Expanding window regression

[manta]

Dear E-views users,
I'm struggling to find a suitable way to do the following

I have to estimate different models (AR(1), multiple linear regression, simple linear regression), using an expanding window regression.
After that, I want to compare the out-of-sample one step ahead forecast among all these models to check whether is the best one in predicting.
I'm using monthly observations. Let me explain how I would like the expanding window regression to work: let's say I want to forecast July with observations from January to June, keep this one step ahead forecast in a series. Than, forecast August using observation from January to July, and add the latter forecast to the same series created before. I've checked the 'roll' examples, but it looks to me that the window size is set to 20, and therefore it is does not allow to change the size. Finally, I want this loop to stop at a certain month, let's say Decemeber, and then start again with the next year (i.e. again July with observations from January up to June etc). I'm struggling since a couple of days but up to now I had no luck.

Thanks
Marco

[manta]

Also, in the EViews language, the procedure described is a 'with coefficients update' or 'without'?

[EViews Gareth]

If you do a search for rolling regression, you should find lots of examples on how to do this.

[manta]

Ok, i changed the mechanic of my analysis, it is now slightly different.
I need to estimate a recursive model for a time series data of asset returns. The dependent variable is the asset return and then I have a set of k variables, a lagged value of the dependent variable (plus an intercept) as regressors. My sample period (monthly observations) starts on Jan 1972. What I need to do is the following:

1)use a moving window regression (window of 60 observations, i.e. 5 years)
2)estimate all the possible model (Jan 1972 Dec 1977) using a subset of the k variables (intercept and lagged values always present) and choose the best model according to thee AIC criterion
3)once the best model is chosen, make one-step ahead prediction with that model
4)go back to step 2 shifting the sample period one month ahead (i.e. Feb 1972, Jan 1978) and then repeat step 2 and 3
5)keep going until the end of the sample (May 2009)

Hope it helps

[mjfl]

hi manta,

have u found how to do it yet?