level
-3.483312
-2.884665
-2.579180
Null Hypothesis H0: Sunspots Cycle (Sep1878 - Jun 1890) has a unit root
Exogenous: Constant
Lag Length: 13 (Automatics-based on AIC, MaxLag = 13)
t-statistics prob*
Augmented Dickey Fuller (ADF) test - 6.602668 0.0000
Test Critical vales 1% level
5% level
10% level
-3.482879
-2.884477
-2.579080
Null Hypothesis H0: Sunspots Cycle (Jun1890 - Sep 1902) has a unit root
Exogenous: Constant
Lag Length: 13 (Automatics-based on AIC, MaxLag = 13)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -7.700490 0.0000
Test Critical vales 1% level
5% level
10% level
-3.480485
-2.883408
-2.578510
Null Hypothesis H0: Sunspots Cycle (Sep 1902 - Dec 1913) has a unit root
Exogenous: Constant
Lag Length: 10 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -8.106940 0.0000
Test Critical vales 1% level
5% level
10% level
-3.484198
-2.885051
-2.579368
204
Null Hypothesis H0: Sunspots Cycle (Dec1913 - May 1923) has a unit root
Exogenous: Constant
Lag Length: 8 (Automatics-based on AIC, MaxLag = 10)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -7.087756 0.0000
Test Critical vales 1% level
5% level
10% level
-3.495021
-2.889753
-2.581890
Null Hypothesis H0: Sunspots Cycle (May 1923 - Sep 1933) has a unit root
Exogenous: Constant
Lag Length: 4 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -10.74242 0.0000
Test Critical vales 1% level
5% level
10% level
-3.486551
-2.886074
-2.579931
Null Hypothesis H0: Sunspots Cycle (Sep 1933 - Jan 1944) has a unit root
Exogenous: Constant
Lag Length: 12 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -6.524422 0.0000
Test Critical vales 1% level
5% level
10% level
-3.490772
-2.887909
-2.580908
Null Hypothesis H0: Sunspots Cycle (Jan 1944 - Feb 1954) has a unit root
Exogenous: Constant
Lag Length: 9 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -7.167179 0.0000
Test Critical vales 1% level
5% level
10% level
-3.490772
-2.889079
-2.580908
Null Hypothesis H0: Sunspots Cycle (Feb 1954 - Oct 1964) has a unit root
Exogenous: Constant
205
Lag Length: 11 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -7.122662 0.0000
Test Critical vales 1% level
5% level
10% level
-3.488063
-2.886732
-2.580281
Null Hypothesis H0: Sunspots Cycle (Oct 1964 - May 1976) has a unit root
Exogenous: Constant
Lag Length: 9 (Automatics-based on AIC, MaxLag = 13)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -6.852907 0.0000
Test Critical vales 1% level
5% level
10% level
-3.482035
-2.884109
-2.578884
Null Hypothesis H0: Sunspots Cycle (May1976 - Mar 1986) has a unit root
Exogenous: Constant
Lag Length: 10 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -7.960910 0.0000
Test Critical vales 1% level
5% level
10% level
-3.493129
-2.888932
-2.581453
Null Hypothesis H0: Sunspots Cycle (Mar1986 - Jun1996) has a unit root
Exogenous: Constant
Lag Length: 10 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -7.082791 0.0000
Test Critical vales 1% level
5% level
10% level
-3.490210
-2.887665
-2.580778
Null Hypothesis H0: Sunspots Cycle (Jun1996 - Jan 2008) has a unit root
Exogenous: Constant
Lag Length: 13 (Automatics-based on AIC, MaxLag = 13)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -7.048743 0.0000
206
Test Critical vales 1% level
5% level
10% level
-3.483751
-2.884856
-2.579282
Null Hypothesis H0: Sunspots Cycle (Aug1755 - Jan 2008) has a unit root
Exogenous: Constant
Lag Length: 4 (Automatics-based on AIC, MaxLag = 12)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -17.69189 0.0000
Test Critical vales 1% level
5% level
10% level
-3.432338
-2.862304
-2.567221
Table 1.3: Descriptive Statistics of Sunspot Cycles
C Duration N Mean Median Std.D Skewness Kurtosis Jur-Bera
1 Aug 1755 - Mar 1766 128 43.81 43.40 24.35 0.376 2.428 4.782
2 Mar 1766 - Aug 1775 144 57.04 51.25 39.21 0.582 2.648 7.035
3 Aug 1775 - Jun 1784 107 68.95 54.30 52.12 0.665 2.686 8.319
4 Jun 1784 - Jun 1798 169 59.92 53.00 54.44 0.512 2.122 12.82
5 Jun 1798 - Sep 1810 148 23.05 23.75 18.71 0.273 1.721 11.96
6 Sep1810 - Dec 1823 160 17.76 14.40 17.98 1.269 4.932 67.81
7 Dec1823 - Oct 1833 119 40.06 42.30 26.01 0.304 2.331 4.055
8 Oct1833 - Sep 1843 120 65.39 59.40 48.31 0.568 2.518 8.018
9 Sep1843 - Mar 1855 139 59.37 54.80 38.41 0.930 3.480 21.37
10 Mar1855 - Feb 1867 144 45.86 40.85 32.57 0.320 1.991 8.558
11 Feb1867 - Sep 1878 140 53.59 39.60 46.81 0.718 2.395 14.18
12 Sep1878 - Jun 1890 142 32.55 27.15 26.99 0.512 2.054 11.52
13 Jun1890 - Sep 1902 148 37.54 30.55 30.63 0.553 2.234 11.16
14 Sep 1902 - Dec 1913 136 32.68 31.95 26.33 0.596 2.793 8.293
15 Dec1913 - May 1923 114 46.53 42.15 33.97 0.703 3.021 9.380
16 May 1923 - Sep 1933 125 39.79 34.40 28.73 0.372 1.922 8.936
17 Sep 1933 - Jan 1944 125 57.64 54.60 40.52 0.400 2.159 7.034
18 Jan 1944 - Feb 1954 122 74.43 61.45 54.71 0.378 1.985 8.145
19 Feb 1954 - Oct 1964 129 89.04 63.60 71.52 0.431 1.816 11.54
20 Oct 1964 - May 1976 140 60.26 57.35 37.29 0.159 1.738 9.887
21 May1976 - Mar 1986 119 83.57 82.70 57.22 0.154 1.616 9.985
22 Mar1986 - Jun1996 124 76.15 60.35 58.50 0.410 1.754 11.49
23 Jun1996 - Jan 2008 140 57.94 48.90 42.67 0.425 2.018 9.834
24 Aug1755 - Jan 2008 3030 52.26 42.10 44.60 1.066 3.741 643.62
Table 1.4: Best Fitted ARMA Model based on DURBIN WATSON STAT
Sunspot Cycle (Aug 1755 - Mar 1766)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HQC DWS
207
1 ARMA (6,5) 0.437 0.424 18.509 -554.25 8.723 8.812 8.759 0.798
2 ARMA (5,5) 0.474 0.461 17.848 -550.72 8.667 8.757 8.704 0.959
3 ARMA (5,4) 0.523 0.511 17.049 -543.86 8.560 8.649 8.597 0.976
4 ARMA (4,5) 0.523 0.511 17.050 -543.79 8.559 8.648 8.595 1.051
5 ARMA(4,4) 0.527 0.516 16.972 -543.77 8.559 8.648 8.595 1.096
Dynamics Forecasted Evolution of Sunspot Cycle (Aug 1755 - Mar 1766)
No# Model RMSE MAE MAPE U
1 ARMA (6,5) 22.96202 18.55645 64.93243 0.251178
2 ARMA (5,5) 25.78878 20.41369 58.87236 0.306819
3 ARMA (5,4) 23.01596 18.60317 66.29410 0.252782
4 ARMA (4,5) 22.59682 18.43576 61.99027 0.255410
5 ARMA(4,4) 25.49785 20.23572 55.73550 0.303910
Sunspot Cycle (Mar 1766 - Aug 1775)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (5,3) 0.439 0.424 29.763 -547.40 9.674 9.770 9.713 0.719
2 ARMA (3,3) 0.524 0.511 27.407 -539.57 9.519 9.615 9.557 0.855
3 ARMA (3,2) 0.497 0.483 28.190 -541.00 9.561 9.657 9.600 0.949
4 ARMA(2,2) 0.575 0.563 25.912 -531.90 9.402 9.498 9.441 1.104
5 ARMA (2,6) 0.540 0.528 26.942 -536.02 9.474 9.560 9.513 1.197
Dynamics Forecasted Evolution of Sunspot Cycle (Mar 1766 - Aug 1775)
No# Model RMSE MAE MAPE U
1 ARMA (5,3) 37.92649 29.55789 161.1660 0.309054
2 ARMA (3,3) 43.21607 33.35958 149.9696 0.395928
3 ARMA (3,2) 38.15855 30.36035 185.7616 0.309936
4 ARMA(2,2) 43.71936 33.78659 145.5506 0.405274
5 ARMA (2,6) 38.07863 30.05593 179.8060 0.312458
Sunspot Cycle (Aug 1775 - Jun 1784)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,5) 0.722 0.714 27.882 -508.60 9.581 9.681 9.622 0.846
2 ARMA(5,6) 0.751 0.743 26.402 -502.95 9.746 9.576 9.516 0.993
3 ARMA (5,4) 0.783 0.777 24.607 -495.37 9.334 9.434 9.375 1.007
4 ARMA (5,5) 0.736 0.728 27.176 -506.63 9.544 9.644 9.585 1.155
5 ARMA (6,4) 0.736 0.728 27.174 -505.94 9.532 9.631 9.572 1.162
Dynamics Forecasted Evolution of Sunspot Cycle (Aug 1775 - Jun 1784)
No# Model RMSE MAE MAPE U
1 ARMA(6,5) 57.08266 43.80487 121.4064 0.415818
2 ARMA(5,6) 60.72247 46.10220 130.2028 0.465018
3 ARMA (5,4) 58.41590 44.52910 107.1816 0.436066
4 ARMA (5,5) 61.03474 46.27304 102.4356 0.469253
5 ARMA(6,4) 56.72805 43.6452 122.6841 0.411658
Sunspot Cycle (Jun 1784 - Jun 1798)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,5) 0.810 0.807 19.962 -784.12 8.901 8.975 8.931 0.904
2 ARMA(6,4) 0.820 0.816 19.476 -743.78 8.849 8.824 8.879 0.977
3 ARMA (5,6) 0.813 0.809 19.849 -746.37 8.880 8.954 8.910 1.078
4 ARMA (3,3) 0.782 0.869 16.415 -713.90 8.496 8.570 8.526 1.150
5 ARMA (5,4) 0.807 0.803 20.153 -749.04 8.912 8.986 8.942 1.172
208
Dynamics Forecasted Evolution of Sunspot Cycle (Jun 1784 - Jun 1798)
No# Model RMSE MAE MAPE U
1 ARMA (6,5) 55.57598 43.24828 171.1271 0.498712
2 ARMA(6,4) 54.53012 42.47551 175.9627 0.481473
3 ARMA (5,6) 53.38909 41.59299 180.1554 0.465585
4 ARMA (3,3) 59.78222 45.79778 141.12189 0.587859
5 ARMA (5,4) 54.15024 42.15394 176.6212 0.477880
Sunspot Cycle (Jun 1798 - Sep 1810)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,4) 0.670 0.693 10.358 -556.64 7.576 7.657 7.609 0.794
2 ARMA(4,4) 0.749 0.743 9.475 -543.74 7.402 7.483 7.435 0.875
3 ARMA (5,3) 0.729 0.723 9.826 -584.79 7.470 7.551 7.503 0.994
4 ARMA (3,3) 0.777 0.772 8.933 -534.55 7.278 7.359 7.311 0.998
5 ARMA (6,2) 0.731 0.723 9.797 -548.48 7.466 7.547 7.499 0.999
Dynamics Forecasted Evolution of Sunspot Cycle (Jun 1798 - Sep 1810)
No# Model RMSE MAE MAPE U
1 ARMA (6,4) 19.64111 16.86298 224.7709 0.416733
2 ARMA(4,4) 22.96767 18.54974 172.6923 0.559893
3 ARMA (5,3) 19.61195 16.77907 222.5072 0.418899
4 ARMA (3,3) 23.56660 18.78161 161.6836 0.592059
5 ARMA (6,2) 19.71194 16.93144 225.7594 0.418814
Sunspot Cycle (Sep1810 - Dec 1823)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,5) 0.521 0.511 12.570 -631.50 7.944 8.021 7.975 1.187
2 ARMA(6,6) 0.503 0.494 12.793 -635.00 7.987 8.064 8.018 1.229
3 ARMA (6,2) 0.490 0.480 12.957 -636.60 8.007 8.084 8.039 1.283
4 ARMA (3,4) 0.507 0.498 12.734 -632.97 7.962 8.039 7.993 1.300
5 ARMA (6,4) 0.543 0.534 12.269 -627.70 7.896 7.973 7.927 1.325
Dynamics Forecasted Evolution of Sunspot Cycle (Sep1810 - Dec 1823)
No# Model RMSE MAE MAPE U
1 ARMA (6,5) 17.85740 13.28661 261.7866 0.429515
2 ARMA(6,6) 18.43876 13.58944 215.0192 0.482890
3 ARMA (6,2) 17.67344 13.27179 251.9070 0.437088
4 ARMA (3,4) 17.58213 13.69847 296.0727 0.418201
5 ARMA (6,4) 17.58113 13.25282 261.6172 0.429342
Sunspot Cycle (Dec1823 - Oct 1833)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (5,5) 0.531 0.518 18.051 -513.30 8.694 8.788 8.732 0.926
2 ARMA(6,4) 0.435 0.421 19.798 -523.20 8.861 8.954 8.898 0.971
3 ARMA (5,3) 0.523 0.511 18.189 -513.19 8.692 8.786 8.730 0.993
4 ARMA (4,4) 0.559 0.548 17.492 -509.35 8.628 8.721 8.666 1.054
5 ARMA (5,2) 0.529 0.516 18.092 -512.59 8.682 8.775 8.720 1.112
Dynamics Forecasted Evolution of Sunspot Cycle (Dec1823 - Oct 1833)
No# Model RMSE MAE MAPE U
1 ARMA (5,5) 28.54622 23.13027 125.1342 0.376961
2 ARMA(6,4) 25.01170 20.63166 160.3185 0.297296
3 ARMA (5,3) 24.84239 20.57404 159.1863 0.294667
209
4 ARMA (4,4) 29.44770 23.99430 142.4501 0.396488
5 ARMA (5,2) 24.99444 20.66547 156.7645 0.297884
Sunspot Cycle (Oct1833 - Sep 1843)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (5,6) 0.663 0.654 28.398 -571.85 9.598 9.690 9.635 0.767
2 ARMA(4,6) 0.657 0.648 28.648 -572.46 9.608 9.701 9.645 0.907
3 ARMA (6,3) 0.688 0.680 27.324 -567.40 9.523 9.616 9.561 0.918
4 ARMA (3,3) 0.695 0.687 27.031 -565.92 9.499 9.592 9.536 0.963
5 ARMA (6,2) 0.746 0.741 24.583 -555.23 9.321 9.413 9.358 0.978
Dynamics Forecasted Evolution of Sunspot Cycle (Oct1833 - Sep 1843)
No# Model RMSE MAE MAPE U
1 ARMA (5,6) 49.58514 38.8095 121.2569 0.371537
2 ARMA(4,6) 47.89669 37.64070 125.6561 0.353636
3 ARMA (6,3) 52.16269 40.57367 114.0542 0.400505
4 ARMA (3,3) 59.74495 45.34109 101.5025 0.518348
