Forecasting previous values

[subbuteo3]

Hello,

Running a really basic garch where my dependent is only regressed on time
y= c + @trend
and my variance is a function of of the autoregressive term and time as well
variance=c+arch+@trend

my sample is set from 1973-1997 but I only have data for y from 1984-1997. I want to "forecast" (backcast) variance and the fitted y to 1973 using my arch model that is computed using the 1984-1997 data.

When I press forecast I cant seem to get fitted values for y or get it to display these values in the graphical output (it will only show 1984-1997). However, it will compute fitted values for the variance back to 1973. What I really want is a graphical representation of y and including the confidence interval brackets representing the changing variance.

How do I get it to forecast these previous values?

[EViews Gareth]

Because in GARCH the Yt is dependent on Yt-1, if there is no value for Y0, you will not be able to forecast Y1 etc... Thus you'll get NAs.

[subbuteo3]

But my equation for y does not contain any autoregressive terms. My equation for variance does contain lagged values of the squared error term (which obviously does not exist before 1984) and yet Eviews produces fitted values for variance before 1984 but not for y.

[EViews Gareth]

Ah, yes, I see what you mean.

You can always calculate it manually:

Code: Select all

series yf = c(1)+c(2)*@trend

[subbuteo3]

Thanks.

I suppose I can make the confidence intervals using my variance forcast and plot them with the fitted ys to make the nice forecast graph.