2102 final exam pass session

37
ACTL2102 Final Exam PASS Session ACTL2102 Final Exam PASS Session October 30, 2014 1/37

Upload: bob

Post on 16-Feb-2016

21 views

Category:

Documents


3 download

DESCRIPTION

.

TRANSCRIPT

Page 1: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

October 30, 2014

1/37

Page 2: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Plan

1 Lectures 1-2: Stochastic Processes and Discrete Time Markov ChainsStuff You Should KnowExample Questions

2 Lecture 3: Exponential Distribution and the Poisson ProcessStuff You Should KnowExample Questions

3 Lectures 4-5: Continuous Time Markov ChainsStuff You Should KnowExample Questions

4 Lectures 6-8: Time Series MathematicsStuff You Should KnowExample Questions

5 Lectures 8-10: Model Selection, Checking, and PredictionStuff You Should KnowExample Questions

6 Lectures 11-12: Pseudo-Random Continuous WalksStuff You Should KnowExample Questions

2/37

Page 3: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 1-2: Stochastic Processes and Discrete Time Markov Chains

Stuff You Should Know

Stuff You Should Know

Definitions:

Independent IncrementsStationary IncrementsMarkov Process

Transition Matrices

State Classifications

Mean Time in Transient States

Periodicity

Limiting Probabilities and their Conditions

Time Reversibility

Gambler’s Ruin

Branching Processes

3/37

Page 4: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 1-2: Stochastic Processes and Discrete Time Markov Chains

Example Questions

PASS 3-4 (modified)

Specify the classes of the following Markov chain, and determinefor each class whether they are transient or recurrent:

1

13

23 0 0

12 0 1

2 00 0 1

434

0 0 12

12

2 Find the probability that the process will return to state 4

eventually if the process starts at state 4.

3 Find the expected number of times state 4 is visited if theprocess starts at state 4.

4/37

Page 5: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 1-2: Stochastic Processes and Discrete Time Markov Chains

Example Questions

Tutorial 2-2

An unfeasibly large organisation has N employees. Each employeehas one one of three possible job classifications and changesclassification (independently) according to a Markov chain withtransition probabilities. 0.7 0.2 0.1

0.2 0.6 0.20.1 0.4 0.5

1 What percentage of employees are in each classification?

2 Is this process time-reversible?

5/37

Page 6: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 1-2: Stochastic Processes and Discrete Time Markov Chains

Example Questions

PASS 4-4

For the following branching process, calculate the probability ofextinction when the number Y of offspring of each individualfollows the following distribution:

1 P(Y = 0) = 13 ,P(Y = 1) = 1

2 ,P(Y = 2) = 16

2 Y ∼ Bin(2, 0.6)

For both questions, assume the starting population X0 = 1

6/37

Page 7: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 1-2: Stochastic Processes and Discrete Time Markov Chains

Example Questions

PASS 4-6

Han Solo moves among n + 1 star systems that are arranged in acircle. At each system it moves either to the next system in theclockwise direction with probability p or the counterclockwisedirection with probability q = 1− p. Starting at a specifiedsystem, call it system 0, find the probability that all star systemshave been visited before revisiting system 0.

7/37

Page 8: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lecture 3: Exponential Distribution and the Poisson Process

Stuff You Should Know

Stuff You Should Know

Exponential Distribution: Properties and Results

Definition of Counting and Poisson Process

Interarrival and Waiting Times

Sum and Thinning of Poisson Process

Non-Homogenous and Compound Poisson

8/37

Page 9: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lecture 3: Exponential Distribution and the Poisson Process

Example Questions

PASS 5-3

The average number of automobiles entering a mountain tunnelper minute period is 1. Excessive number of cars entering thetunnel during a brief period of time produces a hazardoussituation. Assuming the Poisson process:

1 Find the probability that the number of automobiles enteringthe tunnel during a 1 minute period exceeds 2?

2 What is the probability that the number of automobilesentering the tunnel will be less than 3 during a 3-minuteperiod?

3 What is the probability that no car enters the tunnel duringthe first 4 minutes?

9/37

Page 10: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lecture 3: Exponential Distribution and the Poisson Process

Example Questions

PASS 5-6

An insurance portfolio contains policies for three categories ofpolicyholder: A, B and C. The number of claims in a year, N, onan individual policy follows a Poisson distribution with mean λ.Individual claim sizes are assumed to be exponentially distributedwith mean 4 and are independent from claim to claim. Thedistribution of λ, depending on the category of the policyholder, is:

Category Value of λ Proportion of policyholders

A 2 20%

B 3 60%

C 4 20%

Denote by S the total amount claimed by a policyholder in oneyear.

10/37

Page 11: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lecture 3: Exponential Distribution and the Poisson Process

Example Questions

PASS 5-6 continued

1 Prove that E (S) = E [E (S |λ)].

2 Show that E (S |λ) = 4λ and Var(S |λ) = 32λ.

3 Calculate E (S).

4 Calculate Var(S).

11/37

Page 12: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lecture 3: Exponential Distribution and the Poisson Process

Example Questions

Tutorial 3-6

Events occur according to a nonhomogenoous Poisson processwhose mean value function is given by

m(t) = t2 + 2t, t ≥ 0

What is the probability that n events occur between time t = 4and t = 5?