5 ARMA (6,2) 51.83747 40.41983 116.4244 0.396347
Sunspot Cycle (Sep1843 - Mar 1855)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,4) 0.511 0.500 27.162 -655.49 9.489 9.574 9.523 0.779
2 ARMA(4,3) 0.583 0.574 25.080 -644.24 9.327 9.412 9.362 0.913
3 ARMA (4,4) 0.587 0.578 24.946 -644.38 9.329 9.414 9.364 0.929
4 ARMA (5,3) 0.589 0.579 24.912 -643.51 9.317 9.401 9.351 0.981
5 ARMA (3,3) 0.651 0.643 22.946 -632.53 9.159 9.243 9.193 0.987
Dynamics Forecasted Evolution of Sunspot Cycle (Sep1843 - Mar 1855)
No# Model RMSE MAE MAPE U
1 ARMA (6,4) 37.01347 26.67266 62.85951 0.297524
2 ARMA(4,3) 36.85102 26.74996 66.02246 0.296860
3 ARMA (4,4) 43.45420 30.97244 55.55537 0.393127
4 ARMA (5,3) 36.91985 26.71517 64.07458 0.297030
5 ARMA (3,3) 43.32231 30.79417 55.33579 0.393812
Sunspot Cycle (Mar1855 - Feb 1867)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,5) 0.729 0.723 17.138 -614.66 8.593 8.675 8.626 0.993
2 ARMA(6,4) 0.728 0.722 17.161 -614.71 8.593 8.676 8.627 1.178
3 ARMA (4,4) 0.771 0.768 15.689 -306.53 8.424 8.486 8.449 1.222
4 ARMA (6,2) 0.733 0.728 16.997 -613.37 8.574 8.657 8.608 1.222
5 ARMA (5,3) 0.760 0.757 16.067 -607.35 8.477 8.539 8.502 1.285
Dynamics Forecasted Evolution of Sunspot Cycle (Mar1855 - Feb 1867)
No# Model RMSE MAE MAPE U
1 ARMA (6,5) 34.69173 26.74974 165.8685 0.393126
2 ARMA(6,4) 34.05871 26.32748 173.7933 0.318617
3 ARMA (4,4) 53.58256 43.77341 149.2235 0.856167
4 ARMA (6,2) 34.27413 26.50601 172.0539 0.385227
5 ARMA (5,3) 55.30789 45.35533 122.1337 0.903059
Sunspot Cycle (Feb1867 - Sep 1878)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
210
1 ARMA (6,5) 0.724 0.717 24.884 -649.72 9.339 9.423 9.373 0.680
2 ARMA(6,4) 0.750 0.745 23.624 -642.34 9.233 9.316 9.268 0.710
3 ARMA (5,4) 0.737 0.732 24.238 -645.69 9.281 9.365 9.315 0.873
4 ARMA (3,3) 0.801 0.797 21.115 -626.13 9.002 9.086 9.036 0.886
5 ARMA (5,3) 0.769 0.764 22.742 -636.62 9.152 9.236 9.188 0.905
Dynamics Forecasted Evolution of Sunspot Cycle (Feb1867 - Sep 1878)
No# Model RMSE MAE MAPE U
1 ARMA (6,5) 52.74751 41.71703 774.1111 0.490512
2 ARMA(6,4) 51.34360 40.82203 807.2583 0.467277
3 ARMA (5,4) 51.64772 40.83656 788.2092 0.475558
4 ARMA (3,3) 56.57545 42.82194 619.2955 0.576678
5 ARMA (5,3) 50.26837 39.90352 818.8082 0.453341
Sunspot Cycle (Sep1878 - Jun 1890)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (5,4) 0.596 0.587 17.341 -606.7 8.603 8.686 8.637 0.877
2 ARMA(5,3) 0.627 0.619 16.656 -600.50 8.514 8.597 8.548 0.888
3 ARMA (6,6) 0.570 0.560 17.898 -611.79 8.673 8.756 8.707 0.908
4 ARMA (6,4) 0.570 0.560 17.897 -610.87 8.660 8.743 8.694 0.977
5 ARMA (4,4) 0.631 0.623 16.574 -600.41 8.513 8.596 8.547 1.056
Dynamics Forecasted Evolution of Sunspot Cycle (Sep1878 - Jun 1890)
No# Model RMSE MAE MAPE U
1 ARMA (5,4) 28.20895 23.20957 384.5670 0.413804
2 ARMA(5,3) 26.94070 22.33445 409.9513 0.384873
3 ARMA (6,6) 31.12448 25.16658 329.3777 0.494468
4 ARMA (6,4) 27.37615 22.67076 394.1263 0.394094
5 ARMA (4,4) 31.58116 25.16165 317.31740 0.516661
Sunspot Cycle (Jun1890 - Sep 1902)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,4) 0.720 0.714 16.375 -624.57 8.494 8.575 8.527 1.122
2 ARMA (6,2) 0.723 0.717 16.282 -623.79 8.484 8.565 8.517 1.244
3 ARMA (4,4) 0.756 0.751 15.278 -614.38 8.356 8.437 8.389 1.254
4 ARMA (4,3) 0.748 0.743 15.519 -616.10 8.380 8.461 8.413 1.269
5 ARMA (5,4) 0.707 0.698 16.817 -627.95 8.540 8.621 8.573 1.272
Dynamics Forecasted Evolution of Sunspot Cycle (Jun1890 - Sep 1902)
No# Model RMSE MAE MAPE U
1 ARMA (6,4) 34.15608 28.14172 429.9099 0.440684
2 ARMA(6,2) 34.40949 28.24120 424.3318 0.446793
3 ARMA (4,4) 36.86275 29.26016 351.6125 0.520870
4 ARMA (4,3) 33.39690 27.52467 429.3716 0.429832
5 ARMA (5,4) 33.19800 27.39173 441.9179 0.418577
Sunspot Cycle (Sep 1902 - Dec 1913)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (5,6) 0.422 0.409 20.242 -600.95 8.896 8.982 8.931 0.980
2 ARMA(4,6) 0.436 0.423 19.993 -599.17 8.870 8.956 8.905 1.007
3 ARMA (5,5) 0.486 0.474 19.092 -594.02 8.794 8.880 8.829 1.060
4 ARMA (4,4) 0.526 0.515 18.333 -588.36 8.711 8.797 8.756 1.061
5 ARMA (5,3) 0.419 0.406 20.289 -602.26 8.916 9.001 8.950 1.064
211
Dynamics Forecasted Evolution of Sunspot Cycle (Sep 1902 - Dec 1913)
No# Model RMSE MAE MAPE U
1 ARMA (5,6) 26.60654 22.16956 582.2514 0.369668
2 ARMA(4,6) 26.33349 21.89815 586.2954 0.363862
3 ARMA (5,5) 30.20454 25.05934 469.8335 0.471314
4 ARMA (4,4) 30.33743 25.05251 447.7918 0.480269
5 ARMA (5,3) 27.75297 23.29113 547.1076 0.400474
Sunspot Cycle (Dec 1913 - May 1923)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,4) 0.483 0.469 24.750 -526.71 9.311 9.407 9.350 0.687
2 ARMA (5,4) 0.468 0.454 25.110 -528.20 9.337 9.433 9.376 0.743
3 ARMA (4,3) 0.592 0.580 22.006 -513.36 9.077 9.123 9.115 0.744
4 ARMA (5,5) 0.473 0.459 24.984 -528.49 9.342 9.438 9.380 0.810
5 ARMA (5,3) 0.546 0.533 23.207 -519.35 9.182 9.278 9.221 0.953
Dynamics Forecasted Evolution of Sunspot Cycle (Dec 1913 - May 1923)
No# Model RMSE MAE MAPE U
1 ARMA (6,4) 33.70195 26.41578 142.1956 0.341615
2 ARMA(5,4) 33.40646 26.17204 143.7501 0.338521
3 ARMA (4,3) 33.59338 26.19764 139.4187 0.345463
4 ARMA (5,5) 38.64505 30.04723 124.6412 0.434757
5 ARMA (5,3) 33.02818 26.04595 149.5811 0.329163
Sunspot Cycle (May 1923 - Sep 1933)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,5) 0.559 0.548 19.307 -547.01 8.816 8.907 8.853 0.565
2 ARMA(6,4) 0.584 0.574 18.760 -543.59 8.761 8.852 8.798 0.722
3 ARMA (4,4) 0.628 0.618 17.749 -537.13 8.658 8.746 8.695 0.792
4 ARMA (3,3) 0.666 0.658 16.814 -530.06 8.545 8.635 8.582 0.914
5 ARMA (5,3) 0.628 0.619 17.742 -536.60 8.650 8.740 8.686 0.920
Dynamics Forecasted Evolution of Sunspot Cycle (May 1923 - Sep 1933)
No# Model RMSE MAE MAPE U
1 ARMA (6,5) 28.41330 23.46526 286.6359 0.340449
2 ARMA(6,4) 28.92684 23.72642 284.4390 0.351530
3 ARMA (4,4) 34.14833 27.18319 225.7652 0.470814
4 ARMA (3,3) 34.83879 27.46117 226.5581 0.491752
5 ARMA (5,3) 28.94486 23.72149 276.0542 0.354408
Sunspot Cycle (Sep 1933 - Jan 1944)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,4) 0.601 0.592 25.898 -584.19 9.411 9.502 9.448 0.709
2 ARMA(4,4) 0.654 0.646 24.117 -575.64 9.274 9.365 9.311 0.908
3 ARMA (3,3) 0.710 0.703 22.091 -564.41 9.094 9.185 9.131 0.937
4 ARMA (5,3) 0.644 0.635 24.468 -576.77 9.292 9.383 9.329 0.940
5 ARMA (6,2) 0.684 0.676 23.073 -569.45 9.182 9.272 9.218 0.951
Dynamics Forecasted Evolution of Sunspot Cycle (Sep 1933 - Jan 1944)
No# Model RMSE MAE MAPE U
1 ARMA (6,4) 41.14730 32.48182 90.65688 0.355584
2 ARMA(4,4) 49.84429 38.45141 79.25677 0.491490
3 ARMA (3,3) 51.63485 39.71464 103.6560 0.526209
212
4 ARMA (5,3) 39.64346 31.48136 94.59616 0.336577
5 ARMA (6,2) 40.50563 32.10883 92.99856 0.346311
Sunspot Cycle (Jan 1944 - Feb 1954)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,4) 0.638 0.629 33.330 -601.22 9.922 10.014 9.959 0.583
2 ARMA(4,4) 0.719 0.713 29.323 -586.07 9.673 9.765 9.711 0.661
3 ARMA (6,2) 0.677 0.669 31.478 -594.25 9.806 9.898 9.843 0.867
4 ARMA (3,3) 0.750 0.743 27.733 -578.74 9.553 9.645 9.591 0.869
5 ARMA (5,3) 0.675 0.667 31.563 -594.15 9.806 9.898 9.843 0.907
Dynamics Forecasted Evolution of Sunspot Cycle (Jan 1944 - Feb 1954)
No# Model RMSE MAE MAPE U
1 ARMA (6,4) 58.17556 46.03504 487.9669 0.395454
2 ARMA(4,4) 69.35137 53.93403 374.3422 0.539219
3 ARMA (6,2) 52.26395 45.41299 496.7139 0.385152
4 ARMA (3,3) 69.75429 53.79824 380.6224 0.549886
5 ARMA (5,3) 55.25653 44.02761 508.1254 0.366021
Sunspot Cycle (Feb 1954 - Oct 1964)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,5) 0.781 0.775 33.895 -639.27 9.973 10.062 10.009 0.663
2 ARMA(5,4) 0.815 0.808 31.334 -628.68 9.809 9.898 9.845 0.836
3 ARMA (5,3) 0.854 0.850 27.611 -612.63 9.560 9.649 9.561 0.855
4 ARMA (6,4) 0.794 0.792 32.655 -643.31 9.896 9.985 9.932 0.887
5 ARMA (4,3) 0.848 0.844 28.216 -614.84 8.594 9.683 9.630 0.978
Dynamics Forecasted Evolution of Sunspot Cycle (Feb 1954 - Oct 1964)
No# Model RMSE MAE MAPE U
1 ARMA (6,5) 86.23169 66.88760 126.8783 0.522011
2 ARMA(5,4) 82.96249 64.61203 125.9374 0.491867
3 ARMA (5,3) 82.12713 64.21752 127.4141 0.483381
4 ARMA (6,4) 83.16715 65.28142 132.5322 0.489885
5 ARMA (4,3) 81.95903 64.00049 155.9173 0.485348
Sunspot Cycle (Oct 1964 - May 1976)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (5,3) 0.751 0.745 18.814 -609.96 8.771 8.855 8.805 0.947
2 ARMA(6,4) 0.684 0.677 21.205 -626.80 9.011 9.096 9.046 0.980
3 ARMA (4,3) 0.739 0.733 19.278 -613.09 8.816 8.900 8.850 1.048
4 ARMA (3,3) 0.781 0.776 17.662 -601.12 8.645 8.729 8.679 1.085
5 ARMA (6,2) 0.717 0.711 20.050 -619.01 8.900 8.984 8.934 1.114
Dynamics Forecasted Evolution of Sunspot Cycle (Oct 1964 - May 1976)
No# Model RMSE MAE MAPE U
1 ARMA (5,3) 38.99557 32.35205 88.63447 0.328025
2 ARMA(6,4) 39.73977 32.97053 88.18282 0.335997
3 ARMA (4,3) 39.00213 32.19808 88.14809 0.329886
4 ARMA (3,3) 47.63696 38.07346 78.84340 0.454724
5 ARMA (6,2) 39.89365 33.13202 88.41728 0.337784
Sunspot Cycle (May1976 - Mar 1986)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
213
1 ARMA (6,4) 0.695 0.687 31.995 -582.02 9.849 9.943 9.887 0.828
2 ARMA(6,2) 0.705 0.697 31.486 -580.04 9.816 9.909 9.854 1.026
3 ARMA (3,3) 0.773 0.767 27.641 -564.29 9.551 9.644 9.589 1.050
4 ARMA (5,3) 0.731 0.724 30.069 -574.29 9.719 9.813 9.757 1.014
5 ARMA (5,2) 0.746 0.739 29.228 -570.97 9.663 9.757 9.701 1.126
Dynamics Forecasted Evolution of Sunspot Cycle (May1976 - Mar 1986)
No# Model RMSE MAE MAPE U
1 ARMA (6,4) 63.33039 52.49356 120.4046 0.393470
2 ARMA(6,2) 62.69861 52.05456 121.8634 0.386909
3 ARMA (3,3) 75.57275 60.66088 99.64765 0.540982
4 ARMA (5,3) 61.20026 50.64551 121.8423 0.375419
5 ARMA (5,2) 61.73283 51.04290 121.2593 0.380368
Sunspot Cycle (Mar1986 - Jun1996)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,5) 0.758 0.752 29.114 -595.39 9.667 9.759 9.704 0.834
2 ARMA(6,4) 0.775 0.770 28.083 -590.87 9.594 9.686 9.631 0.946
3 ARMA (5,6) 0.766 0.760 28.653 -592.93 9.628 9.719 9.665 1.011
4 ARMA (5,4) 0.764 0.758 28.777 -593.41 9.636 9.727 9.673 1.076
5 ARMA (4,5) 0.781 0.776 27.690 -588.33 9.554 9.645 9.591 1.117
Dynamics Forecasted Evolution of Sunspot Cycle (Mar1986 - Jun1996)
No# Model RMSE MAE MAPE U
1 ARMA (6,5) 67.60252 53.31430 120.4492 0.466438
2 ARMA(6,4) 66.42465 52.50765 122.5996 0.452597
3 ARMA (5,6) 65.05685 51.96270 132.4873 0.439658
4 ARMA (5,4) 65.13699 52.01062 132.6014 0.440693
5 ARMA (4,5) 64.18852 50.90771 126.8260 0.433141
Sunspot Cycle (Jun1996 - Jan 2008)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,4) 0.714 0.708 23.056 -638.59 9.180 9.264 9.214 0.880
2 ARMA(5,4) 0.733 0.727 22.274 -633.79 9.111 9.195 9.145 0.964
3 ARMA (4,3) 0.788 0.783 19.866 -617.50 8.879 8.963 8.913 1.008
4 ARMA (3,4) 0.791 0.787 19.712 -615.96 8.857 8.940 8.890 1.022
5 ARMA (3,3) 0.792 0.787 19.666 -616.22 8.860 8.944 8.894 1.073
Dynamics Forecasted Evolution of Sunspot Cycle (Jun1996 - Jan 2008)
No# Model RMSE MAE MAPE U
1 ARMA (6,4) 45.06462 36.50313 187.5552 0.384967
2 ARMA(5,4) 46.62530 37.34643 178.4829 0.410093
3 ARMA (4,3) 45.43849 36.59640 194.0759 0.394859
4 ARMA (3,4) 43.79293 35.46916 204.3410 0.373450
5 ARMA (3,3) 52.87249 41.31675 160.2933 0.518691
Sunspot Cycle (Aug1755 - Jan 2008)
