Me and my colleague are writing our master thesis and have a few struggles in our econometric procedure. We want to use a dynamic linear model to explain a macroeconomic phenomenon, and are expecting to include lags. According to the AIC, 2 lags is suitable.
In order to check for autocorrelation in our regression model, we want to do a Breuch-Godfrey test. The test acquire to fill in lag order, and this is when we met insecurity. Should we:
1) Use a simple lm to this test and exclude the lags intended to use, or 2) should we include a model including the 2 lags we intend to use. Will the lagged model disturb and give wrong output (as the test requires you to specify lag order)?
Including the lagged model we get 95% significance up to 15 lags, which is a lot more than what the AIC expressed. With the same significance level, our basic linear model shows that 2 lags is suitable.
We have also done a Durbin-Watson test using the lagged model, which indicated no signs of autocorrelation.
We appreciate every answer we can get.
Why not make some simulations using your lagged model and check that the Breuch-Godfrey test distribution is still usable in that context?
Durbin-Watson test shows no autocorrelation then you do not have to worry about it
The number of lags depends on the kind of autocorrelation pattern you wish to test for. One lag if AR(1) errors and so on. Wooldridge's Introductory Econometrics is a very good reading.
Are you sure your time series are stationary? That goes before checking for autocorrelation.
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Writing