time series analysis and forecasting - university ofmath.unm.edu/~alvaro/time series.pdf · time...
TRANSCRIPT
![Page 1: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/1.jpg)
Time Series Analysis and ForecastingMath 667
Al Nosedal
Department of Mathematics
Indiana University of Pennsylvania
Time Series Analysis and Forecasting – p. 1/115
![Page 2: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/2.jpg)
Introduction
Many decision-making applications depend on a forecast ofsome quantity. Here are a couple of examples.
When a service organization, such as a fast-foodrestaurant, plans its staffing over some time period, itmust forecast the customer demand as a function oftime.
Time Series Analysis and Forecasting – p. 2/115
![Page 3: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/3.jpg)
Introduction
Many decision-making applications depend on a forecast ofsome quantity. Here are a couple of examples.
When a service organization, such as a fast-foodrestaurant, plans its staffing over some time period, itmust forecast the customer demand as a function oftime.
When a company plans its ordering or productionschedule for a product it sells to the public, it mustforecast the customer demand for this product so that itcan stock appropriate quantities-neither too much nortoo little.
Time Series Analysis and Forecasting – p. 2/115
![Page 4: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/4.jpg)
Forecasting Methods: An Overview
There are many forecasting methods available, thesemethods can generally be divided into three groups:
1. Judgemental methods.
Time Series Analysis and Forecasting – p. 3/115
![Page 5: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/5.jpg)
Forecasting Methods: An Overview
There are many forecasting methods available, thesemethods can generally be divided into three groups:
1. Judgemental methods.
2. Extrapolation (or Time Series) methods, and
Time Series Analysis and Forecasting – p. 3/115
![Page 6: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/6.jpg)
Forecasting Methods: An Overview
There are many forecasting methods available, thesemethods can generally be divided into three groups:
1. Judgemental methods.
2. Extrapolation (or Time Series) methods, and
3. Econometric (or causal) methods.
Time Series Analysis and Forecasting – p. 3/115
![Page 7: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/7.jpg)
Extrapolation Methods
Extrapolation methods are quantitative methods that usepast data of a time series variable-and nothing else, exceptpossibly time itself-to forecast future values of the variable.The idea is that we can use past movements of a variable,such as a company sales to forecast its future values.
Time Series Analysis and Forecasting – p. 4/115
![Page 8: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/8.jpg)
Extrapolation Methods
There are many extrapolation methods available, includingtrend-based regression, moving averages, andautoregression models. All these extrapolation methodssearch for patterns in the historical series and thenextrapolate these patterns into the future.
Time Series Analysis and Forecasting – p. 5/115
![Page 9: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/9.jpg)
Random Series
The simplest time series model is the random model. Arandom model can be written as
Y (t) = µ + ε(t)(1)
Here, µ is a constant, the average of the Y (t)′s, and ε(t)
is the residual (or error) term. We assume that theresiduals have mean 0, variance σ2, and areprobabilistically independent of one another.
Time Series Analysis and Forecasting – p. 6/115
![Page 10: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/10.jpg)
Random series (Graph)
0 10 20 30 40
−1
01
2
Observation number
Y(t
)
Time Series Analysis and Forecasting – p. 7/115
![Page 11: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/11.jpg)
Random Series (1)
There are two situations where random time series occur.
The first is when the original time series is random. Forexample, when studying the time pattern of diameters ofindividual parts from a manufacturing process, we mightdiscover that the successive diameters behave like arandom time series.
Time Series Analysis and Forecasting – p. 8/115
![Page 12: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/12.jpg)
Random Series (1)
There are two situations where random time series occur.
The first is when the original time series is random. Forexample, when studying the time pattern of diameters ofindividual parts from a manufacturing process, we mightdiscover that the successive diameters behave like arandom time series.
Time Series Analysis and Forecasting – p. 8/115
![Page 13: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/13.jpg)
Random Series (1)
There are two situations where random time series occur.
The first is when the original time series is random. Forexample, when studying the time pattern of diameters ofindividual parts from a manufacturing process, we mightdiscover that the successive diameters behave like arandom time series.
Time Series Analysis and Forecasting – p. 8/115
![Page 14: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/14.jpg)
Random Series (2)
The second situation where a random series occurs ismore common. This is when we fit a model to a timeseries to obtain an equation of the form
Y (t) = fitted part + residual.(2)
Although the fitted part varies from model to model, itsessential feature is that it describes any underlying timeseries pattern in the original data. The residual is thenwhatever is left, and we hope that the series of residualsis random with mean µ = 0. The fitted part is used tomake forecasts, and the residuals, often called noise,are forecasted to be zero.
Time Series Analysis and Forecasting – p. 9/115
![Page 15: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/15.jpg)
Random Series (2)
The second situation where a random series occurs ismore common. This is when we fit a model to a timeseries to obtain an equation of the form
Y (t) = fitted part + residual.(3)
Although the fitted part varies from model to model, itsessential feature is that it describes any underlying timeseries pattern in the original data. The residual is thenwhatever is left, and we hope that the series of residualsis random with mean µ = 0. The fitted part is used tomake forecasts, and the residuals, often called noise,are forecasted to be zero.
Time Series Analysis and Forecasting – p. 9/115
![Page 16: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/16.jpg)
Runs test
For each observation Y (t) we associate a 1 if Y (t) ≥ y and 0if Y (t) < y. A run is a consecutive sequence of 0’s or 1’s. LetT be the number of observations, let Ta above the mean,and let Tb the number below the mean. Also let R be theobserved number of runs. Then it can be show that for arandom series
E(R) =T + 2TaTb
T(4)
Stdev(R) =
√
2TaTb(2TaTb − T )
T 2(T − 1)(5)
When T > 20, the distribution of R is roughly Normal.Time Series Analysis and Forecasting – p. 10/115
![Page 17: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/17.jpg)
Example: Runs test
Suppose that the successive observations are 87, 69, 53,57, 94, 81, 44, 68, and 77, with mean Y = 70. It is possiblethat this series is random. Does the runs test support thisconjecture?