12/37

Page 13: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lecture 3: Exponential Distribution and the Poisson Process

Example Questions

Tutorial 3-9

Insurer A has a combined home insurance and landlords insuranceportfolio. The total number of claims for this portfolio is modelledusing a Poisson process with expected claims 300 per year. Theproportion of landlords insurance claims was 1/5 of the overallclaims. Insurer A sells its home insurance portfolio to insurer B.Insurer B specialises in home insurance and has no landlordsinsurance policies. The expected number of claims for the oldportfolio of insurer B was 120.

1 Define the processes of the number of claims from insurer Afor both the home insurance portfolio and landlords insuranceportfolio before the takeover.

2 Define the processes of the number of claims of both insurerA and insurer B before and after the takeover of the homeinsurance by insurer B.

13/37

Page 14: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 4-5: Continuous Time Markov Chains

Stuff You Should Know

Stuff You Should Know

Definitions and First Principles

The ”Q” (Generator) Matrix and the embedded Markov Chain

Kolmogorov Equations and the ”P(s,t)” transition matrix

Limiting Probabilities

Time Reversibility

Birth-Death Processes (rates, structure, mean time,Kolmogorov equations, balance equations)

HSD models, and computing probabilities using integrals

(Maybe) Calculation methods

14/37

Page 15: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 4-5: Continuous Time Markov Chains

Example Questions

PASS 6-1

A 24 hour convenience store has three cashiers, Ralph, Vincent,Benjamin. Customers arrive according to a Poisson process withrate λ per hour and join the queue if there are already 3 customersserved. Each cashier serves customers with service timesexponential with mean 1

µ per hour.

1 Define a Markov chain to model the number of customers inthe store and write down the corresponding generator matrix(the instantaneous transition rate matrix).

2 Given that there is n (n > 3) customers in the store, determinean expression for the conditional probability that the numberof customers will remain unchanged over the next 15 minutes.

15/37

Page 16: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 4-5: Continuous Time Markov Chains

Example Questions

PASS 6-5 (modified)

Consider the following probability transition rate matrix with state1, 2, 3, 4:

Q =

−0.7 0.4 0.3 00.1 −1 0.3 0.60.1 1.9 −2.2 0.21.7 0.3 0.5 −2.5

1 Find the embedded probability matrix.

2 Find the probability that the fifteen transition will be intostate 3, given we started in state 1 and the thirteenthtransition was into state 2.

3 If possible, find the long-run proportion of time spent in eachstate.

4 Is this process time-reversible?16/37

Page 17: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 4-5: Continuous Time Markov Chains

Example Questions

PASS 7-1

Consider the birth and death process with λi = λ for i = 0, 1, 2, ...and µj = µ for j = 1, 2, 3, .... Denote Ti as the time spent in statei before moving to state i + 1. Find E[Ti ].

17/37

Page 18: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 4-5: Continuous Time Markov Chains

Example Questions

PASS 7-7

Consider a health (H), sickness (S) and death (D) model for anindividual aged x > 0 with the following rates:

The rate at which a healthy individual becomes sick is 0.001x

The rate at which a sick individual recovers is 0.002x

The rate at which a healthy individual dies is 0.001x

The rate at which a sick individual dies is 0.002x

1 Give an expression for the probability that a sick 65 year oldindividual stays sick for at least 1 year and then becomeshealthy and remain so till age 67.

2 Find the probability that a sick 65 year old individual remainssick until he dies.

18/37

Page 19: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 4-5: Continuous Time Markov Chains

Example Questions

Tutorial 5-1 (modified)

Consider two machines both with an exponential lifetime 1/λ.

1 There is a single repairdrone that can service machines at arate µ. Set up the Kolmogorov backward equations. Also, setup the forward equations.

2 Consider instead the case where the inevitable automation ofall our jobs has not yet occurred and the repairdrone isinstead a repairhuman. This repairhuman repairs at a rateµ(t), where t is the time since breakfast. Set up theKolmogorov backward and forward equations.

19/37

Page 20: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 6-8: Time Series Mathematics

Stuff You Should Know

Stuff You Should Know

Classical decomposition model

Removing deterministic seasonality and deterministic trends

Integrated Time Series

SARIMA models (and their subsets)

Causality and Invertibility

Calculating ACF:

Linear Filter MethodYule-Walker Equations

Calculating PACF

20/37

Page 21: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 6-8: Time Series Mathematics

Example Questions

PASS 8-1

Find a filter of the form 1 + αB + βB2 + γB3 that passes lineartrends without distortion and that eliminates arbitrary seasonalcomponents of period 2.

21/37

Page 22: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 6-8: Time Series Mathematics

Example Questions

PASS 10-6

Compute the ACF and the PACF for the following AR(2) processXt = 0.6Xt−1 + 0.2Xt−2 + Zt where Zt ∼W .N(0, σ2).

