Autocorrelation

Question 1

What is autocorrelation?

Answer 1

The disturbance term in one observation is related to the disturbance term in another

Question 2

What are the causes of autocorrelation?

Answer 2

1) Something in u that has an effect (genuine AR1 autocorrelation)
2) Omission of lagged variables

Question 3

What are the consequences of autocorrelation?

Answer 3

Inefficient coefficient estimates (not biased) and incorrect standard errors, meaning t/f-tests are invalid

Question 4

What is the test for autocorrelation?

Answer 4

The Durbin-Watson test: uses a lower and upper limit… below Du means we reject and above DL means we don’t reject… in between means the test is indeterminate. Vice versa for above values of 2.

Question 5

What is the OTHER test for autocorrelation?

Answer 5

A t-test on p (rho) or a joint F-test if more than one ‘p’

Question 6

What is the test for autocorrelation for a lagged dependent variable?

Answer 6

We Durbin’s h test: Use n (number of observations), the estimated variance of the coefficient of the lagged dependent variable and p (rho)=1-0.5d… Null hypothesis is p (rho)=o (no autocorrelation)