Time Series Analysis and Forecasting – p. 11/115
![Page 18: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/18.jpg)
Example: Runs test (solution)
The preceding sequence has five runs: 1; 0 0 0; 1 1; 0 0;and 1. Then, we have T = 9, Ta = 4, Tb = 5 and R = 5. Undera randomness hypothesis,
E(R) = 5.44(6)
Stdev(R) = 1.38(7)
Z =R − E(R)
Stdev(R)= −0.32(8)
Time Series Analysis and Forecasting – p. 12/115
![Page 19: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/19.jpg)
Exercise
Write a function in R that performs a run test. My version ofthis function will be posted on our website tomorrow.
Time Series Analysis and Forecasting – p. 13/115
![Page 20: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/20.jpg)
Autocorrelation (1)
Recall that the successive observations in a random seriesare probabilistically independent of one another. Many timeseries violate this property and are instead autocorrelated.For example, in the most common form of autocorrelation,positive autocorrelation, large observations tend to followlarge observations, and small observations tend to followsmall observations. In this case the runs test is likely to pickit up. Another way to check for the same nonrandomnessproperty is to calculate the autocorrelations of the timeseries.
Time Series Analysis and Forecasting – p. 14/115
![Page 21: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/21.jpg)
Autocorrelation (2)
To understand autocorrelation it is first necessary tounderstand what it means to lag a time series.Y (t) = Yt =(3,4,5,6,7,8,9,10,11,12)Y lagged 1 = Y (t − 1) = Yt−1 =(*,3,4,5,6,7,8,9,10,11)
Time Series Analysis and Forecasting – p. 15/115
![Page 22: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/22.jpg)
Autocovariance Function
The autocovariance function is defined as follows:
γ(k) = E(Y (t) − µ)(Y (t + k) − µ), k = 0,±1,±2, ...(9)
where Y (t) represents the values of the series, µ is themean of the series and k is the lag at which theautocovariance is being considered.
Time Series Analysis and Forecasting – p. 16/115
![Page 23: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/23.jpg)
Autocorrelation Function
A standardized version of γ(k) is the autocorrelationfunction, defined as
ρ =γ(k)
γ(0)(10)
Time Series Analysis and Forecasting – p. 17/115
![Page 24: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/24.jpg)
Estimator of the autocovariance function
The autocovariance function at lag k can be estimated by
γ(k) =1
n
n−k∑
t=1
(Y (t) − Y )(Y (t + k) − Y )(11)
Time Series Analysis and Forecasting – p. 18/115
![Page 25: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/25.jpg)
Estimator of the autocorrelation function
The autocorrelation function estimate at lag k is simply
ρ(k) =γ(k)
γ(0)(12)
Time Series Analysis and Forecasting – p. 19/115
![Page 26: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/26.jpg)
Example
Find the autocorrelation at lag 1 for the following time series:Y (t) = (3, 4, 5, 6, 7, 8, 9, 10, 11, 12)
Answer.ρ(1) = 0.7
Time Series Analysis and Forecasting – p. 20/115
![Page 27: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/27.jpg)
Finding the autocorrelation at lag 1
y<-3:12
### function to find autocorrelation when###lag=1my.auto<-function(x){
n<-length(x)denom<-(x-mean(x))%*%(x-mean(x))/nnum<-(x[-n]-mean(x))%*%(x[-1]-mean(x))/nresult<-num/denom
return(result)}
my.auto(y)
Time Series Analysis and Forecasting – p. 21/115
![Page 28: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/28.jpg)
An easier way of finding autocorrelations
Using R, we can easily find the autocorrelation at any lag k.
y<-3:12
auto.lag1<-acf(y,lag=1)$acf[2]
auto.lag1
Time Series Analysis and Forecasting – p. 22/115
![Page 29: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/29.jpg)
Graph of autocorrelations
A graph of the lags against the correspondingautocorrelations is called a correlogram. The following linesof code can be used to make a correlogram in R.### autocorrelation function for###a random series
y<-rnorm(25)
acf(y,lag=8,main=’Random Series N(0,1)’)
Time Series Analysis and Forecasting – p. 23/115
![Page 30: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/30.jpg)
Correlogram (Graph)
0 2 4 6 8
−0.
40.
00.
40.
8
Lag
AC
F
Random Series N(0,1)
Time Series Analysis and Forecasting – p. 24/115
![Page 31: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/31.jpg)
Example: Stereo Retailers
Monthly sales for a chain of stereo retailers are listed in thefile stereo.txt. They cover the period from the beginning of1995 to the end of 1998, during which there was no upwardor downward trend in sales and no clear seasonal peaks orvalleys. This behavior is apparent in the time series chart ofsales shown in our next slide. It is possible that this series israndom. Do autocorrelations support this conclusion?
Time Series Analysis and Forecasting – p. 25/115
![Page 32: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/32.jpg)
Time Series Plot (stereo.txt)
0 10 20 30 40
150
200
250
1:N
sale
s
Time Series Analysis and Forecasting – p. 26/115
![Page 33: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/33.jpg)
Time Series Plot (stereo.txt)
Time
new
.sal
es
1995 1996 1997 1998 1999
150
200
250
Time Series Analysis and Forecasting – p. 27/115
![Page 34: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/34.jpg)
Correlogram (stereo.txt)
0 2 4 6 8 10 12
−0.
20.
20.
61.