No# Model R2 ADJ R2 SE Reg Log L AIC SIC HIC DWS
1 ARMA (6,6) 0.709 0.708 24.088 -13941.6 9.205 9.213 9.208 0.818
2 ARMA(6,5) 0.712 0.711 23.969 -13926.0 9.195 9.203 9.196 0.823
3 ARMA (6,4) 0.724 0.724 23.438 -13858.0 9.150 9.158 9.153 0.881
4 ARMA (5,5) 0.736 0.740 22.921 -13790.9 9.106 9.113 9.108 0.893
5 ARMA (5,4) 0.732 0.732 23.105 -13814.7 9.121 9.129 9.124 0.933
214
Dynamics Forecasted Evolution of Sunspot Cycle (Aug1755 - Jan 2008)
No# Model RMSE MAE MAPE U
1 ARMA (6,6) 44.59329 35.32464 465.4666 0.372824
2 ARMA(6,5) 44.55660 35.32846 468.5673 0.371180
3 ARMA (6,4) 44.54175 35.33142 469.6662 0.370626
4 ARMA (5,5) 44.59715 35.30753 464.3263 0.373367
5 ARMA (5,4) 44.54582 35.31252 468.8461 0.370987
Table 1.5: AR(p) - GRACH(1, 1) diagnostic checking and forecasting
evolution of sunspots cycles
Dynamics Forecasted Evolution
RMSE MAE MAPE U
27.61325 21.41193 56.70953 0.348243
Sunspot Cycle (Mar 1766 - Aug 1775)
Method : ML ARCH- Normal Distribution
GARCH = 11.338 + 0.227*Resid(-1)^2 + 0.775*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.824378
0.226696
0.774968
0.047120
0.140947
0.110833
17.49543
1.608383
6.992243
0.0000
0.1078
0.0000
R-Square 0.51076 Durbin-Watson Stat 1.360450
Adjusted R-Square 0.506394 Akaike Info Criterion 9.108496
S.E. of Regression 27.54760 Schwarz – Criterion 9.228505
Log likelihood -514.1843 Hannan-Quinn Criter 9.157201
Dynamics Forecasted Evolution
RMSE MAE MAPE U
48.95338 37.48065 115.7679 0.501998
Sunspot Cycle (Aug 1775 - Jun 1784)
Method : ML ARCH- Normal Distribution
GARCH = 16.066 + 0.2738*Resid(-1)^2 + 0.7406*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
Sunspot Cycle (Aug 1755 - Mar 1766)
Method : ML ARCH- Normal Distribution
GARCH = 12.140 + 0.182*Resid(-1)^2 + 0.776*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.833626
0.181855
0.776212
0.045517
0.077604
0.080125
18.31450
2.340945
9.687482
0.0000
0.0192
0.0000
R-Square 0.637689 Durbin-Watson Stat 1.582369
Adjusted R-Square 0.634814 Akaike Info Criterion 8.192929
S.E. of Regression 14.73627 Schwarz – Criterion 8.304337
Log likelihood -519.3475 Hannan-Quinn Criter 8.238195
215
AR(2)
Resid(-1)^2
GARCH(-1)
0.927650
0.273756
0.740552
0.043225
0.141865
0.112860
21.46112
1.929694
6.561700
0.0000
0.0536
0.0000
R-Square 0.762877 Durbin-Watson Stat 1.525458
Adjusted R-Square 0.760619 Akaike Info Criterion 8.905641
S.E. of Regression 25.50199 Schwarz – Criterion 9.030540
Log likelihood -417.4518 Hannan-Quinn Criter 8.95674
Dynamics Forecasted Evolution
RMSE MAE MAPE U
81.25105 62.90952 80.41010 0.859162
Sunspot Cycle (Jun 1784 - Jun 1798)
Method : ML ARCH- Normal Distribution
GARCH = 13.999 + 0.3505*Resid(-1)^2 + 0.6393*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.999993
0.350577
0.639280
1.39E-05
0.095717
0.066112
7213617
3.662653
9.669657
0.0000
0.0002
0.0000
R-Square 0.857579 Durbin-Watson Stat 1.85337
Adjusted R-Square 0.856726 Akaike Info Criterion 8.250305
S.E. of Regression 17.19911 Schwarz – Criterion 8.342905
Log likelihood -692.1507 Hannan-Quinn Criter 8.287884
Dynamics Forecasted Evolution
RMSE MAE MAPE U
69.70664 53.90236 84.66272 0.837476
Sunspot Cycle (Jun 1798 - Sep 1810)
Method : ML ARCH- Normal Distribution
GARCH = -0.0848 + 0.5254*Resid(-1)^2 + 0.6533*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.923855
0.525379
0.653287
0.019588
0.136829
0.058282
47.16485
13.83969
11.20897
0.0000
0.0001
0.0000
R-Square 0.789767 Durbin-Watson Stat 1.352767
Adjusted R-Square 0.788327 Akaike Info Criterion 6.837493
S.E. of Regression 8.607955 Schwarz – Criterion 6.938750
Log likelihood -500.9745 Hannan-Quinn Criter 6.878633
Dynamics Forecasted Evolution
RMSE MAE MAPE U
29.84910 23.36523 87.63108 0.9998639
Sunspot Cycle (Sep1810 - Dec 1823)
Method : ML ARCH- Normal Distribution
GARCH = 19.466+ 0.1187*Resid(-1)^2 + 0.8839*GARCH(-1)
216
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.779258
0.118733
0.883948
0.065885
0.023108
0.021690
11.82746
5.138217
40.75431
0.0000
0.0000
0.0000
R-Square 0.409344 Durbin-Watson Stat 1.604565
Adjusted R-Square 0.405605 Akaike Info Criterion 7.369178
S.E. of Regression 13.86322 Schwarz – Criterion 7.735277
Log likelihood -606.1342 Hannan-Quinn Criter 7.678200
Dynamics Forecasted Evolution
RMSE MAE MAPE U
23.87355 16.61689 81.56241 0.867214
Sunspot Cycle (Dec1823 - Oct 1833)
Method : ML ARCH- Normal Distribution
GARCH = 99.1634+ 0.1971*Resid(-1)^2 + 0.5125*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(3)
Resid(-1)^2
GARCH(-1)
0.717154
0.197108
0.512512
0.065084
0.088165
0.197635
11.01844
2.23579
2.593227
0.0000
0.0254
0.0095
R-Square 0.512241 Durbin-Watson Stat 1.152532
Adjusted R-Square 0.508072 Akaike Info Criterion 8.668157
S.E. of Regression 18.24313 Schwarz – Criterion 8.784927
Log likelihood -510.7554 Hannan-Quinn Criter 8.715574
Dynamics Forecasted Evolution
RMSE MAE MAPE U
26.97423 22.29147 148.6877 0.342102
Sunspot Cycle (Oct1833 - Sep 1843)
Method : ML ARCH- Normal Distribution
GARCH = 28.7902 + 0.3570*Resid(-1)^2 + 0.6217*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.999994
0.357004
0.621699
2.15E-05
0.168878
0.162796
46403.42
2.113973
3.818886
0.0000
0.0345
0.0001
R-Square 0.722466 Durbin-Watson Stat 1.277954
Adjusted R-Square 0.720114 Akaike Info Criterion 9.065138
S.E. of Regression 25.55909 Schwarz – Criterion 9.181284
Log likelihood -538.9083 Hannan-Quinn Criter 9.112305
Dynamics Forecasted Evolution
RMSE MAE MAPE U
76.51305 59.98412 82.81522 0.863572
Sunspot Cycle (Sep1843 - Mar 1855)
Method : ML ARCH- Normal Distribution
217
GARCH = 10.6170 + 0.1696*Resid(-1)^2 + 0.8260*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.892649
0.169634
0.826009
0.0406181
0.049536
0.057499
19.32935
3.424440
14.25403
0.0000
0.0006
0.0000
R-Square 0.660677 Durbin-Watson Stat 1.512800
Adjusted R-Square 0.658200 Akaike Info Criterion 8.797048
S.E. of Regression 22.45658 Schwarz – Criterion 8.902604
Log likelihood -606.3948 Hannan-Quinn Criter 8.839943
Dynamics Forecasted Evolution
RMSE MAE MAPE U
54.20801 41.01302 59.41300 0.590749
Sunspot Cycle (Mar1855 - Feb 1867)
Method : ML ARCH- Normal Distribution
GARCH = 4.3495 + 0.1627*Resid(-1)^2 + 0.8320*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.939019
0.162706
0.823016
0.031614
0.077090
0.068796
29.70280
2.110597
12.09390
0.0000
0.0348
0.0000
R-Square 0.793971 Durbin-Watson Stat 1.647376
Adjusted R-Square 0.792520 Akaike Info Criterion 8.060437
S.E. of Regression 14.83685 Schwarz – Criterion 8.163556
Log likelihood -575.3515 Hannan-Quinn Criter 8.102339
Dynamics Forecasted Evolution
RMSE MAE MAPE U
46.84207 36.86088 163.9772 0.677314
Sunspot Cycle (Feb1867 - Sep 1878)
Method : ML ARCH- Normal Distribution
GARCH = 4.6125 + 0.2670*Resid(-1)^2 + 0.7675*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.939751
0.267000
0.767531
0.029445
0.101809
0.069816
31.91558
2.622550
10.99364
0.0000
0.0085
0.0000
R-Square 0.793359 Durbin-Watson Stat 1.523139
Adjusted R-Square 0.791862 Akaike Info Criterion 8.558675
S.E. of Regression 21.35894 Schwarz – Criterion 8.663734
Log likelihood -594.1073 Hannan-Quinn Criter 8.601368
Dynamics Forecasted Evolution
RMSE MAE MAPE U
66.48149 48.35055 184.8052 0.845125
Sunspot Cycle (Sep1878 - Jun 1890)
218
Method : ML ARCH- Normal Distribution
GARCH = 1.6702 + 0.3061*Resid(-1)^2 + 0.7502*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.890694
0.306069
0.750207
0.044479
0.087816
0.062365
20.02497
3.485335
12.02933
0.0000
0.0005
0.0000
R-Square 0.670852 Durbin-Watson Stat 1.505106
Adjusted R-Square 0.668501 Akaike Info Criterion 7.968635
S.E. of Regression 15.54072 Schwarz – Criterion 8.072714
Log likelihood -560.7731 Hannan-Quinn Criter 8.010928
Dynamics Forecasted Evolution
RMSE MAE MAPE U
39.33113 29.41939 110.6451 0.841212
Sunspot Cycle (Jun1890 - Sep 1902)
Method : ML ARCH- Normal Distribution
GARCH = 4.8575 + 0.2405*Resid(-1)^2 + 0.7672*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.903458
0.240492
0.767205
0.039764
0.117877
0.090512
22.72052
2.040189
8.476297
0.0000
0.0413
0.0000
R-Square 0.764905 Durbin-Watson Stat 1.687333
Adjusted R-Square 0.763295 Akaike Info Criterion 8.021805
S.E. of Regression 14.90059 Schwarz – Criterion 8.123062
Log likelihood -588.6136 Hannan-Quinn Criter 8.062945
Dynamics Forecasted Evolution
RMSE MAE MAPE U
41.43307 30.79801 168.5865 0.703145
Sunspot Cycle (Sep 1902 - Dec 1913)
Method : ML ARCH- Normal Distribution
GARCH = 1.0075 + 0.2232*Resid(-1)^2 + 0.8033*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.803072
0.250610
0.789312
0.040800
0.065314
0.035812
19.68333
3.836991
22.04036
0.0000
0.0001
0.0000
R-Square 0.454177 Durbin-Watson Stat 1.238019
Adjusted R-Square 0.450104 Akaike Info Criterion 8.165793
S.E. of Regression 19.52373 Schwarz – Criterion 8.272876
Log likelihood -550.2739 Hannan-Quinn Criter 8.209309
Dynamics Forecasted Evolution
RMSE MAE MAPE U
37.13987 28.22192 163.2702 0.755452
219
Sunspot Cycle (Dec1913 - May 1923)
Method : ML ARCH- Normal Distribution
GARCH = 19.1145 + 0.2182*Resid(-1)^2 + 0.7588*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(3)
Resid(-1)^2
GARCH(-1)
0.856972
0.218223
0.758773
0.05336
0.120566
0.103634
16.05851
1.809981
7.321672
0.0000
0.0703
0.0000
R-Square 0.597334 Durbin-Watson Stat 1.053902
Adjusted R-Square 0.593739 Akaike Info Criterion 8.844163
S.E. of Regression 21.65239 Schwarz – Criterion 8.964172
Log likelihood -499.1173 Hannan-Quinn Criter 8.892868
Dynamics Forecasted Evolution
RMSE MAE MAPE U
50.69620 39.37540 80.53190 0.747014
Sunspot Cycle (May 1923 - Sep 1933)
Method : ML ARCH- Normal Distribution
GARCH = 10.9234 + 0.2912*Resid(-1)^2 + 0.6939*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.908219
0.291231
0.693905
0.036128
0.137997
0.129705
25.13905
2.110422
5.349893
0.0000
0.0348
0.0000
R-Square 0.697778 Durbin-Watson Stat 1.287276
Adjusted R-Square 0.69532 Akaike Info Criterion 8.168819
S.E. of Regression 15.85856 Schwarz – Criterion 8.281952
Log likelihood -505.5512 Hannan-Quinn Criter 8.214779
Dynamics Forecasted Evolution
RMSE MAE MAPE U
41.61186 32.01947 136.7743 0.701884
Sunspot Cycle (Sep 1933 - Jan 1944)
Method : ML ARCH- Normal Distribution
GARCH = 5.3230 + 0.1840*Resid(-1)^2 + 0.8284*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(3)
Resid(-1)^2
GARCH(-1)
0.920729
0.184035
0.828438
0.040200
0.101441
0.074076
22.90396
1.81214
11.18357
0.0000
0.0696
0.0000
R-Square 0.660145 Durbin-Watson Stat 1.147542
Adjusted R-Square 0.657382 Akaike Info Criterion 8.915888
S.E. of Regression 23.71966 Schwarz – Criterion 9.029021
Log likelihood -552.2430 Hannan-Quinn Criter 8.961848
Dynamics Forecasted Evolution
220
RMSE MAE MAPE U
66.90030 53.93148 94.93236 0.874060
Sunspot Cycle (Jan 1944 - Feb 1954)
Method : ML ARCH- Normal Distribution
GARCH = 14.0134+ 0.1392*Resid(-1)^2 + 0.8575*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.955562
0.139156
0.857469
0.035750
0.066739
0.055771
26.72916
2.085097
15.37487
0.0000
0.0371
0.0000
R-Square 0.784001 Durbin-Watson Stat 1.316423
Adjusted R-Square 0.782201 Akaike Info Criterion 9.158523
S.E. of Regression 25.53388 Schwarz – Criterion 9.273342
Log likelihood -553.6699 Hannan-Quinn Criter 9.205200
