You are on page 1of 2

Subject 300 - Econometrics

Exercise 11 - Serial Correlation and Nonlinear


Time Series Models
A. Sancetta
December 2007

1. Consider a model with a single regressor and an MA(1) disturbance term


yt = βxt + ut , t = 1, . . . , T,
ut = εt + θεt−1 ,
where εt , t = 1, .., T, is a sequence of serially uncorrelated disturbances with
mean zero and variance σ 2 , and ε0 = 0. By formulating the appropriate LM
test, determine whether the Durbin-Watson test is appropriate for MA(1)
disturbances.
2. Regressing yt on yt−1 in a sample of size 25 yields a coefficient of φ̂ = 0.8 and
a standard error of 0.12. Within the framework of an AR(1) model test

(i) H0 : φ = 0.6 against H1 : φ > 0.6.


(ii) H0 : φ = 1 against H1 : φ < 1.

Would your conclusions have been different if the results had been from a
regression of yt on yt−1 and a constant term?

3. Show that the process


yt = εt + βεt−1 εt−2 , εt ∼ NID(0, σ 2 ), t = 1, ..., T
with ε0 and ε−1 fixed and known, is white noise but not strict white noise.
Compare the minimum mean square estimator of yT +1 with the minimum
mean square linear estimator.
4. Consider the GARCH model

yt = σ t εt , εt ∼ NID (0, 1)
σt = γ + αyt−1 + βσ 2t−1
2 2
γ > 0, α ≥ 0, β ≥ 0.

(i) Show that yt is a martingale difference and derive an expression for


its (unconditional) variance.

(ii) Find the autocorrelation function of (a) yt and (b) yt2 . Discuss.

(iii) Write down the log-likelihood function for the GARCH-M model

yt = δσ t + ut , t = 1, ..., T

in which δ is an unknown parameter and ut follows a GARCH(1,1) process.


State any assumptions you make.

You might also like