0
Lag
AC
F
Stereo Sales
Time Series Analysis and Forecasting – p. 28/115
![Page 35: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/35.jpg)
How large is a "large" autocorrelation?
Suppose that Y1, ..., YT are independent and identicallydistributed random variables with arbitrary mean. Then itcan be shown that
E(ρk) '−1
T(13)
V ar(ρk) '1
T(14)
Thus, having plotted the correlogram, we can plotapproximate 95% confidence limits at −1/T ± 2/
√T , which
are often further approximated to ±2/√
T .
Time Series Analysis and Forecasting – p. 29/115
![Page 36: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/36.jpg)
Conclusion (Stereo Sales)
In this case,√
V (ρk) = 1/√
48. If the series is truly random,then only an occasional autocorrelation should be largerthan two standard errors in magnitude. Therefore, anyautocorrelation that is larger than two standard errors inmagnitude is worth our attention. The only "large"autocorrelation for the sales data is the first, or lag 1. Itseems that the pattern of sales is not completely random.
Time Series Analysis and Forecasting – p. 30/115
![Page 37: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/37.jpg)
Example: Demand
The dollar demand for a certain class of parts at a localretail store has been recorded for 82 consecutive days.(See the file demand.txt) A time series plot of thesedemands appears in the next slide. The store managerwants to forecast future demands. In particular, he wants toknow whether is any significant time pattern to the historicaldemands or whether the series is essentially random.
Time Series Analysis and Forecasting – p. 31/115
![Page 38: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/38.jpg)
Time Series Plot of Demand for Parts
Time Series Plot of Demand for Parts
Time
Dai
ly d
eman
d
0 20 40 60 80
150
200
250
300
350
Time Series Analysis and Forecasting – p. 32/115
![Page 39: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/39.jpg)
Solution
A visual inspection of the time series graph in the previousslide shows that demands vary randomly around the samplemean of $247.54 (shown as the horizontal centerline). Thevariance appears to be constant through time, and there areno obvious time series patterns. To check formally whetherthis apparent randomness holds, we perform the runs testand calculate the first 10 autocorrelations. (The associatedcorrelogram appears in the next slide). The p-value for theruns test is relatively large 0.118 and none of theautocorrelations is significantly large. These findings areconsistent with randomness.
Time Series Analysis and Forecasting – p. 33/115
![Page 40: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/40.jpg)
Graph
0 2 4 6 8 10
−0.
20.
20.
40.
60.
81.
0
Lag
AC
F
Correlogram for Demand Data
Time Series Analysis and Forecasting – p. 34/115
![Page 41: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/41.jpg)
Conclusion
For all practical purposes there is no time series pattern tothese demand data. It is as if every day’s demand is anindependent draw from a distribution with mean $247.54and standard deviation $47.78. Therefore, the managermight as well forecast that demand for any day in the futurewill be $247.54. If he does so, about 95% of his forecastsshould be within two standard deviations (about $95) of theactual demands.
Time Series Analysis and Forecasting – p. 35/115
![Page 42: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/42.jpg)
A simulation
It is often helpful to study a time series model by simulation.This enables the main features of the model to be ’observed’in plots, so that when historical data exhibit similar featuresthe model may be selected as a potential candidate. Thefollowing commands can be used to simulate random walkdata for y.
set.seed(2) #we do this to get#the same random numbersy<-e<-rnorm(1000)for (t in 2:1000){ y[t] <- y[t-1] + e[t]}plot(y, type=’l’,col=’blue’)
Time Series Analysis and Forecasting – p. 36/115
![Page 43: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/43.jpg)
Simulation (plot)
0 200 400 600 800 1000
020
4060
Index
y
Time Series Analysis and Forecasting – p. 37/115
![Page 44: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/44.jpg)
Autocorrelations for Random Walk
0 5 10 15 20 25 30
0.0
0.2
0.4
0.6
0.8
1.0
Lag
AC
F
Autocorrelations for Random Walk
Time Series Analysis and Forecasting – p. 38/115
![Page 45: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/45.jpg)
Random Walk Model
Random series are sometimes building blocks for other timeseries models. The model we now discuss, the random walkmodel, is an example of this. In a random walk model theseries itself is not random. However, its differences - that is,the changes from one period to the next - are random. Thistype of behavior is typical of stock price data. For example,the graph in the next slide shows monthly Dow Jonesaverages from January 1988 through March 1992. (See thefile dow.txt)
Time Series Analysis and Forecasting – p. 39/115
![Page 46: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/46.jpg)
Time series chart of closing Dow average
Time Series Plot of Dow Jones Index
Time
clos
ing
Dow
avg
1988 1989 1990 1991 1992
2000
2400
2800
3200
Time Series Analysis and Forecasting – p. 40/115
![Page 47: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/47.jpg)
More about Dow Jones
This series is not random, as can be seen from its gradualupward trend (Although the runs test and autocorrelationsare not shown here, they confirm that the series is notrandom.)If we were standing in March 1992 and were asked toforecast the Dow Jones average for the next few months, itis intuitive that we would not use the average of the historicalvalues as our forecast. This forecast would probably be toolow because the series has an upward trend. Instead, wewould base our forecast on the most recent observation.This is exactly what the random walk model does.
Time Series Analysis and Forecasting – p. 41/115
![Page 48: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/48.jpg)
Random walk (equation)
An equation for the random walk model is
Yt = Yt−1 + µ + εt(15)
where µ is a constant and εt is a random series with mean 0and some standard deviation σ. If we let DYt = Yt − Yt−1, thechange in the series from time t to time t − 1, then we canwrite the random walk model as
DYt = µ + εt(16)
This implies that the differences form a random series withmean µ and standard deviation σ.