Dynamics Forecasted Evolution
RMSE MAE MAPE U
88.43862 70.49887 127.5823 0.627014
Sunspot Cycle (Feb 1954 - Oct 1964)
Method : ML ARCH- Normal Distribution
GARCH = 6.4314 + 0.2348*Resid(-1)^2 + 0.7856*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.933207
0.234765
0.785609
0.025720
0.089860
0.068461
36.28382
2.612577
11.47520
0.0000
0.0090
0.0000
R-Square 0.863349 Durbin-Watson Stat 1.593149
Adjusted R-Square 0.862273 Akaike Info Criterion 9.061693
S.E. of Regression 26.54211 Schwarz – Criterion 9.172538
Log likelihood -579.4792 Hannan-Quinn Criter 9.106732
Dynamics Forecasted Evolution
RMSE MAE MAPE U
110.3128 84.95945 101.2019 0.910967
Sunspot Cycle (Oct 1964 - May 1976)
Method : ML ARCH- Normal Distribution
GARCH = 18.0514 + 0.0448*Resid(-1)^2 + 0.8955*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.953305
0.044846
0.895546
0.034193
0.049779
0.087291
27.88030
0.900918
10.25934
0.0000
0.3676
0.0000
R-Square 0.795757 Durbin-Watson Stat 1.563318
Adjusted R-Square 0.794277 Akaike Info Criterion 8.478663
S.E. of Regression 16.91348 Schwarz – Criterion 8.583721
Log likelihood -588.5064 Hannan-Quinn Criter 8.521355
221
Dynamics Forecasted Evolution
RMSE MAE MAPE U
59.36716 47.77713 71.04261 0.921958
Sunspot Cycle (May1976 - Mar 1986)
Method : ML ARCH- Normal Distribution
GARCH = 16.4323 + 0.1773*Resid(-1)^2 + 0.8218*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.99992
0.183786
0.183309
1.41E-05
0.086960
0.060714
70702.72
2.113467
12.19096
0.0000
0.0346
0.0000
R-Square 0.776317 Durbin-Watson Stat 1.485692
Adjusted R-Square 0.774405 Akaike Info Criterion 9.325120
S.E. of Regression 27.17647 Schwarz – Criterion 9.441890
Log likelihood -549.8446 Hannan-Quinn Criter 9.372536
Dynamics Forecasted Evolution
RMSE MAE MAPE U
92.03096 73.19906 81.00142 0.908955
Sunspot Cycle (Mar1986 - Jun1996)
Method : ML ARCH- Normal Distribution
GARCH = 4.1310 + 0.1844*Resid(-1)^2 + 0.8332*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.937443
0.184360
0.833166
0.046924
0.083106
0.061641
19.97779
2.218375
13.51636
0.0000
0.0265
0.0000
R-Square 0.824317 Durbin-Watson Stat 1.593557
Adjusted R-Square 0.822877 Akaike Info Criterion 9.106281
S.E. of Regression 24.62191 Schwarz – Criterion 9.220002
Log likelihood -559.5894 Hannan-Quinn Criter 9.152477
Dynamics Forecasted Evolution
RMSE MAE MAPE U
84.07413 63.14946 90.53969 0.901397
Sunspot Cycle (Jun1996 - Jan 2008)
Method : ML ARCH- Normal Distribution
GARCH = 13.5042 + 0.2016*Resid(-1)^2 + 0.7890*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.915932
0.201586
0.789028
0.048648
0.082006
0.074668
18.81267
2.458183
10.56714
0.0000
0.0140
0.0000
R-Square 0.743622 Durbin-Watson Stat 1.666736
Adjusted R-Square 0.741764 Akaike Info Criterion 8.801897
222
S.E. of Regression 21.68285 Schwarz – Criterion 8.906956
Log likelihood -611.1328 Hannan-Quinn Criter 8.844590
Dynamics Forecasted Evolution
RMSE MAE MAPE U
60.54590 46.23687 105.7920 0.925559
Sunspot Cycle (Aug1755 - Jan 2008)
Method : ML ARCH- Normal Distribution
GARCH = 4.8151 + 0.2384*Resid(-1)^2 + 0.7817*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
Resid(-1)^2
GARCH(-1)
0.907941
0.238439
0.781679
0.007499
0.015721
0.011738
121.0810
15.16704
66.59540
0.0000
0.0000
0.0000
R-Square 0.794168 Durbin-Watson Stat 1.483204
Adjusted R-Square 0.794100 Akaike Info Criterion 8.464186
S.E. of Regression 20.23097 Schwarz – Criterion 8.474084
Log likelihood -12818.2 Hannan-Quinn Criter 8.467725
Dynamics Forecasted Evolution
RMSE MAE MAPE U
52.01005 37.04478 238.9409 0.612406
Table 1.6: Normality test for AR (p)-GARCH (1, 1) model of sunspot cycles
Duration AR-GARCH Mean Median Std.D Skewness Kurtosis Jur-Bera
Aug 1755 - Mar 1766 AR(2)-G(1,1) 0.142 0.073 0.992 0.3053 3.1963 2.1936
Mar 1766 - Aug 1775 AR(2)-G(1,1) 0.152 0.196 0.993 -0.481 3.0194 0.0450
Aug 1775 - Jun 1784 AR(2)-G(1,1) 0.177 0.012 0.985 1.1670 7.1835 102.31
Jun 1784 - Jun 1798 AR(2)-G(1,1) -0.04 -0.115 1.025 -0.1061 4.0362 7.8780
Jun 1798 - Sep 1810 AR(2)-G(1,1) 0.179 0.051 0.960 0.4299 5.0149 30.269
Sep1810 - Dec 1823 AR(2)-G(1,1) 0.266 0.032 0.964 0.9403 4.1162 31.8829
Dec1823 - Oct 1833 AR(3)-G(1,1) 0.102 0.061 0.997 0.4802 3.5677 6.1715
Oct1833 - Sep 1843 AR(2)-G(1,1) 0.044 0.014 1.017 0.3433 2.7908 2.5764
Sep1843 - Mar 1855 AR(2)-G(1,1) 0.178 0.073 0.985 0.882 5.0451 42.226
Mar1855 - Feb 1867 AR(2)-G(1,1) 0.120 0.020 0.993 0.4680 3.292 5.5845
Feb1867 - Sep 1878 AR(2)-G(1,1) 0.146 0.095 0.990 0.4649 3.9492 10.299
Sep1878 - Jun 1890 AR(2)-G(1,1) 0.226 0.083 0.974 0.5547 3.2357 7.6100
Jun1890 - Sep 1902 AR(2)-G(1,1) 0.149 0.149 0.990 0.4791 4.1853 14.327
Sep 1902 - Dec 1913 AR(2)-G(1,1) 0.243 0.169 1.010 0.2638 3.1268 1.6680
Dec1913 - May 1923 AR(3)-G(1,1) 0.102 0.061 0.997 0.4800 3.5680 6.1715
May 1923 - Sep 1933 AR(2)-G(1,1) 0.169 0.078 0.986 0.034 2.8421 0.1552
Sep 1933 - Jan 1944 AR(3)-G(1,1) 0.227 0.119 0.973 0.1460 2.6545 1.0660
Jan 1944 - Feb 1954 AR(2)-G(1,1) 0.145 0.106 0.988 0.8278 4.5516 26.171
Feb 1954 - Oct 1964 AR(2)-G(1,1) 0.198 0.104 0.981 0.1059 3.1181 0.3159
Oct 1964 - May 1976 AR(2)-G(1,1) 0.131 -0.023 0.990 0.5212 3.4326 7.3407
May1976 - Mar 1986 AR(2)-G(1,1) 0.143 0.053 0.989 0.2668 3.5184 2.7442
Mar1986 - Jun1996 AR(2)-G(1,1) 0.129 0.060 1.001 0.6264 4.1593 15.0531
Jun1996 - Jan 2008 AR(2)-G(1,1) 0.151 -0.013 0.988 0.6279 3.8123 13.046
Aug1755 - Jan 2008 AR(2)-G(1,1) 0.135 0.050 0.991 0.5309 4.123 301.6
223
TABLE 1.7: ARMA (p,q) - GRACH(1, 1) of sunspots cycles
Dynamics Forecasted Evolution
RMSE MAE MAPE U
32.67503 25.67541 56.50119 0.453746
Sunspot Cycle (Mar 1766 - Aug 1775)
Method : ML ARCH- Normal Distribution
GARCH = 13.5351 + 0.1668*Resid(-1)^2 + 0.8192*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(5)
MA(1)
GARCH(-1)
0.533846
0.536496
0.819233
0.095292
0.101409
0.081786
5.602184
5.290412
10.01677
0.0000
0.0000
0.0000
R-Square 0.557460 Durbin-Watson Stat 1.516418
Adjusted R-Square 0.549487 Akaike Info Criterion 9.266430
S.E. of Regression 26.31767 Schwarz – Criterion 9.410440
Log likelihood -522.1865 Hannan-Quinn Criter 9.324875
Dynamics Forecasted Evolution
RMSE MAE MAPE U
43.75567 33.31524 124.7012 0.406759
Sunspot Cycle (Aug 1775 - Jun 1784)
Method : ML ARCH- Normal Distribution
GARCH = 4.6145 + 0.2983*Resid(-1)^2 + 0.7611*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(1)
MA(1)
GARCH(-1)
0.971002
-0.653490
0.761148
0.012483
0.071574
0.080710
77.78319
-9.130225
9.430652
0.0000
0.0000
0.0000
R-Square 0.827458 Durbin-Watson Stat 1.430916
Adjusted R-Square 0.824172 Akaike Info Criterion 8.610944
S.E. of Regression 21.85616 Schwarz – Criterion 8.760823
Log likelihood -454.6855 Hannan-Quinn Criter 8.671730
Sunspot Cycle (Aug 1755 - Mar 1766)
Method : ML ARCH- Normal Distribution
GARCH = 13.1418 + 0.1512*Resid(-1)^2 + 0.7964*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
MA(2)
GARCH(-1)
0.942377
-0.344875
0.796377
0.031142
0.120105
0.092550
30.26041
-2.871453
8.604828
0.0000
0.0041
0.0000
R-Square 0.658394 Durbin-Watson Stat 1.449450
Adjusted R-Square 0.652883 Akaike Info Criterion 8.158435
S.E. of Regression 14.36707 Schwarz – Criterion 8.292124
Log likelihood -516.1398 Hannan-Quinn Criter 8.212754
224
Dynamics Forecasted Evolution
RMSE MAE MAPE U
82.97762 64.99252 88.91661 0.907096
Sunspot Cycle (Jun 1784 - Jun 1798)
Method : ML ARCH- Normal Distribution
GARCH = 13.0234 + 0.3321*Resid(-1)^2 + 0.6463*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
MA(2)
GARCH(-1)
0.963098
-0.460583
0.646251
0.011744
0.087155
0.108109
82.00897
-5.284628
5.975009
0.0000
0.0000
0.0000
R-Square 0.871309 Durbin-Watson Stat 1.297425
Adjusted R-Square 0.869758 Akaike Info Criterion 8.038823
S.E. of Regression 16.39826 Schwarz – Criterion 8.149944
Log likelihood -673.2805 Hannan-Quinn Criter 8.083918
Dynamics Forecasted Evolution
RMSE MAE MAPE U
69.78911 53.91451 82.29495 0.841223
Sunspot Cycle (Jun 1798 - Sep 1810)
Method : ML ARCH- Normal Distribution
GARCH = 4.1899 + 0.2784*Resid(-1)^2 + 0.6817*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(3)
MA(3)
GARCH(-1)
0.935858
-0.283606
0.68167
0.031534
0.098061
0.101790
29.67735
-2.892135
6.696870
0.0000
0.0038
0.0000
R-Square 0.767958 Durbin-Watson Stat 1.007526
Adjusted R-Square 0.764758 Akaike Info Criterion 7.075953
S.E. of Regression 9.074554 Schwarz – Criterion 7.197462
Log likelihood -517.6205 Hannan-Quinn Criter 7.125322
Dynamics Forecasted Evolution
RMSE MAE MAPE U
28.93611 22.58619 85.55745 0.922685
Sunspot Cycle (Sep1810 - Dec 1823)
Method : ML ARCH- Normal Distribution
GARCH = 1.5456 + 0.1084*Resid(-1)^2 + 0.8929*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(3)
MA(3)
GARCH(-1)
0.948818
-0.504034
0.892899
0.037621
0.088419
0.018509
25.22040
-5.700533
48.24127
0.0000
0.0000
0.0000
R-Square 0.557202 Durbin-Watson Stat 1.471146
Adjusted R-Square 0.551562 Akaike Info Criterion 7.372483
225
S.E. of Regression 12.04143 Schwarz – Criterion 7.487802
Log likelihood -583.7966 Hannan-Quinn Criter 7.419310
Dynamics Forecasted Evolution
RMSE MAE MAPE U
25.17780 17.81173 79.46162 0.972678
Sunspot Cycle (Dec1823 - Oct 1833)
Method : ML ARCH- Normal Distribution
GARCH = 97.4526 + 0.3116*Resid(-1)^2 + 0.3843*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(3)
MA(3)
GARCH(-1)
1.000000
-0.492451
0.744837
0.0145516
0.104489
0.206923
68.89037
-4.712929
3.599420
0.0000
0.0000
0.0003
R-Square 0.571467 Durbin-Watson Stat 1.197600
Adjusted R-Square 0.567804 Akaike Info Criterion 8.546026
S.E. of Regression 17.09971 Schwarz – Criterion 8.662796
Log likelihood -503.4886 Hannan-Quinn Criter 8.593443
Dynamics Forecasted Evolution
RMSE MAE MAPE U
35.06270 27.96897 113.0161 0.486173
Sunspot Cycle (Oct1833 - Sep 1843)
Method : ML ARCH- Normal Distribution
GARCH = 18.1207 + 0.4027*Resid(-1)^2 + 0.6657*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(3)
MA(2)
GARCH(-1)
0.717264
0.457177
0.665718
0.074861
0.102429
0.060208
9.604436
4.463361
7.722201
0.0000
0.0000
0.0000
R-Square 0.624065 Durbin-Watson Stat 0.974799
Adjusted R-Square 0.617639 Akaike Info Criterion 9.356377
S.E. of Regression 29.87390 Schwarz – Criterion 9.495752
Log likelihood -555.3826 Hannan-Quinn Criter 9.412978
Dynamics Forecasted Evolution
RMSE MAE MAPE U
70.49102 53.86759 79.03446 0.727303
Sunspot Cycle (Sep1843 - Mar 1855)
Method : ML ARCH- Normal Distribution
GARCH = 8.0428 + 0.1739*Resid(-1)^2 + 0.8282*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(4)
MA(2)
GARCH(-1)
0.820607
0.713805
0.828213
0.067158
0.084455
0.059027
12.21903
8.451875
14.03098
0.0000
0.0000
0.0000
226
R-Square 0.660677 Durbin-Watson Stat 1.434357
Adjusted R-Square 0.655573 Akaike Info Criterion 8.765867
S.E. of Regression 22.54270 Schwarz – Criterion 8.892535
Log likelihood -603.2278 Hannan-Quinn Criter 8.817342
Dynamics Forecasted Evolution
RMSE MAE MAPE U
55.68955 42.46238 61.37514 0.609577
Sunspot Cycle (Mar1855 - Feb 1867)
Method : ML ARCH- Normal Distribution
GARCH = 5.2998 + 0.1693*Resid(-1)^2 + 0.8236*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(4)
MA(4)
GARCH(-1)
0.935145
-0.261549
0.823638
0.027516
0.096453
0.085286
33.90542
-2.711680
4.657320
0.0000
0.0067
0.0000
R-Square 0.765643 Durbin-Watson Stat 1.103142