22/37

Page 23: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 6-8: Time Series Mathematics

Example Questions

Tutorial 8-3

Consider the AR(2) process {Xt} satisfying:

Xt − φXt−1 − φ2Xt−2 = Zt

For what values ofφ is this a causal process?

23/37

Page 24: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 6-8: Time Series Mathematics

Example Questions

UKCT6 10/07-10 (modified)

The time series Xt is assumed to be stationary and to follow anARMA(2,1) process defined by:

Xt = 1 +8

15Xt−1 −

1

15Xt−2 + Zt −

1

7Zt−1

1 Determine the roots of the characteristic polynomial, andexplain how their values relate to the stationarity of theprocess.

2 Find the ACF for lags 0, 1, and 2.

3 Determine the mean and variance of Xt .

24/37

Page 25: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 8-10: Model Selection, Checking, and Prediction

Stuff You Should Know

Stuff You Should Know

Model selection

Looking at picturesAIC, BIC (AKA SBC), AICc

Model parameter estimation (finding and using SACF andSPACF)

Residual Analysis: Portmanteau/Ljung-Box

Testing for non-stationarity

Looking at the diagramDickey-Fuller and Augmented Dickey-Fuller tests

Cointegrated Time Series

Markov Property

Forecasting

Box-JenkinsBest Linear Predictor

25/37

Page 26: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 8-10: Model Selection, Checking, and Prediction

Example Questions

Mock Exam Q6 & Q7

No. I can’t be bothered typing them out. Just... open the .pdf foryourself.

26/37

Page 27: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 8-10: Model Selection, Checking, and Prediction

Example Questions

PASS 12-5

Consider the following 2 time series:

Xt = 2t2 + 3 + Zt

Yt = 3(t − 1)2 + Zt

1 We say that (Xt ,Yt) are integrated of order d . Find d .

2 Are Xt and Yt cointegrated? If so, give the cointegrationvector. If not, give reasons why not.

27/37

Page 28: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 8-10: Model Selection, Checking, and Prediction

Example Questions

Tutorial 10-2 (Modified)

Explain briefly whether the following processes are Markov:

1 AR(4)

2 ARMA (1,1)

28/37

Page 29: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 8-10: Model Selection, Checking, and Prediction

Example Questions

Tutorial 10-3 (Modified)

Suppose {Xt} is a stationary time series with mean µ and ACFρ(.). Find the best linear predictor of Xn+h of the form aXn + b.

29/37

Page 30: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 8-10: Model Selection, Checking, and Prediction

Example Questions

Tutorial 10-4 (Modified)

Consider an ARIMA(1,1,1) model.

1 Use the Box-Jenkins apporach to derive the one-step andtwo-step ahead forecasts, assuming the parameter values areknown.

2 Evaluate the prediction variance Var( Xn+1 − x̂n(1)), assumingthe parameter values are known.

30/37

Page 31: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 11-12: Pseudo-Random Continuous Walks

Stuff You Should Know

Stuff You Should Know

Definitions of Brownian Motion

Properties of Brownian Motion

Stochastic Differential Equations (!)

Stochastic Integrals (!)

Monte Carlo Simulation

Linear Congruential Formula

Inverse-Transform (Discrete and Continuous)

Accept-Reject (Discrete and Continuous)

Variance Reduction Techniques

Importance Sampling

Number of Simulations

31/37

Page 32: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 11-12: Pseudo-Random Continuous Walks

Example Questions

Lecture 11-30

Evaulate E[B8t ], where Bt is standard Brownian Motion.

32/37

Page 33: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 11-12: Pseudo-Random Continuous Walks

Example Questions

Bonus Question

Consider the geometric Brownian Motion

Y (t) = exp(φt + ψXt)

where Xt has the stochastic differential equation

dXt = µdt + σdBt

and where Bt is standard Brownian motion.Express dYt as a stochastic differential equation.

33/37

Page 34: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 11-12: Pseudo-Random Continuous Walks

Example Questions

Another Bonus Question

1 Consider the linear congruential formula. Where a = 11,c = 37, m = 100 and x0 = 85, generate 3 random numberson U[0, 1].

2 Use these numbers to generate samplings from a Bin(3, 0.5)and an exponential distribution with mean 0.5.

3 If possible, use the exponential random variable samplesobtained to generate samplings from a Gamma(2, 3) randomvariable. Use the Acceptance-Rejection method.

34/37

Page 35: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 11-12: Pseudo-Random Continuous Walks

Example Questions

PASS 13-5

Suppose you are simulating a set or normally distributed randomvariables with mean 65 and standard deviation 15. Find thenumber of simulations required so that you will be in a 0.5% bandof the true value, 95% of the time.

35/37

Page 36: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 11-12: Pseudo-Random Continuous Walks

Example Questions

Lazy Questions:

1 Explain how using antithetical variates reduces estimatevariance.

2 Explain how using control variates reduces estimate variance.

3 Explain how importance sampling can increase computationalefficiency for determining expectations.

36/37

Page 37: 2102 Final Exam PASS Session

ACTL2102 Final Exam PASS Session

Lectures 11-12: Pseudo-Random Continuous Walks

Example Questions

WE’RE DONE HERE!

GOOD LUCK!

37/37