Time Series Analysis and Forecasting – p. 42/115
![Page 49: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/49.jpg)
Random walk (cont.)
An estimate of µ is the average of the differences, labeledYD, and an estimate of σ is the sample standard deviation ofthe differences , labeled sD. In words, a series that behavesaccording to this random walk model has randomdifferences, and the series tends to trend upward (if µ > 0) ordownward (if µ < 0) by an amount µ each period. If we arestanding in period t and want to make a forecast Ft+1 of Yt+1,then a reasonable forecast is
Ft+1 = Yt + YD(17)
That is, we add the estimated trend to the currentobservation to forecast the next observation.
Time Series Analysis and Forecasting – p. 43/115
![Page 50: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/50.jpg)
Example: Dow Jones revisited
Given the monthly Dow Jones data in the file dow.txt, checkthat it satisfies the assumptions of a random walk, and usethe random walk model to forecast the value for April 1992.
Time Series Analysis and Forecasting – p. 44/115
![Page 51: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/51.jpg)
Solution
We have already seen that the Dow Jones series itself is notrandom, due to the upward trend, so we form the differencesdiff.dow<-diff(index)ts.diff<-ts(diff.dow)plot(ts.diff,col=’blue’,ylab=’first difference’, main=’Time SeriesPlot of Differences’)abline(h=mean(diff.dow),col=’red’)
Time Series Analysis and Forecasting – p. 45/115
![Page 52: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/52.jpg)
Time Series Plot of Dow Differences
Time Series Plot of Differences
Time
first
diff
eren
ce
0 10 20 30 40 50
−20
00
100
200
Time Series Analysis and Forecasting – p. 46/115
![Page 53: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/53.jpg)
Dow Differences (cont.)
It appears to be a much more random series, varyingaround the mean difference 26. The runs test shows thatthere is absolutely no evidence of nonrandom differences.runs(diff.dow)$T [1] 50$Ta [1] 26$Tb [1] 24$R [1] 26$Z [1] 0.01144965$‘E(R)‘ [1] 25.96$‘P value‘ [1] 0.9908647
Time Series Analysis and Forecasting – p. 47/115
![Page 54: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/54.jpg)
Dow Differences (cont.)
Similarly, the autocorrelations are all small except for arandom "blip" at lag 11. Because there is probably noreason to believe that values 11 months apart really arerelated, we would tend to ignore this autocorrelation.
Time Series Analysis and Forecasting – p. 48/115
![Page 55: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/55.jpg)
Correlogram for Dow Differences
0 2 4 6 8 10 12
−0.
20.
20.
61.
0
Lag
AC
F
Autocorrelations for Dow Differences
Time Series Analysis and Forecasting – p. 49/115
![Page 56: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/56.jpg)
Forecast of April 1992
Assuming the random walk model is adequate, the forecastof April 1992 made in March 1992 is the observed Marchvalue, 3247.42, plus the mean difference, 26, or 3273.42. Ameasure of the forecast accuracy is provided by thestandard deviation, sD=84.65, of the differences. Providedthat our assumptions hold, we can be 95 % confident thatour forecast is off by no more than 2sD, or about 170.
Time Series Analysis and Forecasting – p. 50/115
![Page 57: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/57.jpg)
Exercise
Write a function in R that gives you a one-step-aheadforecast for a Random Walk Model.
Time Series Analysis and Forecasting – p. 51/115
![Page 58: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/58.jpg)
One solution
forecast<-function(y){
diff<-diff(y)y.diff.bar<-mean(diff) #average differencelast<-length(y) #last observationF.next<-y[last]+y.diff.barnew.y<-c(y,F.next)
list(’Y(t)’=y,’Y(t+1)’=new.y,’F(t+1)’=F.next)
}
Time Series Analysis and Forecasting – p. 52/115
![Page 59: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/59.jpg)
Exercise
Using your one-step-ahead forecast function, write anotherfunction in R that computes forecast for times:t+1,t+2,...,t+N, where t represents the length of your originaltime series.
Time Series Analysis and Forecasting – p. 53/115
![Page 60: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/60.jpg)
One solution
forecast.N<-function(y,N){
original<-y #original time series
for (i in 1:N){new<-forecast(original)$’Y(t+1)’original<-new
}
list(’Y(t)’=y,’N’=N,’Y(t+N)’=original)
}
Time Series Analysis and Forecasting – p. 54/115
![Page 61: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/61.jpg)
Forecasts for Dow Jones Avg
Dow Jones Avg with forecasts
Time
Dow
Jon
es a
vera
ge
0 10 20 30 40 50
2000
2500
3000
Time Series Analysis and Forecasting – p. 55/115
![Page 62: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/62.jpg)
Another simulation
set.seed(1)y <- e <- rnorm(100)for (t in 2:100) {y[t]<-0.7*y[t-1] + e[t]}plot(y,type=’l’,col=’blue’)acf(y)
Time Series Analysis and Forecasting – p. 56/115
![Page 63: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/63.jpg)
Autoregression Models
A regression-based extrapolation method is to regress thecurrent value of the time series on past (lagged) values.This is called autoregression, where the "auto" means thatthe explanatory variables in the equation are lagged valuesof the response variable, so that we are regressing theresponse variable on lagged versions of itself. Some trialand error is generally required to see how many lags areuseful in the regression equation. The following exerciseillustrates the procedure.
Time Series Analysis and Forecasting – p. 57/115
![Page 64: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/64.jpg)
Exercise
Plot your time series
Make a correlogram
Fit an autoregressive model of order p (p suggested byyour correlogram)
Determine if your model is adequate
Fit another model if necessary
Time Series Analysis and Forecasting – p. 58/115
![Page 65: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/65.jpg)
Exercise: Hammers
A retailer has recorded its weekly sales of hammers (unitspurchased) for the past 42 weeks. (See the file hammers.txt)How useful is autoregression for modeling these data andhow would it be used for forecasting?