Adjusted R-Square 0.762319 Akaike Info Criterion 8.205885
S.E. of Regression 15.88000 Schwarz – Criterion 8.329627
Log likelihood -584.8237 Hannan-Quinn Criter 8.256166
Dynamics Forecasted Evolution
RMSE MAE MAPE U
50.33423 40.31416 150.4605 0.766811
Sunspot Cycle (Feb1867 - Sep 1878)
Method : ML ARCH- Normal Distribution
GARCH = 5.5674 + 0.2023*Resid(-1)^2 + 0.8006*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(5)
MA(1)
GARCH(-1)
0.810985
0.372377
0.800595
0.040918
0.097022
0.075716
19.81984
2.105510
10.57360
0.0000
0.0001
0.0000
R-Square 0.786394 Durbin-Watson Stat 1.271423
Adjusted R-Square 0.783275 Akaike Info Criterion 8.551210
S.E. of Regression 21.79505 Schwarz – Criterion 8.677281
Log likelihood -592.5847 Hannan-Quinn Criter 8.602442
Dynamics Forecasted Evolution
RMSE MAE MAPE U
68.21752 50.46720 163.9832 0.874876
Sunspot Cycle (Sep1878 - Jun 1890)
Method : ML ARCH- Normal Distribution
GARCH = 2.0310 + 0.3175*Resid(-1)^2 + 0.7260*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
227
AR(2)
MA(2)
GARCH(-1)
0.977805
-0.480838
0.726032
0.020228
0.086600
0.087548
48.33850
-5.52371
8.292922
0.0000
0.0000
0.0000
R-Square 0.721790 Durbin-Watson Stat 1.329812
Adjusted R-Square 0.717787 Akaike Info Criterion 7.826804
S.E. of Regression 14.33897 Schwarz – Criterion 7.951698
Log likelihood -549.7031 Hannan-Quinn Criter 7.877555
Dynamics Forecasted Evolution
RMSE MAE MAPE U
40.16201 30.27203 105.1265 0.876150
Sunspot Cycle (Jun1890 - Sep 1902)
Method : ML ARCH- Normal Distribution
GARCH = 3.4242 + 0.2769*Resid(-1)^2 + 0.7417*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
MA(2)
GARCH(-1)
0.960991
-0.535407
0.741708
0.013212
0.090477
0.072713
72.73458
-5.917594
10.20053
0.0000
0.0000
0.0000
R-Square 0.801219 Durbin-Watson Stat 1.326042
Adjusted R-Square 0.798477 Akaike Info Criterion 7.805326
S.E. of Regression 13.74870 Schwarz – Criterion 7.926835
Log likelihood -571.5942 Hannan-Quinn Criter 7.854695
Dynamics Forecasted Evolution
RMSE MAE MAPE U
44.99864 33.86295 108.7426 0.836094
Sunspot Cycle (Sep 1902 - Dec 1913)
Method : ML ARCH- Normal Distribution
GARCH = 0.3897 + 0.2509*Resid(-1)^2 + 0.7845*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
MA(2)
GARCH(-1)
0.986754
-0.552862
0.784486
0.024632
0.080533
0.051384
40.05999
-6.244711
15.26708
0.0000
0.00000
.0000
R-Square 0.590129 Durbin-Watson Stat 1.287636
Adjusted R-Square 0.583965 Akaike Info Criterion 8.016758
S.E. of Regression 16.98195 Schwarz – Criterion 8.145258
Log likelihood -539.1395 Hannan-Quinn Criter 8.069877
Dynamics Forecasted Evolution
RMSE MAE MAPE U
36.86236 27.61363 117.5639 0.744690
Sunspot Cycle (Dec1913 - May 1923)
Method : ML ARCH- Normal Distribution
GARCH = 20.9228 + 0.5071*Resid(-1)^2 + 0.5522*GARCH(-1)
228
Variable Coefficient Std. Error z-statistics Prob
AR(3)
MA(3)
GARCH(-1)
0.913874
-0.467862
0.552189
0.023667
0.115770
0.119920
38.61419
-4.041320
4.604632
0.0000
0.0000
0.0001
R-Square 0.578780 Durbin-Watson Stat 0.802072
Adjusted R-Square 0.571190 Akaike Info Criterion 8.751704
S.E. of Regression 22.24516 Schwarz – Criterion 8.895714
Log likelihood -492.8471 Hannan-Quinn Criter 8.810150
Dynamics Forecasted Evolution
RMSE MAE MAPE U
53.34212 41.91097 79.70615 0.826760
Sunspot Cycle (May 1923 - Sep 1933)
Method : ML ARCH- Normal Distribution
GARCH = 9.9187 + 0.3086*Resid(-1)^2 + 0.6774*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
MA(2)
GARCH(-1)
0.958244
-0.459230
0.677347
0.020393
0.111471
0.123939
46.98964
-4.119713
5.465159
0.0000
0.0000
0.0000
R-Square 0.726700 Durbin-Watson Stat 1.002388
Adjusted R-Square 0.722219 Akaike Info Criterion 8.072810
S.E. of Regression 15.14237 Schwarz – Criterion 8.208569
Log likelihood -498.5507 Hannan-Quinn Criter 8.127962
Dynamics Forecasted Evolution
RMSE MAE MAPE U
44.927819 35.31193 116.2730 0.814867
Sunspot Cycle (Sep 1933 - Jan 1944)
Method : ML ARCH- Normal Distribution
GARCH = 7.7621+ 0.1524*Resid(-1)^2 + 0.8407*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
MA(2)
GARCH(-1)
0.974799
-0.419687
0.840734
0.023533
0.100495
0.106100
41.42327
-4.176213
7.923969
0.0000
0.0000
0.0000
R-Square 0.755215 Durbin-Watson Stat 1.176361
Adjusted R-Square 0.751202 Akaike Info Criterion 8.664634
S.E. of Regression 20.21283 Schwarz – Criterion 8.800393
Log likelihood -535.5396 Hannan-Quinn Criter 8.719786
Dynamics Forecasted Evolution
RMSE MAE MAPE U
67.16046 54.07226 96.16188 0.888186
Sunspot Cycle (Jan 1944 - Feb 1954)
229
Method : ML ARCH- Normal Distribution
GARCH = 9.4407 + 0.2034*Resid(-1)^2 + 0.8127*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(5)
MA(3)
GARCH(-1)
0.854521
0.346542
0.812699
0.052139
0.098825
0.072919
16.38938
3.506625
11.14520
0.0000
0.0005
0.0000
R-Square 0.658214 Durbin-Watson Stat 0.962410
Adjusted R-Square 0.652470 Akaike Info Criterion 9.418094
S.E. of Regression 32.25406 Schwarz – Criterion 9.555997
Log likelihood -568.5037 Hannan-Quinn Criter 9.474106
Dynamics Forecasted Evolution
RMSE MAE MAPE U
89.57113 72.19864 122.4058 0.895654
Sunspot Cycle (Feb 1954 - Oct 1964)
Method : ML ARCH- Normal Distribution
GARCH = 3.7640 + 0.3486*Resid(-1)^2 + 0.6959*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(5)
MA(3)
GARCH(-1)
0.854124
0.497641
0.695872
0.035637
0.108373
0.102716
23.96715
4.591918
6.774746
0.0000
0.0000
0.0000
R-Square 0.844706 Durbin-Watson Stat 0.794019
Adjusted R-Square 0.842241 Akaike Info Criterion 9.055507
S.E. of Regression 28.40691 Schwarz – Criterion 9.188522
Log likelihood -578.0802 Hannan-Quinn Criter 9.109554
Dynamics Forecasted Evolution
RMSE MAE MAPE U
113.2087 88.62173 87.83657 0.939852
Sunspot Cycle (Oct 1964 - May 1976)
Method : ML ARCH- Normal Distribution
GARCH = 618.4769 + 0.1576*Resid(-1)^2 – 1.0532*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
MA(2)
GARCH(-1)
0.981676
-0.364651
0.902341
0.020894
0.100502
0.073421
46.98375
-3.628281
12.28997
0.0000
0.0003
0.0000
R-Square 0.816936 Durbin-Watson Stat 1.324666
Adjusted R-Square 0.814264 Akaike Info Criterion 8.389168
S.E. of Regression 16.07089 Schwarz – Criterion 8.515238
Log likelihood -581.2417 Hannan-Quinn Criter 8.440399
Dynamics Forecasted Evolution
RMSE MAE MAPE U
63.56454 51.95468 74.98162 0.333498
230
Sunspot Cycle (May1976 - Mar 1986)
Method : ML ARCH- Normal Distribution
GARCH = 14.7176+ 0.1767*Resid(-1)^2 + 0.8190*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
MA(2)
GARCH(-1)
0.969742
-0.420841
0.818978
0.20513
0.116761
0.071133
47.27442
-3.604301
11.51331
0.0000
0.0003
0.0000
R-Square 0.814647 Durbin-Watson Stat 1.138317
Adjusted R-Square 0.811452 Akaike Info Criterion 9.215569
S.E. of Regression 24.84504 Schwarz – Criterion 9.295692
Log likelihood -538.7563 Hannan-Quinn Criter 9.212468
Dynamics Forecasted Evolution
RMSE MAE MAPE U
91.74100 72.87236 80.66910 0.800143
Sunspot Cycle (Mar1986 - Jun1996)
Method : ML ARCH- Normal Distribution
GARCH = 11.4294 + 0.1612*Resid(-1)^2 + 0.8380*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(6)
MA(1)
GARCH(-1)
0.834506
0.380998
0.838039
0.047834
0.108274
0.045692
17.44584
3.518828
18.34105
0.0000
0.0000
0.0000
R-Square 0.792093 Durbin-Watson Stat 1.463125
Adjusted R-Square 0.788657 Akaike Info Criterion 9.169600
S.E. of Regression 26.89540 Schwarz – Criterion 9.306065
Log likelihood -562.5152 Hannan-Quinn Criter 9.225035
Dynamics Forecasted Evolution
RMSE MAE MAPE U
87.08569 66.45792 75.91525 0.772314
Sunspot Cycle (Jun1996 - Jan 2008)
Method : ML ARCH- Normal Distribution
GARCH = 10.1606 + 0.1681*Resid(-1)^2 + 0.8193*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
MA(2)
GARCH(-1)
0.969016
-0.498636
0.819252
0.021416
0.110312
0.059092
45.24690
-4.520227
13.86391
0.0000
0.0000
0.0000
R-Square 0.803463 Durbin-Watson Stat 1.279801
Adjusted R-Square 0.800593 Akaike Info Criterion 8.561970
S.E. of Regression 19.05363 Schwarz – Criterion 8.688040
Log likelihood -593.3379 Hannan-Quinn Criter 8.613201
Dynamics Forecasted Evolution
231
RMSE MAE MAPE U
65.31769 50.61469 93.96324 0.802811
Sunspot Cycle (Aug1755 - Jan 2008)
Method : ML ARCH- Normal Distribution
GARCH = 4.2183 + 0.2292*Resid(-1)^2 + 0.7863*GARCH(-1)
Variable Coefficient Std. Error z-statistics Prob
AR(2)
MA(2)
GARCH(-1)
0.895023
0.083322
0.780889
0.008475
0.018512
0.011874
105.6073
4.500889
65.76430
0.0000
0.0000
0.0000
R-Square 0.797498 Durbin-Watson Stat 1.474460
Adjusted R-Square 0.794662 Akaike Info Criterion 8.458505
S.E. of Regression 20.21142 Schwarz – Criterion 8.470418
Log likelihood -12808.63 Hannan-Quinn Criter 8.462788
Dynamics Forecasted Evolution
RMSE MAE MAPE U
54.37098 38.55961 205.0083 0.778433
Table 1.8: Test of Normality ARMA (p, q)-GARCH (1,1) process of Sunspot Cycles
cycles ARMA-GARCH Mean Median Std.D Skewness Kurtosis Jur-Bera
1 ARMA(2,2)-G(1,1) 0.143 -0.202 0.991 0.3875 3.4860 4.4639
2 ARMA(5,1)-G(1,1) 0.167 0.107 0.991 0.256 2.6522 1.913
3 ARMA(1,1)-G(1,1) 0.197 0.045 0.983 0.9322 6.1950 61.007
4 ARMA(2,2)-G(1,1) 0.140 0.121 0.991 -0.283 4.226 12.839
5 ARMA(3,3)-G(1,1) 0.153 0.061 0.988 0.1280 3.501 1.9499
6 ARMA(3,3)-G(1,1) 0.190 -0.027 0.983 1.1371 4.882 58.097
7 ARMA(3,3)-G(1,1) 0.097 0.018 1.009 0.4005 2.8365 3.3141
8 ARMA(3,2)-G(1,1) 0.305 0.178 0.954 0.6251 3.3824 8.5454
9 ARMA(4,2)-G(1,1) 0.181 0.093 0.984 0.8470 4.5254 30.096
10 ARMA(4,4)-G(1,1) 0.145 0.045 0.992 0.089 2.8840 0.2716
11 ARMA(5,1)-G(1,1) 0.249 0.322 0.968 0.046 2.9809 0.2716
12 ARMA(2,2)-G(1,1) 0.130 -0.003 0.994 0.6397 3.0275 9.6901
13 ARMA(2,2)-G(1,1) 0.140 0.0615 0.992 0.4077 3.8820 808934
14 ARMA(2,2)-G(1,1) 0.030 -0.112 0.996 0.9258 4.1931 27.4940
15 ARMA(3,3)-G(1,1) 0.238 0.098 0.973 0.4546 3.1964 4.1101
16 ARMA(2,2)-G(1,1) 0.165 0.148 0.987 0.1321 2.6787 0.9011
17 ARMA(2,2)-G(1,1) 0.167 0.095 0.985 0.2786 3.2299 1.8924
18 ARMA(5,3)-G(1,1) 0.258 0.100 0.965 0.2909 2.6303 2.4153
19 ARMA(5,3)-G(1,1) 0.258 0.297 0.966 -0.1425 2.4166 2.2661
20 ARMA(2,2)-G(1,1) 0.125 0.017 0.996 0.4562 3.1727 5.0298
21 ARMA(2,2)-G(1,1) 0.142 0.129 0.989 0.0006 3.7090 2.4922
22 ARMA(6,1)-G(1,1) 0.206 0.108 0.977 0.1528 3.6409 2.6051
23 ARMA(2,2)-G(1,1) 0.118 -0.050 0.993 0.6868 4.5192 24.4696
1-23 ARMA(2,2)-G(1,1) 0.142 0.318 0.990 0.5121 3.8258 218.54
Table 1.9: Root Mean Square Error (RMSE) of ARMA, AR-GARCH and ARMA-GARCH of sunspot
cycles
232
Duration ARMA (p, q) AR(p)–GARCH (1,1) ARMA(p,q) )–GARCH (1,1)
Aug 1755 - Mar 1766 (6,5) 22.96202 AR(2) 27.61330 (6,5) 32.67503
Mar 1766 - Aug 1775 (5,3) 37.92649 AR(2) 48.9534 (5,1) 43.75567
Aug 1775 - Jun 1784 (6,4) 56.72805 AR(2) 81.25105 (1,1) 82.97762
Jun 1784 - Jun 1798 (5,4) 54.15024 AR(2) 69.70664 (2,2) 69.78911
Jun 1798 - Sep 1810 (5,3) 19.61195 AR(2) 29.84910 (3,3) 28.93611
Sep1810 - Dec 1823 (3,4) 17.58213 AR(2) 23.87355 (3,3) 25.17780
Dec1823 - Oct 1833 (5,3) 24.84239 AR(3) 26.97423 (3,3) 35.06270
Oct1833 - Sep 1843 (4,6) 47.89669 AR(2) 76.51305 (3,2) 70.49102
Sep1843 - Mar 1855 (4,3) 36.85102 AR(2) 54.20801 (4,2) 55.68955
Mar1855 - Feb 1867 (6,4) 34.05871 AR(2) 46.84207 (4,4) 50.33423
Feb1867 - Sep 1878 (6,4) 51.34360 AR(2) 66.48149 (5,1) 68.21752
Sep1878 - Jun 1890 (5,3) 26.94070 AR(2) 39.33113 (2,2) 40.16201
Jun1890 - Sep 1902 (5,4) 33.19800 AR(2) 41.43307 (2,2) 44.99864
Sep 1902 - Dec 1913 (4,6) 26.33349 AR(2) 37.13987 (2,2) 36.86236
