In the course of our seminar paper, my group and I wanted to compare the forecasts of both the rw with drift and a VECM (including macroeconomic variables like unempl, exch... etc, 1 lag) model to the actual values of the S&P 500 close values for a certain period. Unfortunately, we have encountered some problems with our analysis.
Everything up to the forecasts themselves (Model specs, unit root tests, cointegration tests etc.) has already been done. We are struggling though with the "manual" computation of our forecasts. We have data for 1990M01 - 1999M12 for all our variables and would ideally like to forecast e.g. 1999M11 - 1999M12. We understand, we need to do one step ahead forecasts, moving the forecasting window one step further at a time. Thus, our approach was the following.
1.) Expand our workfile to 2000m02.
2.) Estimate Random Walk Model & VECM model for estimation period 1990M01 - 1999M10
3.) Forecast specification for RW
4.) Same for our VECM
Up to this stage, fine, for this period (M11) the RW, as expected beats the VECM. However, when doing the next step, adjusting the estimation sample to 1990m01 - 1999m11 and the forecasting window to 1999m12 2000m02, we find that the VECM beats our RW model.
Basically, we would like to forecast 1999m11 and 1999m12 with both our models and then compare it to the actual values, hopefully to find that the RW model beats our VECM in forecasting the stock index. Is there anything we are missing in specifying the forecasts / forecast evaluation?
Thank you very much in advance for any input!
For econometric discussions not necessarily related to EViews.
1 post • Page 1 of 1
Who is online
Users browsing this forum: Vermeiren and 2 guests