Time Series Analysis and Forecasting – p. 59/115
![Page 66: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/66.jpg)
Exercise: Hammers (cont.)
Plot your time series
Make a correlogram
Fit an autoregressive model of order p (p suggested byyour correlogram)
Determine if your model is adequate
Fit another model if necessary
Time Series Analysis and Forecasting – p. 60/115
![Page 67: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/67.jpg)
Regression-Based Trend Models
Many time series follow a long-term trend except for randomvariation. This trend can be upward or downward. Astraightforward way to model this trend is to estimate aregression equation for Yt, using time t as the singleexplanatory variable. In this "section" we will discuss the twomost frequently used trend models, linear trend andexponential trend.
Time Series Analysis and Forecasting – p. 61/115
![Page 68: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/68.jpg)
Linear Trend
A linear trend means that the time series variable changesby a constant amount each time period. The relevantequation is
Yt = α + βt + εt(18)
where, as in previous regression equations, α is theintercept, β is the slope, and εt is an error term.
Time Series Analysis and Forecasting – p. 62/115
![Page 69: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/69.jpg)
Exercise: Reebok
The file reebok.txt includes quarterly sales data for Reebokfrom the first quarter 1986 through second quarter 1996.Sales increase from $ 174.52 million in the first quarter to$ 817.57 million in the final quarter. How well does a lineartrend fit these data? Are the residuals from this fit random?
Time Series Analysis and Forecasting – p. 63/115
![Page 70: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/70.jpg)
Exercise: Reebok (cont.)
Plot your time series
Fit a linear regression model and interpret your results
Find the residuals
Determine if the residuals are random
Time Series Analysis and Forecasting – p. 64/115
![Page 71: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/71.jpg)
Exponential Trend
In contrast to a linear trend, an exponential trend isappropriate when the time series changes by a constantpercentage (as opposed to a constant dollar amount) eachperiod. Then the appropriate regression equation is
Yt = c exp(bt)ut(19)
where c and b are constants, and ut represents amultiplicative error term. By taking logarithms of both sides,and letting a = ln(c) and εt = ln(ut), we obtain a linearequation that can be estimated by the usual linearregression method. However, note that the responsevariable is now the logarithm of Yt:
ln(Yt) = a + bt + εt(20)Time Series Analysis and Forecasting – p. 65/115
![Page 72: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/72.jpg)
Exercise: Intel
The file intel.txt contains quarterly sales data for the chipmanufacturing firm Intel from the beginning of 1986 throughthe second quarter of 1996. Each sales value is expressedin millions of dollars. Check that an exponential trend fitsthese sales data fairly well. Then estimate the relationshipand interpret it.
Time Series Analysis and Forecasting – p. 66/115
![Page 73: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/73.jpg)
Exercise: Intel (cont)
Plot your time series
If your original time series shows an exponential trend,apply ln to the series
Use the transformed series to estimate the relationship
Express the estimated relationship in the original scale
Find an estimate of the standard deviation
Time Series Analysis and Forecasting – p. 67/115
![Page 74: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/74.jpg)
Moving Averages
Perhaps the simplest and one of the most frequently usedextrapolation methods is the method of moving averages. Toimplement the moving averages method, we first choose aspan, the number of terms in each moving average. Let’ssay the data are monthly and we choose a span of 6months. Then the forecast of next month’s value is theaverage of the most recent 6 month’s values. For example,we average January-June to forecast July, we averageFebruary-July to forecast August, and so on. This procedureis the reason for the term moving averages.
Time Series Analysis and Forecasting – p. 68/115
![Page 75: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/75.jpg)
Exercise (toy example)
Suppose that Y (t) = [4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15].
Plot your time series
Write a function in R that finds a one-step-ahead movingaverage forecast for a span=2
Write a function in R that does a moving average"smoothing" to a time series
Time Series Analysis and Forecasting – p. 69/115
![Page 76: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/76.jpg)
Exercise: Dow Jones revisited
We again look at the Dow Jones monthly data from January1988 through March 1992. (See the file dow.txt). How welldo moving averages track this series when the span is 3months; when the span is 12 months?
Time Series Analysis and Forecasting – p. 70/115
![Page 77: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/77.jpg)
Exponential Smoothing
There are two possible criticisms of the moving averagesmethod. First, it puts equal weight on each value in a typicalmoving average when making a forecast. Many peoplewould argue that if next month’s forecast is to be based onthe previous 12 months’ observations, then more weightought to be placed on the more recent observations. Thesecond criticism is that moving averages method requires alot of data storage. This is particularly true for companiesthat routinely make forecasts of hundreds or even thousandsof items. If 12-month moving averages are used for 1000items, then 12000 values are needed for next month’sforecasts. This may or may not be a concern consideringtoday’s relatively inexpensive computer storage capabilities.
Time Series Analysis and Forecasting – p. 71/115
![Page 78: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/78.jpg)
Exponential Smoothing (cont.)
Exponential smoothing is a method that addresses both ofthese criticisms. It bases its forecasts on a weighted averageof past observations, with more weight put on the morerecent observations, and it requires very little data storage.There are many versions of exponential smoothing. Thesimplest is called simple exponential smoothing. It isrelevant when there is no pronounced trend or seasonality inthe series. If there is a trend but no seasonality, then Holt’smethod is applicable. If, in addition, there is seasonality,then Winter’s method can be used.