Dec1913 - May 1923 (5,4) 33.40646 AR(3) 50.69620 (3,3) 53.34212
May 1923 - Sep 1933 (6,5) 28.41330 AR(2) 41.61186 (2,2) 44.92782
Sep 1933 - Jan 1944 (6,2) 40.50563 AR(3) 66.90030 (2,2) 67.16046
Jan 1944 - Feb 1954 (6,4) 58.17556 AR(2) 88.43862 (5,3) 89.57113
Feb 1954 - Oct 1964 (4,3) 81.95903 AR(2) 110.3128 (5,3) 113.2087
Oct 1964 - May 1976 (4,3) 39.00213 AR(2) 59.36716 (2,2) 63.56454
May1976 - Mar 1986 (5,3) 61.20026 AR(2) 92.03096 (2,2) 91.74100
Mar1986 - Jun1996 (5,6) 65.05685 AR(2) 84.07413 (6,1) 87.08569
Jun1996 - Jan 2008 (6,4) 45.06462 AR(2) 60.54590 (2,2) 65.31769
Aug1755 - Jan 2008 (6,4) 44.54175 AR(2) 52.01005 (2,2) 54.37098
Table 1.10: Mean Absolute Error (MAE) of ARMA, AR-GARCH and ARMA-GARCH of sunspot
cycles
Duration ARMA (p, q) AR(p)–GARCH (1,1) ARMA(p,q) )–GARCH (1,1)
Aug 1755 - Mar 1766 (4,5) 18.43576 AR(2) 21.4119 (6,5) 25.67541
Mar 1766 - Aug 1775 (5,3) 29.55789 AR(2) 37.4807 (5,1) 33.31524
Aug 1775 - Jun 1784 (6,4) 43.64520 AR(2) 62.90952 (1,1) 64.99252
Jun 1784 - Jun 1798 (5,6) 41.59299 AR(2) 53.90236 (2,2) 53.91451
Jun 1798 - Sep 1810 (5,3) 16.77907 AR(2) 23.36523 (3,3) 22.58619
Sep1810 - Dec 1823 (6,4) 13.25282 AR(2) 16.61689 (3,3) 22.58619
Dec1823 - Oct 1833 (5,3) 20.57404 AR(3) 22.29147 (3,3) 27.96897
Oct1833 - Sep 1843 (4,6) 37.64070 AR(2) 59.98412 (3,2) 53.86759
Sep1843 - Mar 1855 (6,4) 26.67266 AR(2) 41.01302 (4,2) 42.46238
Mar1855 - Feb 1867 (6,4) 26.32748 AR(2) 36.86088 (4,4) 40.31416
Feb1867 - Sep 1878 (6,4) 40.82203 AR(2) 48.35055 (5,1) 50.46720
Sep1878 - Jun 1890 (5,3) 22.33445 AR(2) 29.41939 (2,2) 30.27203
Jun1890 - Sep 1902 (5,4) 27.39173 AR(2) 30.79801 (2,2) 33.86295
Sep 1902 - Dec 1913 (4,6) 21.89815 AR(2) 28.22192 (2,2) 27.61363
Dec1913 - May 1923 (5,3) 26.04595 AR(3) 39.37540 (3,3) 41.91097
May 1923 - Sep 1933 (6,5) 23.46526 AR(2) 32.01947 (2,2) 35.31193
Sep 1933 - Jan 1944 (6,2) 31.48136 AR(3) 32.01947 (2,2) 54.07226
Jan 1944 - Feb 1954 (5,3) 44.02761 AR(2) 70.49887 (5,3) 72.19864
Feb 1954 - Oct 1964 (4,3) 64.00049 AR(2) 84.95945 (5,3) 88.62173
Oct 1964 - May 1976 (4,3) 32.19808 AR(2) 47.77713 (2,2) 51.95468
May1976 - Mar 1986 (5,3) 50.64551 AR(2) 73.19906 (2,2) 72.87236
233
Mar1986 - Jun1996 (4,5) 50.90771 AR(2) 63.14946 (6,1) 66.45792
Jun1996 - Jan 2008 (3,4) 35.46916 AR(2) 46.23687 (2,2) 50.61469
Aug1755 - Jan 2008 (6,4) 35.30753 AR(2) 37.04478 (2,2) 38.55961
Table1.11: Mean Absolute Percentage Error (MAPE) of ARMA, AR-GARCH and ARMA-GARCH of
sunspot cycles
Duration ARMA (p, q) AR(p)–GARCH (1,1) ARMA(p,q) )–GARCH (1,1)
Aug 1755 - Mar 1766 (4,4) 55.73550 AR(2) 56.7095 (6,5) 56.50119
Mar 1766 - Aug 1775 (2,2) 145.5506 AR(2) 115.7679 (5,1) 124.7012
Aug 1775 - Jun 1784 (5,5) 102.4356 AR(2) 80.41010 (1,1) 88.91661
Jun 1784 - Jun 1798 (3,3) 141.1219 AR(2) 84.66272 (2,2) 82.29495
Jun 1798 - Sep 1810 (5,3) 222.5072 AR(2) 87.63108 (3,3) 85.55745
Sep1810 - Dec 1823 (6,6) 215.0192 AR(2) 81.56241 (3,3) 79.46162
Dec1823 - Oct 1833 (5,5) 125.1342 AR(3) 148.6877 (3,3) 113.0161
Oct1833 - Sep 1843 (3,3) 101.5025 AR(2) 82.81522 (3,2) 79.03446
Sep1843 - Mar 1855 (3,3) 55.33579 AR(2) 59.41300 (4,2) 61.37514
Mar1855 - Feb 1867 (4,4) 149.2235 AR(2) 163.9772 (4,4) 150.4605
Feb1867 - Sep 1878 (3,3) 619.2955 AR(2) 184.8052 (5,1) 163.9832
Sep1878 - Jun 1890 (4,4) 317.3174 AR(2) 110.6451 (2,2) 105.1265
Jun1890 - Sep 1902 (4,4) 351.6125 AR(2) 168.5865 (2,2) 108.7426
Sep 1902 - Dec 1913 (5,5) 469.8335 AR(2) 163.2702 (2,2) 117.5639
Dec1913 - May 1923 (5,5) 124.6412 AR(3) 80.53190 (3,3) 79.70615
May 1923 - Sep 1933 (4,4) 225.7652 AR(2) 136.7743 (2,2) 116.2730
Sep 1933 - Jan 1944 (4,4) 79.25677 AR(3) 94.93236 (2,2) 96.16188
Jan 1944 - Feb 1954 (4,4) 374.3422 AR(2) 127.5823 (5,3) 122.4058
Feb 1954 - Oct 1964 (5,4) 125.9374 AR(2) 101.2019 (5,3) 87.83657
Oct 1964 - May 1976 (3,3) 78.84340 AR(2) 71.04261 (2,2) 74.98162
May1976 - Mar 1986 (3,3) 99.64765 AR(2) 81.00142 (2,2) 80.66910
Mar1986 - Jun1996 (6,5) 120.4492 AR(2) 90.53969 (6,1) 75.91525
Jun1996 - Jan 2008 (5,4) 178.4829 AR(2) 105.7920 (2,2) 93.96324
Aug1755 - Jan 2008 (5,5) 464.3263 AR(2) 238.9409 (2,2) 205.0083
Table 1.12: Theil’s U-Statistics (U test) of ARMA, AR-GARCH and ARMA-GARCH of sunspot cycles
Duration ARMA (p, q) AR(p)–GARCH (1,1) ARMA(p,q) )–GARCH (1,1)
Aug 1755 - Mar 1766 (6,5) 0.251178 AR(2) 0.34824 (6,5) 0.453746
Mar 1766 - Aug 1775 (5,3) 0.309054 AR(2) 0.501998 (5,1) 0.406759
Aug 1775 - Jun 1784 (6,4) 0.411658 AR(2) 0.859162 (1,1) 0.907096
Jun 1784 - Jun 1798 (5,6) 0.465585 AR(2) 0.837476 (2,2) 0.841223
Jun 1798 - Sep 1810 (6,4) 0.416733 AR(2) 0.999864 (3,3) 0.922685
Sep1810 - Dec 1823 (3,4) 0.418201 AR(2) 0.867214 (3,3) 0.972678
Dec1823 - Oct 1833 (5,3) 0.294667 AR(3) 0.342102 (3,3) 0.486173
Oct1833 - Sep 1843 (4,6) 0.353636 AR(2) 0.863572 (3,2) 0.727303
Sep1843 - Mar 1855 (4,3) 0.296860 AR(2) 0.590749 (4,2) 0.609577
Mar1855 - Feb 1867 (6,4) 0.318617 AR(2) 0.677314 (4,4) 0.766811
Feb1867 - Sep 1878 (6,4) 0.467277 AR(2) 0.845125 (5,1) 0.874876
Sep1878 - Jun 1890 (5,3) 0.384873 AR(2) 0.841212 (2,2) 0.876150
Jun1890 - Sep 1902 (4,3) 0.429832 AR(2) 0.703145 (2,2) 0.836094
Sep 1902 - Dec 1913 (4,6) 0.363862 AR(2) 0.755452 (2,2) 0.744690
Dec1913 - May 1923 (5,3) 0.329163 AR(3) 0.747014 (3,3) 0.826760
234
May 1923 - Sep 1933 (6,5) 0.340449 AR(2) 0.701884 (2,2) 0.814867
Sep 1933 - Jan 1944 (5,3) 0.336577 AR(3) 0.874060 (2,2) 0.888186
Jan 1944 - Feb 1954 (5,3) 0.366021 AR(2) 0.627014 (5,3) 0.895654
Feb 1954 - Oct 1964 (5,3) 0.483381 AR(2) 0.910967 (5,3) 0.939852
Oct 1964 - May 1976 (4,3) 0.329886 AR(2) 0.921958 (2,2) 0.333498
May1976 - Mar 1986 (5,3) 0.375419 AR(2) 0.908955 (2,2) 0.800143
Mar1986 - Jun1996 (4,5) 0.433141 AR(2) 0.901397 (6,1) 0.772314
Jun1996 - Jan 2008 (3,4) 0.373450 AR(2) 0.925559 (2,2) 0.802811
Aug1755 - Jan 2008 (6,4) 0.370626 AR(2) 0.612406 (2,2) 0.778433
Table 2.1: The numerical relationship between different fractal dimensions, Hurst exponent,
autocorrelation coefficient and spectral exponents can be express as,
Cy
cles
Duration FDS FDA HS HA αS αA CS∇ CA∇ Persistency
1 Aug 1755 - Mar 1766 1.181 1.367 0.819 0.633 2.638 2.266 0.556 0.202 Persistent
2 Mar 1766 - Aug 1775 1.27 1.407 0.73 0.593 2.246 2.186 0.376 0.138 Persistent
3 Aug 1775 - Jun 1784 1.327 1.587 0.672 0.413 2.344 1.826 0.269 -0.114 Persistent
4 Jun 1784 - Jun 1798 1.318 1.446 0.682 0.554 2.364 2.108 0.287 0.078 Persistent
5 Jun 1798 - Sep 1810 1.099 1.380 0.901 0.620 2.802 2.240 0.744 0.181 Persistent
6 Sep1810 - Dec 1823 1.136 1.363 0.864 0.637 2.728 2.274 0.656 0.209 Persistent
7 Dec1823 - Oct 1833 1.177 1.163 0.823 0.837 2.646 2.674 0.565 0.595 Persistent
8 Oct1833 - Sep 1843 1.296 1.223 0.704 0.772 2.408 2.554 0.327 0.458 Persistent
9 Sep1843 - Mar 1855 1.275 1.470 0.725 0.530 2.45 2.060 0.366 0.042 Persistent
10 Mar1855 - Feb 1867 1.267 1.313 0.733 0.687 2.466 2.374 0.381 0.296 Persistent
11 Feb1867 - Sep 1878 1.298 1.347 0.702 0.653 2.404 2.306 0.323 0.236 Persistent
12 Sep1878 - Jun 1890 1.165 1.463 0.835 0.537 2.67 2.074 0.591 0.053 Persistent
13 Jun1890 - Sep 1902 1.252 1.376 0.748 0.624 2.496 2.248 0.410 0.188 Persistent
14 Sep 1902 - Dec 1913 1.141 1.400 0.859 0.600 2.718 2.200 0.645 0.149 Persistent
15 Dec1913 - May 1923 1.235 1.479 0.765 0.521 2.53 2.042 0.443 0.030 Persistent
16 May 1923 - Sep 1933 1.149 1.391 0.851 0.604 2.702 2.218 0.627 0.155 Persistent
17 Sep 1933 - Jan 1944 1.206 1.526 0.794 0.474 2.588 1.948 0.503 -0.035 Persistent
18 Jan 1944 - Feb 1954 1.194 1.373 0.806 0.627 2.612 2.254 0.528 0.193 Persistent
19 Feb 1954 - Oct 1964 1.266 1.410 0.734 0.590 2.468 2.180 0.383 0.133 Persistent
20 Oct 1964 - May 1976 1.193 1.355 0.807 0.645 2.614 2.290 0.530 0.223 Persistent
21 May1976 - Mar 1986 1.198 1.187 0.802 0.813 2.604 2.626 0.520 0.543 Persistent
22 Mar1986 - Jun1996 1.225 1.436 0.775 0.563 2.55 2.128 0.464 0.091 Persistent
23 Jun1996 - Jan 2008 1.213 1.451 0.787 0.549 2.574 2.098 0.489 0.070 Persistent
24 Aug1755 - Jan 2008 1.002 1.354 0.998 0.649 2.996 2.110 0.994 0.229 Persistent
Table 2.2: The numerical relationship between different Higuchi’s fractal dimensions FDH and spectral
exponents αH of sunspot cycles can be express as,
Cycles Duration FDH αH
1 Aug 1755 - Mar 1766 1.621 1.8
2 Mar 1766 - Aug 1775 1.620 1.8
3 Aug 1775 - Jun 1784 1.432 2.1
4 Jun 1784 - Jun 1798 1.454 2.1
5 Jun 1798 - Sep 1810 1.486 2.0
6 Sep1810 - Dec 1823 1.648 1.7
7 Dec1823 - Oct 1833 1.623 1.8
235
8 Oct1833 - Sep 1843 1.507 2.0
9 Sep1843 - Mar 1855 1.550 1.9
10 Mar1855 - Feb 1867 1.465 2.0
11 Feb1867 - Sep 1878 1.455 2.1
12 Sep1878 - Jun 1890 1.580 1.8
13 Jun1890 - Sep 1902 1.588 1.8
14 Sep 1902 - Dec 1913 1.679 1.6
15 Dec1913 - May 1923 1.564 1.9
16 May 1923 - Sep 1933 1.521 2.0
17 Sep 1933 - Jan 1944 1.503 2.0
18 Jan 1944 - Feb 1954 1.448 2.1
19 Feb 1954 - Oct 1964 1.354 2.3
20 Oct 1964 - May 1976 1.542 1.9
21 May1976 - Mar 1986 1.442 2.1
22 Mar1986 - Jun1996 1.402 2.2
23 Jun1996 - Jan 2008 1.481 2.0
Table 2.3: The numerical relationship between self-similar and self-affine fractal dimensions, Hurst
exponent, Probability distribution persistency of ENSO cycles
Cycles Duration FDS HES axd+ c FDA HEA Persistency
1 1866-1872 1.059 0.941 19.95x1.059 + 0.17 1.208 0.792 Persistent
2 1873-1879 1.010 0.990 84.75x1.01 + 0.03 1.090 0.910 Persistent
3 1880-1886 1.002 0.998 83.43x1.002 + 0.005 1.361 0.640 Persistent
4 1887-1893 1.006 0.994 84.08x1.006+ 0.02 1.156 0.844 Persistent
5 1894-1900 1.004 0.996 83.79x1.004+ 0.01 1.129 0.871 Persistent
6 1901-1907 1.005 0.995 83.45x1.005+ 0.01 1.171 0.829 Persistent
7 1908-1914 1.003 0.997 83.55x1.003+ 0.008 1.014 0.986 Persistent
8 1915-1921 1.007 0.993 84.37x1.007+ 0.02 1.041 0.959 Persistent
9 1922-1928 1.003 0.997 83.5x1.003+ 0.009 1.264 0.736 Persistent
10 1929-1935 1.001 0.999 83.24x1.001+ 0.004 1.314 0.686 Persistent
11 1936-1942 1.005 0.995 83.91x1.005+ 0.01 1.001 0.999 Persistent
12 1943-1949 1.002 0.998 83.29x1.002+ 0.006 1.291 0.709 Persistent
13 1950-1956 1.003 0.997 83.51x1.003+ 0.008 1.016 0.984 Persistent
14 1957-1963 1.001 0.999 83.28x1.001+ 0.004 1.160 0.840 Persistent
15 1964-1970 1.005 0.995 83.76x1.005+ 0.015 1.202 0.798 Persistent
16 1971-1977 1.006 0.994 84.09x1.006+ 0.02 1.088 0.913 Persistent
17 1978-1984 1.005 0.995 83.8x1.005+ 0.016 1.122 0.879 Persistent
18 1985-1991 1.005 0.995 83.96x1.005+ 0.017 1.113 0.887 Persistent
19 1992-1998 1.007 0.993 84.32x1.007+ 0.025 1.118 0.882 Persistent
20 1999-2005 1.001 0.999 83.19x1.001+ 0.002 1.072 0.930 Persistent
21 2006-2012 1.006 0.994 84.12x1.006+ 0.022 1.178 0.822 Persistent
1-21 1866-2012 1 1 1763.8x1+ 0.0005 1.224 0.776 Persistent
22 1981-2000 1 1 239.1x1+ 0.001 1.268 0.732 Persistent
Table 2.4: The numerical relationship between self-similar and self-affine fractal dimensions, Hurst
exponents and persistency in the same interval of sunspots and ENSO
Parameters Sunspots ENSO Persistency
Cycle Duration FDS FDA HS HA FDS FDA HS HA Persistent
1 May1976-Mar 1986 1.198 1.187 0.802 0.813 1.001 1.205 0.999 0.795 Persistent
236
2 Mar1986- Jun1996 1.225 1.436 0.775 0.563 1.002 1.080 0.998 0.920 Persistent
3 Jun1996- Jan 2008 1.213 1.451 0.787 0.549 1.002 1.105 0.998 0.895 Persistent
Table 3.1: Sunspot Cycles probability distribution
Cycles Duration MEAN Std.
Dev
Distribution ADT KST CST Parameters
1 Aug 1755 -
Mar 1766
42.001 27.925 Generalized Pareto 12.039 0.463 -- α =
β = 63.963
γ = -0.700
2 Mar 1766 -
Aug 1775
56.959 40.33 Generalized Pareto 4.4108 0.065 -- α =
β = 85.884
γ = -0.5299
3 Aug 1775 -
Jun 1784
68.504 52.54 Generalized Pareto 4.559 0.0741 -- α = −
β = 101.44
γ = -0.42722
4 Jun 1784 -
Jun 1798
60.801 43.02 Generalized Pareto 0.717 0.049 3.642 α = −
β = 95.497
γ = -0.4873
5 Jun 1798 -
Sep 1810
24.068 16.45 Generalized Pareto 2.019 0.111 7.354 α = −
β = 49.461
γ = -0.6527
6 Sep1810 -
Dec 1823
17.763 17.70 Generalized Pareto 2.269 0.122 8.269 α = −
β = 24.961
γ = -0.17961
7 Dec1823 -
Oct 1833
39.303 27.17 Generalized Pareto 12.387 0.081
--- α = −1.1533
β = 72.997
γ = -0.77102
8 Oct1833 -
Sep 1843
65.208 48.12 Generalized Pareto 4.362 0.052
--- α = −2.6559
β = 101.25
γ = -0. 48812
9 Sep1843 -
Mar 1855
59.369 40.15 Gen. Extreme Value 0.385 0.052 5.845 k = 0.03061
σ = 29.598
μ = 0. 03061
10 Mar1855 -
Feb 1867
47.813 31.34 Generalized Pareto 0.437 0.048
9.065 α = −4.037
β = 83.244
γ = -0. 66846
11 Feb1867 -
Sep 1878
54.822 42.35 Generalized Pareto 1.415 0.086
9.837 α = −6.1558
β = 77.048
γ = -0. 28971
12 Sep1878 -
Jun 1890
33.771 24.19 Generalized Pareto 1.703 0.090
16.321 α = −4.6614
β = 53.533
γ = -0. 43861
13 Jun1890 -
Sep 1902
37.789 28.99 Generalized Pareto 1.334 0.077
25.193 α = −4.8239
β = 61.465
γ = -0. 45076
14 Sep 1902 -
Dec 1913
32.844 25.44 Generalized Pareto 12.826 0.083
-- α = −4.907
β = 57.984
γ = -0. 5426
15 Dec1913 -
May 1923
46.156 34.88 Generalized Pareto 4.317 0.053
-- α = −0.89396
β = 70.49
γ = -0. 48655
237
16 May 1923 –
Sep 1933
40.119 27.45 Generalized Pareto 0.7126 0.067
10.4 α = −2.9088
β = 68.284
γ = -0. 59905
17 Sep 1933 -
Jan 1944
57.191 40.39 Generalized Pareto 4.267 0.045
-- α = −2.9908
β = 97.886
γ = -0. 61441
18 Jan 1944 -
Feb 1954
75.713 51.60 Generalized Pareto 0.539 0.055
6.982 α = −7.0494
β= 130.47
γ = -0. 60125
19 Feb 1954 -
Oct 1964
92.006 69.93 Generalized Pareto 1.656 0.085
13.806 α = −12.116
β= 151.2
γ = -0. 49471
20 Oct 1964 -
May 1976
59.579 37.30 Generalized Pareto 4.497 0.054
-- α = −0.22406
β = 109.56
γ = -0. 81137
21 May1976 -
Mar 1986
86.675 52.80 Generalized Pareto 1.295 0.073
8.389 α = −7.8826
β = 162.73
γ = -0. 77936
22 Mar1986 -
Jun1996
79.273 52.67 Generalized Pareto 1.613 0.085
7.543 α = −7.1456
β= 125.54
γ = -0. 50715
23 Jun1996 -
Jan 2008
57.983 41.21 Generalized Pareto 4.675 0.056
-- α = −4.2933
β = 96.833
γ = -0. 55605
24 Jan 2008-
Process
30.289 46.53 Gen. Extreme Value 5.508 0.154 -- k = -0.62362
σ = 49.473
μ = 22.155
1-23 Aug1755 -
Jan 2008
52.693 44.15 Generalized Pareto 1.699 0.027
15.904 α = 1.8212
β = 67.815
γ = -0. 24514
Table 3.2: The Probability distributions of ENSO cycles (1866-2012)
Cycles Duration MEAN Std.
Dev
Distribution ADT KST CST Parameters
1 1866-1872 0.159 1.141 Gen. Extreme Value 0.296 0.067 4.265 k = -0.0971
σ = 1.0101
μ= -0. 3343
2 1873-1879 0.748 2.295 Generalized Pareto 22.606 0.087 -- α = −1.513
β = 7.4476
γ = -2.8151
3 1880-1886 -0.127 1.063 Gen. Extreme Value 0.1704 0.051 2.123 k= -0.54425
σ = 1.1686
μ = -0.3677
4 1887-1893 0.166 1.172 Gen. Extreme Value 0.407 0.064
2.725 k = -0. 4319
σ = 1.232
μ = -0. 163
5 1894-1900 -0.325 1.230 Gen. Extreme Value 0.235 0.087 3.488 k =-0. 2931
σ = 1.2269
μ =-0.754
6 1901-1907 -0.092 1.280 Logistic 0.208 0.060 3.023 σ = 621
μ =-0.0918
238
7 1908-1914 -0.200 1.076 Gen. Extreme Value 0.286 0.050 5.924 k =-0.13346
σ = 0.9908
μ =-0.6531
8 1915-1921 0.373 1.381 Gen. Extreme Value 0.282 0.061 3.390 k = 0.17923
σ = 1.3295
μ =-0.186
9 1922-1928 0.035 0.905 Gen. Extreme Value 0.526 0.078 5.526 k = -0.4696
σ = 0.99794
μ = -0.2058
10 1929-1935 0.070 0.881 Gen. Extreme Value 4.394 0.072
-- k =-0. 3838
σ = 0.87404
μ =-0. 1935
11 1936-1942 -0.274 1.154 Gen. Extreme Value 0.123 0.055 0.683 k = -0. 2916
σ = 1.1584
μ = -0. 679
12 1943-1949 -0.267 1.024 Generalized Pareto 8.366 0.070 -- α = −1.3903
β = 2.7067
γ = -1.0708
13 1950-1956 -0.101 1.723 Generalized Pareto 11.787 0.060 -- α = −1.6119
β = 3.8763
γ = -1.0454
14 1957-1963 -0.100 0.775 Gen. Extreme Value 4.044 0.044 -- k =-0. 4323
σ = 0.81599
μ = -0.3173
15 1964-1970 -0.127 0.892 Gen. Extreme Value 0.283 0.054
2.79 k =-0. 2158
σ = 0.87581
μ = -0.4734
16 1971-1977 -0.047 1.694 Generalized Pareto 8.145 0.069 -- α = −1.9633
= 4.8623
γ = -1.1491
17 1978-1984 -0.494 1.067 Gen. Extreme Value 8.564 0.101
-- k = -0. 7049
σ = 1.152
μ = -0.6404
18 1985-1991 -0.261 1.082 Gen. Extreme Value 0.232 0.056 4.020 k = -0.268
σ = 1.1043
μ = -0.6579
19 1992-1998 -0.714 1.182 Gen. Extreme Value 0.230 0.050 1.070 k = -0. 3241
σ = 1.253
μ = -0.1198
20 1999-2005 -0.114 0.965 Gen. Extreme Value 0.190 0.051 3.800 k = -0.2671
σ = 0.96223
μ = -0.4649
21 2006-2012 0.390 1.109 Gen. Extreme Value 0.272 0.058 2.845 k = -0. 2063
σ = 1.1028
μ= -0.0493
22 1981-2000 -0.387 1.213 Gen. Extreme Value 0.278 0.034 6.370 k = -0.4334
σ = 1.3029
μ = -0.7295
1-21 1866-2012 -0.057 1.147 Gen. Extreme Value 8.982 0.021 -- k = -0. 2773
σ = 1.1037
μ = -0.4582
Table 3.3: The probability distribution of active ENSO cycles (1981-2000)
239
Cycles Duration Mean Std.
Dev
Distribution ADT KST CST Parameters
1 May1976 -
Mar 1986
0.500 1.021 Gen. Extreme Value 11.881 0.06571 -- k = -0. 5799
σ = 1.1017
μ = -0. 7021
2 Mar1986 -
Jun1996
0.527 1.097 Gen. Extreme Value 0.25186 0.04856
3.3612 k = -0. 2392
σ = 1. 0941
μ = -0. 9415
3 Jun1996 -
Jan 2008
0.180 1.107 Gen. Extreme Value 0.21305 0.03841
0.21305 k= -0.40288
σ = 1. 0818
μ = -0. 1928
Table 3.4: Probability distribution of Fractal dimension of Sunspot cycles (1-23 and total 1-23) and Enso
cycles (1-22 and total 1-21)
Cycles FD Mean St. D Distribution ADT KST CST Parameters
Sunspots
FDS 1.212 0.0751 Gen. Extreme Value 0.17822 0.0791
0.17981 k = -0. 491
σ = 0.0818
μ = 1.1928
Sunspot FDA 1.387 0.0974 Gen. Extreme Value 4.2742 0.15377 -- k =-0. 4567
σ = 0.102
μ = 1.3606
ENSO FDS 0.580 0.3135 Generalized Pareto 8.2478 0.07311
-- α = 1.0007
β =0.00288
γ = 0.4908
ENSO FDA 1.158 0.093 Gen. Extreme Value 0.09756 0.0685
0.16655 k= -0. 1378
σ = 0.0911
μ = 1.117
Table 3.5: the fractional differencing parameter and heavy tails parameter of sunspot cycles
Cycles Duration HES 0 < dS <1/2 1< βS < 2 HEA 0 < dA <1/2 1< βA < 2
1 Aug 1755 - Mar 1766 0.819 0.319 1.362 0.633 0.133 1.734
2 Mar 1766 - Aug 1775 0.73 0.230 1.540 0.593 0.093 1.814
3 Aug 1775 - Jun 1784 0.672 0.172 1.656 0.500 0.00 2.000
4 Jun 1784 - Jun 1798 0.682 0.182 1.636 0.554 0.054 1.892
5 Jun 1798 - Sep 1810 0.901 0.401 1.198 0.620 0.120 1.760
6 Sep1810 - Dec 1823 0.864 0.364 1.272 0.637 0.137 1.726
7 Dec1823 - Oct 1833 0.823 0.323 1.354 0.837 0.337 1.326
8 Oct1833 - Sep 1843 0.704 0.204 1.592 0.772 0.272 1.456
9 Sep1843 - Mar 1855 0.725 0.225 1.550 0.530 0.030 1.940
10 Mar1855 - Feb 1867 0.733 0.233 1.534 0.687 0.187 1.626
11 Feb1867 - Sep 1878 0.702 0.202 1.596 0.653 0.153 1.694
12 Sep1878 - Jun 1890 0.835 0.335 1.330 0.537 0.037 1.926
13 Jun1890 - Sep 1902 0.748 0.248 1.504 0.624 0.124 1.752
14 Sep 1902 - Dec 1913 0.859 0.359 1.282 0.600 0.100 1.800
15 Dec1913 - May 1923 0.765 0.265 1.470 0.521 0.021 1.958
16 May 1923 - Sep 1933 0.851 0.351 1.298 0.604 0.104 1.792
17 Sep 1933 - Jan 1944 0.794 0.294 1.412 0.500 0.00 2.000
18 Jan 1944 - Feb 1954 0.806 0.306 1.388 0.627 0.127 1.746
240
19 Feb 1954 - Oct 1964 0.734 0.234 1.532 0.590 0.090 1.820
20 Oct 1964 - May 1976 0.807 0.307 1.386 0.645 0.145 1.710
21 May1976 - Mar 1986 0.802 0.302 1.396 0.813 0.313 1.374
22 Mar1986 - Jun1996 0.775 0.275 1.450 0.563 0.060 1.874
23 Jun1996 - Jan 2008 0.787 0.287 1.426 0.549 0.049 1.902
24 Aug1755 - Jan 2008 0.998 0.498 1.004 0.649 0.149 1.702
Table 3.6: the fractional differencing parameter and heavy tails parameter of ENSO cycles
Cycles Duration HES 0 < dS <1/2 1< βS < 2 HEA 0 < dA <1/2 1< βA < 2
1 1866-1872 0.941 0.441 1.118 0.792 0.292 1.416
2 1873-1879 0.990 0.490 1.020 0.910 0.410 1.180
3 1880-1886 0.998 0.498 1.004 0.640 0.140 1.720
4 1887-1893 0.994 0.494 1.012 0.844 0.344 1.312
5 1894-1900 0.996 0.496 1.008 0.871 0.371 1.258
6 1901-1907 0.995 0.495 1.010 0.829 0.329 1.342
7 1908-1914 0.997 0.497 1.006 0.986 0.486 1.028
8 1915-1921 0.993 0.493 1.014 0.959 0.459 1.082
9 1922-1928 0.997 0.497 1.006 0.736 0.236 1.528
10 1929-1935 0.999 0.499 1.002 0.686 0.186 1.628
11 1936-1942 0.995 0.495 1.010 0.999 0.499 1.002
12 1943-1949 0.998 0.498 1.004 0.709 0.209 1.582
13 1950-1956 0.997 0.497 1.006 0.984 0.484 1.032
14 1957-1963 0.999 0.499 1.002 0.840 0.340 1.320
15 1964-1970 0.995 0.495 1.010 0.798 0.298 1.404
16 1971-1977 0.994 0.494 1.012 0.913 0.413 1.174
17 1978-1984 0.995 0.495 1.010 0.879 0.379 1.242
18 1985-1991 0.995 0.495 1.010 0.887 0.387 1.226
19 1992-1998 0.993 0.493 1.014 0.882 0.382 1.236
20 1999-2005 0.999 0.499 1.002 0.930 0.430 1.140
21 2006-2012 0.994 0.494 1.012 0.822 0.322 1.356
22 1981-2000 1 0.50 1 0.732 0.232 1.536
23 1866-2012 1 0.50 1 0.776 0.276 1.448
Table 3.7: The strength of long range-correlation of sunspot cycles
Cycles Duration HES 1 < 𝛾𝑆 < 3 HEA -1 < 𝛾𝐴< 1
1 Aug 1755 - Mar 1766 0.819 2.638 0.633 0.266
2 Mar 1766 - Aug 1775 0.73 2.460 0.593 0.186
3 Aug 1775 - Jun 1784 0.672 2.344 0.500 0.00
4 Jun 1784 - Jun 1798 0.682 2.364 0.554 0.108
5 Jun 1798 - Sep 1810 0.901 2.802 0.620 0.240
6 Sep1810 - Dec 1823 0.864 2.728 0.637 0.274
7 Dec1823 - Oct 1833 0.823 2.646 0.837 0.674
8 Oct1833 - Sep 1843 0.704 2.408 0.772 0.544
9 Sep1843 - Mar 1855 0.725 2.450 0.530 0.060
10 Mar1855 - Feb 1867 0.733 2.466 0.687 0.374
11 Feb1867 - Sep 1878 0.702 2.404 0.653 0.306
12 Sep1878 - Jun 1890 0.835 2.670 0.537 0.074
241
13 Jun1890 - Sep 1902 0.748 2.496 0.624 0.248
14 Sep 1902 - Dec 1913 0.859 2.718 0.600 0.200
15 Dec1913 - May 1923 0.765 2.530 0.521 0.042
16 May 1923 - Sep 1933 0.851 2.702 0.604 0.208
17 Sep 1933 - Jan 1944 0.794 2.588 0.500 0.00
18 Jan 1944 - Feb 1954 0.806 2.612 0.627 0.254
19 Feb 1954 - Oct 1964 0.734 2.468 0.590 0.180
20 Oct 1964 - May 1976 0.807 2.614 0.645 0.290
21 May1976 - Mar 1986 0.802 2.604 0.813 0.626
22 Mar1986 - Jun1996 0.775 2.550 0.563 0.126
23 Jun1996 - Jan 2008 0.787 2.574 0.549 0.098
24 Aug1755 - Jan 2008 0.998 2.996 0.649 0.298
Table 3.8: The strength of long range-correlation of ENSO cycles (1-23)
Cycles Duration HES 1 < 𝛾𝑆 < 3 HEA -1 < 𝛾𝐴< 1
1 1866-1872 0.941 2.882 0.792 0.584
2 1873-1879 0.990 2.980 0.910 0.820
3 1880-1886 0.998 2.996 0.640 0.280
4 1887-1893 0.994 2.988 0.844 0.688
5 1894-1900 0.996 2.992 0.871 0.742
6 1901-1907 0.995 2.990 0.829 0.658
7 1908-1914 0.997 2.994 0.986 0.972
8 1915-1921 0.993 2.986 0.959 0.918
9 1922-1928 0.997 2.994 0.736 0.472
10 1929-1935 0.999 2.998 0.686 0.372
11 1936-1942 0.995 2.990 0.999 0.998