Time Series Analysis and Forecasting – p. 72/115
![Page 79: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/79.jpg)
Simple Exponential Smoothing
We now examine simple exponential smoothing in somedetail. We first introduce two new terms. Every exponentialmodel has at least one smoothing constant, which is alwaysbetween 0 and 1. Simple exponential smoothing has asingle smoothing constant denoted by α. The second newterm is Lt, called the level of the series at time t. This valueis not observable but can only be estimated. Essentially, it iswhere we think the series would be at time t if there were norandom noise.
Time Series Analysis and Forecasting – p. 73/115
![Page 80: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/80.jpg)
Simple Exponential Smoothing (cont.)
Then the simple exponential smoothing method is definedby the following two equations, where Ft+k is the forecast ofYt+k made at time t:
Lt = αYt + (1 − α)Lt−1(21)
Ft+k = Lt(22)
Time Series Analysis and Forecasting – p. 74/115
![Page 81: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/81.jpg)
Example:Exxon
The file exxon.txt contains data on quarterly sales (inmillions of dollars) for the period from 1986 through thesecond quarter of 1996. Does a simple exponentialsmoothing model track these data well? How do theforecasts depend on the smoothing constant α?
Time Series Analysis and Forecasting – p. 75/115
![Page 82: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/82.jpg)
Example: Exxon (cont.)
Simple Exponential Smoothing with α = 0.01
Holt−Winters filtering
Time
Obs
erve
d / F
itted
1986 1990 1994 1998
2000
025
000
3000
0
obsforecasts
Time Series Analysis and Forecasting – p. 76/115
![Page 83: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/83.jpg)
Example: Exxon (R code)
exxon<-read.table(file="exxon.txt",header=TRUE)sales<-exxon$Salesnew.sales<-ts(sales,start=c(1986,1),end=c(1996,2),freq=4)mod1<-HoltWinters(new.sales,alpha=0.1,beta=FALSE,gamma=FALSE)plot(mod1,xlim=c(1986,1998))lines(predict(mod1,n.ahead=6),col="red")legend(1992,20000,c("obs","forecasts"),col=c("black","red"),lty=c(1,1),bty="n")
Time Series Analysis and Forecasting – p. 77/115
![Page 84: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/84.jpg)
Exxon (changingα)
The following lines of code produce a "movie" that illustratesthe effect of changing α.n<-100for (i in 1:n){
mod1<-HoltWinters(new.sales,alpha=(1/n)*i,beta=FALSE,gamma=FALSE)plot(mod1,xlim=c(1986,1997))Sys.sleep(0.3)
}
Time Series Analysis and Forecasting – p. 78/115
![Page 85: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/85.jpg)
Choosingα.
What value of α should we use? There is no universallyaccepted answer to this question. Some practitionersrecommend always using a value around 0.1 or 0.2. Othersrecommend experimenting with different values of α until ameasure such as the Mean Square Error (MSE) isminimized.R can find this optimal value of α as follows:mod2<-HoltWinters(new.sales,beta=FALSE,gamma=FALSE)
Time Series Analysis and Forecasting – p. 79/115
![Page 86: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/86.jpg)
Holt’s Model for Trend
The simple exponential smoothing model generally workswell if there is no obvious trend in the series. But if there is atrend, then this method consistently lags behind it. Holt’smethod rectifies this by dealing with trend explicitly. Inaddition to the level of the series, Lt, Holt’s method includesa trend term, Tt, and a corresponding smoothing constant β.The interpretation of Lt is exactly as before.
Time Series Analysis and Forecasting – p. 80/115
![Page 87: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/87.jpg)
Holt’s Model for Trend (cont.)
The interpretation of Tt is that it represents an estimate ofthe change in the series from one period to the next. Theequations for Holt’s model are as follows:
Lt = αYt + (1 − α)(Lt−1 + Tt−1)(23)
Tt = β(Lt − Lt−1) + (1 − β)Tt−1(24)
Ft+k = Lt + kTt(25)
Time Series Analysis and Forecasting – p. 81/115
![Page 88: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/88.jpg)
Example: Dow Jones revisited
We return to the Dow Jones data (see file dow.txt). Again,these are average monthly closing prices from January 1988through March 1992. Recall that there is a definite upwardtrend in this series. In this example we investigate whethersimple exponential smoothing can capture the upward trend.Then we see whether Holt’s exponential smoothing methodcan make an improvement.
Time Series Analysis and Forecasting – p. 82/115
![Page 89: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/89.jpg)
Example: Dow Jones (cont.)
Simple Exponential Smoothing with optimal α
Holt−Winters filtering
Time
Obs
erve
d / F
itted
1988 1990 1992 1994
2000
2500
3000
3500
Time Series Analysis and Forecasting – p. 83/115
![Page 90: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/90.jpg)
Example: Dow Jones (R code)
dow<-read.table(file="dow.txt",header=TRUE)index<-dow$DowDJI<-ts(index,start=c(1988,1),end=c(1992,3),freq=12)mod1<-HoltWinters(DJI,beta=FALSE,gamma=FALSE)
### predictionsplot(mod1,xlim=c(1988,1994),ylim=c(1800,3700))preds<-predict(mod1,n.ahead=12)lines(preds,col="red")
Time Series Analysis and Forecasting – p. 84/115
![Page 91: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/91.jpg)
Dow Jones (Holt’s Model)
Holt’s Model with optimal α and β.
Holt−Winters filtering
Time
Obs
erve
d / F
itted
1988 1990 1992 1994
2000
2500
3000
3500
Time Series Analysis and Forecasting – p. 85/115
![Page 92: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/92.jpg)
R code (Holt’s Model)
mod2<-HoltWinters(DJI,gamma=FALSE)#Fitting Holt’s model
### predictions
plot(mod2,xlim=c(1988,1994),ylim=c(1800,3700))preds<-predict(mod2,n.ahead=12)lines(preds,col=’red’)
Time Series Analysis and Forecasting – p. 86/115
![Page 93: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/93.jpg)
Winter’s Model
So far we have said practically nothing about seasonality.Seasonality is defined as the consistent month-to-month (orquarter-to-quarter) differences that occur each year. Forexample, there is seasonality in beer sales - high in thesummer months, lower in other months.How do we know whether there is seasonality in a timeseries? The easiest way is to check whether a plot of thetime series has a regular pattern of ups and /or downs inparticular months or quarters.
Time Series Analysis and Forecasting – p. 87/115
![Page 94: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/94.jpg)
Winter’s Model (cont.)
Winter’s exponential smoothing model is very similar toHolt’s model- it again has a level and a trend terms andcorresponding smoothing constants α and β- but it also hasseasonal indexes and a corresponding smoothing constantγ. This new smoothing constant γ controls how quickly themethod reacts to perceived changes in the pattern ofseasonality.
Time Series Analysis and Forecasting – p. 88/115
![Page 95: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/95.jpg)
Example: Coke
The data in the coke.txt file represent quarterly sales (inmillions of dollars) for Coca Cola from quarter 1 of 1986through quarter 2 of 1996. As we might expect, there hasbeen an upward trend in sales during this period, and thereis also a fairly regular seasonal pattern, as shown in the nextslide. Sales in the warmer quarters, 2 and 3, areconsistently higher than in colder quarters, 1 and 4. Howwell can Winter’s method track this upward trend andseasonal pattern?
Time Series Analysis and Forecasting – p. 89/115
![Page 96: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/96.jpg)
Coke sales with forecasts
Holt−Winters filtering
Time
Obs
erve
d / F
itted
1986 1990 1994 1998
1000
3000
5000
Time Series Analysis and Forecasting – p. 90/115
![Page 97: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/97.jpg)
R code
mod.coke<-HoltWinters(sales)# Holt-Winter’s model
### predictions
plot(mod.coke,ylim=c(1000,6000),xlim=c(1986,1998))
preds<-predict(mod.coke,n.ahead=6)
lines(preds,col=’red’)
Time Series Analysis and Forecasting – p. 91/115
![Page 98: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/98.jpg)
ARIMA-Models
Forecasting based on ARIMA (autoregressive integratedmoving averages) models, commonly known as theBox-Jenkins approach, comprises the following stages:
Model Identification
Parameter estimation
Diagnostic checking
Iteration
These stages are repeated until a "suitable" model for thegiven data has been identified.
Time Series Analysis and Forecasting – p. 92/115
![Page 99: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/99.jpg)
Identification
Various methods can be useful for identification:
Time Series Plot
ACF / Correlogram
PACF (Partial Autocorrelation Function)
Test for White Noise
Time Series Analysis and Forecasting – p. 93/115
![Page 100: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/100.jpg)
Important Processes
White Noise
Moving Average Process (MA)
Autoregressive Process (AR)
Time Series Analysis and Forecasting – p. 94/115
![Page 101: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/101.jpg)
White Noise
White Noise. A stationary time series for which Y (t) andY (t + k) are uncorrelated, i.e., E[Y (t) − µ][Y (t + k) − µ] = 0
for all integers k 6= 0, is called white noise. Such process issometimes loosely termed a "purely random process".
Time Series Analysis and Forecasting – p. 95/115
![Page 102: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/102.jpg)
Moving Average Process
Moving Average Process. A moving average process oforder q, denoted MA(q), is defined by the equation
Y (t) = µ + β0εt + β1εt−1 + ... + βqεt−q(26)
where εt is a white noise process.
Time Series Analysis and Forecasting – p. 96/115
![Page 103: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/103.jpg)
Autoregressive Process
Consider a time series Y (t) that satisfies the differenceequation
Y (t) = α1Y (t − 1) + α2Y (t − 2) + ... + αpY (t − p) + εt(27)
where εt is a white noise process with zero mean and finitevariance σ2. The time series Y (t) is called an autoregressiveprocess of order p and is denoted AR(p).
Time Series Analysis and Forecasting – p. 97/115
![Page 104: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/104.jpg)
Analysis of ACF and PACF
A first step in analyzing time series is to examine theautocorrelations (ACF ) and partial autocorrelations (PACF ).R provides the functions acf() and pacf() for computing andplotting ACF and PACF. The order of "pure" AR and MA
processes can be identified from the ACF and PACF .