12 1943-1949 0.998 2.996 0.709 0.418
13 1950-1956 0.997 2.994 0.984 0.968
14 1957-1963 0.999 2.998 0.840 0.680
15 1964-1970 0.995 2.990 0.798 0.596
16 1971-1977 0.994 2.988 0.913 0.826
17 1978-1984 0.995 2.990 0.879 0.758
18 1985-1991 0.995 2.990 0.887 0.774
19 1992-1998 0.993 2.986 0.882 0.764
20 1999-2005 0.999 2.998 0.930 0.860
21 2006-2012 0.994 2.988 0.822 0.644
22 1981-2000 1 3 0.732 0.464
23 1866-2012 1 3 0.776 0.552
Table 3.9 (a): Augmented Dickey Fuller (ADF) test (stationary)
Self-Similar Tail Parameter (βS) Sunspots Cycles
Null Hypothesis H0: Self-Similar Tail Parameter (βS) of Sunspots Cycles has
a unit root
Exogenous: Constant
Lag Length: 0 (Automatics-based on SIC, MaxLag = 5)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -5.041712 0.0005
242
Test Critical vales 1% level
5% level
10% level
-3.759246
-2.98064
-3.638752
Table 3.9(b): Augmented Dickey Fuller (ADF) test (stationary) Self-Affine Tail
Parameter (βA) Sunspots Cycles
Null Hypothesis H0: Self-Affine Tail Parameter (βA) of Sunspots Cycles has a unit root
Exogenous: Constant
Lag Length: 0 (Automatics-based on SIC, MaxLag = 5)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -4.762830 0.0010
Test Critical vales 1% level
5% level
10% level
-3.759246
-2.998064
-2.638752
Table 3.10 (a): Augmented Dickey Fuller (ADF) test (stationary) Self-Similar Tail
Parameter (βS) ENSO Cycles
Null Hypothesis H0: Self-Similar Tail Parameter (βS) of ENSO Cycles has a unit root
Exogenous: Constant
Lag Length: 0 (Automatics-based on SIC, MaxLag = 4)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -21.04327 0.000
Test Critical vales 1% level
5% level
10% level
-3.769597
-3.004861
-2.642242
Table 3.10 (b): Augmented Dickey Fuller (ADF) test (stationary) Self-Affine Tail
Parameter (βA) ENSO Cycles
Null Hypothesis H0: Self-Affine Tail Parameter (βA) of ENSO Cycles has a unit root
Exogenous: Constant
Lag Length: 0 (Automatics-based on SIC, MaxLag = 4)
t-statistics prob*
Augmented Dickey Fuller (ADF) test - 5.635757 0.0002
Test Critical vales 1% level
5% level
10% level
-3.769597
-3.004861
-2.642242
Table 3.11 (a): Augmented Dickey Fuller (ADF) test (stationary) Self-Similar strength
of long range-correlation (𝛾 S) Sunspots Cycles
243
Null Hypothesis H0: Self-Similar strength of long range-correlation (𝛾 S) of Sunspots Cycles has a unit
root
Exogenous: Constant
Lag Length: 0 (Automatics-based on SIC, MaxLag = 5)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -3.159720 0.0360
Test Critical vales 1% level
5% level
10% level
-3.752946
-2.998064
-2.638752
Table 3.11 (b): Augmented Dickey Fuller (ADF) test (stationary) Self-Affine strength of long
range-correlation (𝛾 A) Sunspots Cycles
Null Hypothesis H0: Self-Affine strength of long range-correlation (𝛾 A) of Sunspots Cycles has a unit root
Exogenous: Constant
Lag Length: 0 (Automatics-based on SIC, MaxLag = 5)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -3.958201 0.0063
Test Critical vales 1% level
5% level
10% level
-3.759246
-2.998064
-2.638752
Table 3.12 (a): Augmented Dickey Fuller (ADF) test (stationary) Self-Similar strength of long
range-correlation (𝛾 S) ENSO Cycles
Null Hypothesis H0: Self-Similar strength of long range-correlation (𝛾 S) of ENSO Cycles has a unit root
Exogenous: Constant
Lag Length: 0 (Automatics-based on SIC, MaxLag = 4)
t-statistics prob*
Augmented Dickey Fuller (ADF) test -21.04327 0.000
Test Critical vales 1% level
5% level
10% level
-3.769597
-3.004861
-2.642242
Table 3.12 (b): Augmented Dickey Fuller (ADF) test (stationary) Self-Affine strength of long
range-correlation (𝛾 A) ENSO Cycles
Null Hypothesis H0: Self-Affine strength of long range-correlation (𝛾 A) of ENSO Cycles has a unit root
Exogenous: Constant
Lag Length: 0 (Automatics-based on SIC, MaxLag = 4)
t-statistics prob*
Augmented Dickey Fuller (ADF) test - 5.635757 0.0002
244
Test Critical vales 1% level
5% level
10% level
-3.769597
-3.004861
-2.642242
Table 4.2(a): showing the first contour of AR9114
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
−𝐻𝜑
𝐸𝑇
||
−𝐻∅𝐸𝑇
𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -5×1040 3.5 0.16×1033 4200 -2.60×108 0.00020
2 -4.5×1040 3 0.15×1033 4200 -2.14×108 0.000164
3 -4×1040 2.5 0.14×1033 4200 -3.06×108 0.00017
4 -3.5×1040 2 0.13×1033 4200 -1.28×108 0.000071
5 -3×1040 1.5 0.12×1033 4200 -8.929×1010 0.0496
Table 4.2(b): showing the second contour of AR9114
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
−𝐻𝜑
𝐸𝑇
||
−𝐻∅𝐸𝑇
𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -5×1040 3.5 0.16×1033 4300 -2.54×108 0.00014
2 -4.5×1040 3 0.15×1033 4300 -2.09×108 0.000161
3 -4×1040 2.5 0.14×1033 4300 -1.66×108 0.000922
4 -3.5×1040 2 0.13×1033 4300 -1.25×108 0.000694
5 -3×1040 1.5 0.12×1033 4300 -8.721×1010 0.04845
Table 4.2(c): showing the third contour of AR9114
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
−𝐻𝜑
𝐸𝑇
||
−𝐻∅𝐸𝑇
𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -5×1040 3.5 0.16×1033 4400 -2.49×108 0.0001382
2 -4.5×1040 3 0.15×1033 4400 -2.05×108 0.0001139
3 -4×1040 2.5 0.14×1033 4400 -1.62×108 0.00090
4 -3.5×1040 2 0.13×1033 4400 -1.22×108 0.0000678
5 -3×1040 1.5 0.12×1033 4400 -8.523×1010 0.04735
Table 4.2(d): showing the forth contour of AR9114
245
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
−𝐻𝜑
𝐸𝑇
||
−𝐻∅𝐸𝑇
𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -5×1040 3.5 0.16×1033 4500 -2.43×108 0.000135
2 -4.5×1040 3 0.15×1033 4500 -2×109 0.000111
3 -4×1040 2.5 0.14×1033 4500 -1.59×109 0.000833
4 -3.5×1040 2 0.13×1033 4500 -1.197×109 0.0000665
5 -3×1040 1.5 0.12×1033 4500 -8.333×1010 0.04629
Table 4.3(a): showing the first contour of AR9114
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
+𝐻𝜑
𝐸𝑇
||
+𝐻∅𝐸𝑇
+𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -2.4×1041 3.5 0.16×1033 4200 -1.25×1010 0.004661
2 -2×1041 3 0.15×1033 4200 -9.523×1011 0.3553
3 -1.5×1041 2.5 0.14×1033 4200 -6.377×1011 0.23795
4 -1×1041 2 0.13×1033 4200 -3.663×1011 0.13668
5 -0.5×1041 1.5 0.12×1033 4200 -1.488×1011 0.0555
Table 4.3(b): showing the second contour of AR9114
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
+𝐻𝜑
𝐸𝑇
||
+𝐻∅𝐸𝑇
+𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -2.4×1041 3.5 0.16×1033 4300 -1.22×1010 0.004552
2 -2×1041 3 0.15×1033 4300 -9.302×1011 0.34709
3 -1.5×1041 2.5 0.14×1033 4300 -6.229×1011 0.23242
4 -1×1041 2 0.13×1033 4300 -3.578×1011 0.13351
5 -0.5×1041 1.5 0.12×1033 4300 -1.453×1011 0.054216
Table 4.3(c): showing the third contour of AR9114
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
+𝐻𝜑
𝐸𝑇
||
+𝐻∅𝐸𝑇
+𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -2.4×1041 3.5 0.16×1033 4400 -1.193×1010 0.004451
2 -2×1041 3 0.15×1033 4400 -9.091×1011 0.33922
3 -1.5×1041 2.5 0.14×1033 4400 -6.088×1011 0.22716
4 -1×1041 2 0.13×1033 4400 -3.497×1011 0.13049
5 -0.5×1041 1.5 0.12×1033 4400 -1.420×1011 0.052985
246
Table 4.3(d): showing the forth contour of AR9114
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
+𝐻𝜑
𝐸𝑇
||
+𝐻∅𝐸𝑇
+𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -2.4×1041 3.5 0.16×1033 4500 -1.167×1010 0.004354
2 -2×1041 3 0.15×1033 4500 -8.889×1011 0.331679
3 -1.5×1041 2.5 0.14×1033 4500 -5.952×1011 0.220896
4 -1×1041 2 0.13×1033 4500 -3.419×1011 0.127575
5 -0.5×1041 1.5 0.12×1033 4500 -1.389×1011 0.051828
Table 4.4 (a): showing the first contour of AR10696
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
−𝐻𝜑
𝐸𝑇
||
−𝐻∅𝐸𝑇
𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -2×1041 2.5 0.16×1033 4200 -3.16×1010 0.0176
2 -1.5×1041 2 0.15×1033 4200 -2.86×1010 0.0159
3 -1×1041 1.5 0.14×1033 4200 -2.42×1010 0.0134
4 -10×1041 1 0.13×1033 4200 -1.83×1010 0.0102
5 -10.5×1041 0.5 0.12×1033 4200 -1.04×1010 0.0578
Table 4.4 (b): showing the second contour of AR10696
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
−𝐻𝜑
𝐸𝑇
||
−𝐻∅𝐸𝑇
𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -8.5×1041 2.5 0.16×1033 4300 -3.03×1010 0.0168
2 -9×1041 2 0.15×1033 4300 -2.79×1010 0.0155
3 -9.5×1041 1.5 0.14×1033 4300 -2.367×1010 0.0132
4 -10×1041 1 0.13×1033 4300 -1.789×1010 0.0099
5 -10.5×1041 0.5 0.12×1033 4300 -1.017×1010 0.0057
Table 4.4 (c): showing the third contour of AR10696
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
−𝐻𝜑
𝐸𝑇
||
−𝐻∅𝐸𝑇
𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -8.5×1041 2.5 0.16×1033 4400 -3.018×1010 0.01677
247
2 -9×1041 2 0.15×1033 4400 -2.727×1010 0.01515
3 -9.5×1041 1.5 0.14×1033 4400 -2.313×1010 0.01285
4 -10×1041 1 0.13×1033 4400 -1.748×1010 0.0971
5 -10.5×1041 0.5 0.12×1033 4400 -9.943×1011 0.552
Table 4.4 (d): showing the forth contour of AR10696
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
−𝐻𝜑
𝐸𝑇
||
−𝐻∅𝐸𝑇
𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -8.5×1041 2.5 0.16×1033 4500 -2.951×1010 0.01394
2 -9×1041 2 0.15×1033 4500 -2.667×1010 0.01482
3 -9.5×1041 1.5 0.14×1033 4500 -2.262×1010 0.01257
4 -10×1041 1 0.13×1033 4500 -1.709×1010 0.00949
5 -10.5×1041 0.5 0.12×1033 4500 -9.722×1011 0.5401
Table 4.5 (a): showing the first contour of AR10696
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
+𝐻𝜑
𝐸𝑇
||
+𝐻∅𝐸𝑇
+𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -2×1041 2.5 0.16×1033 4200 -7.44×1011 0.4089
2 -2.5×1041 2 0.15×1033 4200 -7.937×1011 0.5023
3 -3×1041 1.5 0.14×1033 4200 -7.653×1011 0.4844
4 -3.5×1041 1 0.13×1033 4200 -6.410×1011 0.4057
5 -4×1041 0.5 0.12×1033 4200 -3.968×1011 0.5211
Table 4.5 (b): showing the second contour of AR10696
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
+𝐻𝜑
𝐸𝑇
||
+𝐻∅𝐸𝑇
+𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -2×1041 2.5 0.16×1033 4300 -7.27×1011 0.4601
2 -2.5×1041 2 0.15×1033 4300 -7.75×1011 0.4905
3 -3×1041 1.5 0.14×1033 4300 -7.48×1011 0.4734
4 -3.5×1041 1 0.13×1033 4300 -6.26×1011 0.3962
5 -4×1041 0.5 0.12×1033 4300 -3.88×1011 0.2456
Table 4.5 (c): showing the third contour of AR10696
248
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
+𝐻𝜑
𝐸𝑇
||
+𝐻∅𝐸𝑇
+𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -2×1041 2.5 0.16×1033 4400 -7.10×1011 0.4494
2 -2.5×1041 2 0.15×1033 4400 -7.58×1011 0.4797
3 -3×1041 1.5 0.14×1033 4400 -7.31×1011 0.4627
4 -3.5×1041 1 0.13×1033 4400 -6.12×1011 0.3873
5 -4×1041 0.5 0.12×1033 4400 -3.79×1011 0.2399
Table 4.5 (d): showing the forth contour of AR10696
S.no Helicity (H)
unit Mx2/h
Tangential
velocity (φ)
unit ∅o/h
Energy (E)
unit egrs
Temperature
(T) unit K
+𝐻𝜑
𝐸𝑇
||
+𝐻∅𝐸𝑇
+𝐻𝑚𝑎𝑥∅𝑚𝑎𝑥
𝐸𝑚𝑎𝑥𝑇𝑎𝑣𝑔
||
1 -2×1041 2.5 0.16×1033 4500 -6.94×1011 0.4392
2 -2.5×1041 2 0.15×1033 4500 -7.41×1011 0.4690
3 -3×1041 1.5 0.14×1033 4500 -7.14×1011 0.4519
4 -3.5×1041 1 0.13×1033 4500 -5.98×1011 0.3785
5 -4×1041 0.5 0.12×1033 4500 -3.70×1011 0.2342