Time Series Analysis and Forecasting – p. 98/115
![Page 105: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/105.jpg)
Simulating AR(1) and MA(1)
#simulation of an AR(1) process
sim.ar<-arima.sim(list(ar=c(0.7)),n=1000)
#simulation of a MA(1) process
sim.ma<-arima.sim(list(ma=c(0.6)),n=1000)
Time Series Analysis and Forecasting – p. 99/115
![Page 106: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/106.jpg)
ACF and PACF for our simulations
# setting for a "matrix" of plots
par(mfrow=c(2,1))
## AR process
acf(sim.ar,main="ACF of AR(1) process")
pacf(sim.ar, main="PACF of AR(1) process")
Time Series Analysis and Forecasting – p. 100/115
![Page 107: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/107.jpg)
ACF and PACF for our simulations
# setting for a "matrix" of plots
par(mfrow=c(2,1))
## MA process
acf(sim.ma,main="ACF of MA(1) process")
pacf(sim.ma, main="PACF of MA(1) process")
Time Series Analysis and Forecasting – p. 101/115
![Page 108: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/108.jpg)
Simulating AR(2) and MA(2)
#simulation of an AR(2) process
sim.ar<-arima.sim(list(ar=c(0.4,0.4)),n=1000)
#simulation of a MA(2) process
sim.ma<-arima.sim(list(ma=c(0.6,-0.4)),n=1000)
Time Series Analysis and Forecasting – p. 102/115
![Page 109: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/109.jpg)
ACF and PACF for our simulations
# setting for a "matrix" of plots
par(mfrow=c(2,1))
## AR process
acf(sim.ar,main="ACF of AR(2) process")
pacf(sim.ar, main="PACF of AR(2) process")
Time Series Analysis and Forecasting – p. 103/115
![Page 110: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/110.jpg)
ACF and PACF for our simulations
# setting for a "matrix" of plots
par(mfrow=c(2,1))
## MA process
acf(sim.ma,main="ACF of MA(2) process")
pacf(sim.ma, main="PACF of MA(2) process")
Time Series Analysis and Forecasting – p. 104/115
![Page 111: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/111.jpg)
Simulating AR(3) and MA(3)
#simulation of an AR(3) process
sim.ar<-arima.sim(list(ar=c(0.4,0.3,0.2)),n=1000)
#simulation of a MA(3) process
sim.ma<-arima.sim(list(ma=c(0.6,-0.4,0.5)),n=1000)
Time Series Analysis and Forecasting – p. 105/115
![Page 112: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/112.jpg)
ACF and PACF for our simulations
# setting for a "matrix" of plots
par(mfrow=c(2,1))
## AR process
acf(sim.ar,main="ACF of AR(3) process")
pacf(sim.ar, main="PACF of AR(3) process")
Time Series Analysis and Forecasting – p. 106/115
![Page 113: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/113.jpg)
ACF and PACF for our simulations
# setting for a "matrix" of plots
par(mfrow=c(2,1))
## MA process
acf(sim.ma,main="ACF of MA(3) process")
pacf(sim.ma, main="PACF of MA(3) process")
Time Series Analysis and Forecasting – p. 107/115
![Page 114: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/114.jpg)
Important conclusions from simulations
We know that for the moving average model MA(q) thetheoretical ACF has the property that ρh = 0 for h > q, sowe can expect the sample ACF rh ≈ 0 when h > q.
If the process Y (t) is really AR(p), then there is nodependence of Y (t) on Y (t − p − 1), Y (t − p − 2), ... onceY (t − 1), ..., Y (t − p) have been taken into account. Thuswe might expect that αh ≈ 0 for h > p, i.e. that the PACF
cuts off beyond p.
Time Series Analysis and Forecasting – p. 108/115
![Page 115: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/115.jpg)
Behavior of Theoretical ACF and PACF
MA(q): ACF cuts off after lag q.
MA(q): PACF shows exponential decay and/ordampened sinusoid.
AR(p): ACF shows exponential decay and/or dampenedsinusoid.
AR(p): PACF cuts off after lag q.
ARMA(p, q):ACF shows exponential decay and/ordampened sinusoid.
ARMA(p, q):PACF shows exponential decay and/ordampened sinusoid.
Time Series Analysis and Forecasting – p. 109/115
![Page 116: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/116.jpg)
Fitting
Suppose that the observed series has, if necessary, beendifferenced to reduce it to stationarity and that we wish to fitan ARMA(p, q) model with known p and q to the stationaryseries. Fitting the model then amounts to estimating thep + q + 1 parameters, the coefficients for the AR(p) process,the coefficients for the MA(q) process and σ2.
The three main methods of estimation are (Gaussian)maximum likelihood, least squares and solving theYule-Walker equations. (We are not going to talk aboutthese methods in detail, Sorry!)
Time Series Analysis and Forecasting – p. 110/115
![Page 117: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/117.jpg)
Diagnostics
Having estimated parameters we need to check whether themodel fits, and, if it doesn’t, detect where it is going wrong.Specific checks could include:
Time Plot of Residuals. A plot of residuals, et, against t
should not show any structure. Similarly for a plot of et
against Y (t).
ACF and PACF of Residuals. The hope is that these willshow the White Noise pattern: approximatelyindependent, mean zero, normal values, each withvariance n−1.
Time Series Analysis and Forecasting – p. 111/115
![Page 118: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/118.jpg)
Diagnostics (cont.)
The Box-Pierce (and Ljung-Box) test examines the Nullof independently distributed residuals. It’s derived fromthe idea that the residuals of a "correctly specified"model are independently distributed. If the residuals arenot, then they come from a miss-specified model.
Time Series Analysis and Forecasting – p. 112/115
![Page 119: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/119.jpg)
Summary of R commands
arima.sim() was used to simulate ARIMA(p, d, q) models
acf() plots the Autocorrelation Function
pacf() plots the Partial Autocorrelation Function
arima(data, order = c(p, d, q)) this function can be used toestimate the parameters of an ARIMA(p, d, q) model
tsdiag() produces a diagnostic plot of fitted time seriesmodel
predict() can be used for predicting future values of thelevels under the model
Time Series Analysis and Forecasting – p. 113/115
![Page 120: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/120.jpg)
Exercises
Fit an ARIMA model to ts1.txt
Fit an ARIMA model to ts2.txt
Fit an ARIMA model to ts3.txt
Fit an ARIMA model to ts4.txt
Time Series Analysis and Forecasting – p. 114/115
![Page 121: Time Series Analysis and Forecasting - University ofmath.unm.edu/~alvaro/time series.pdf · Time Series Analysis and Forecasting Math 667 Al Nosedal Department of Mathematics Indiana](https://reader034.vdocuments.net/reader034/viewer/2022051405/5a71d9ac7f8b9aa2538d2fc8/html5/thumbnails/121.jpg)
THANK YOU!!!
Time Series Analysis and Forecasting – p. 115/115