forecasting

[maxmarzo]

Dear All,
I am looking for help about forecasting.
I am estimating a SURE model with inst option. My variable is in first
difference. I need to get forecast for the variable in levels, but the model
is estimated in first difference for unit root issues.
Is there a way to convert the forecasted values (both in sample and out of
sample) from first differences to levels ?


Do you have worked examples for such problems ?

Is the command FORECAST good also for in-sample forecasts ?
Each time I try to use it, when I try to compare forecasted values with
actual values I get always the time series of the actual values, for in
sample forecasts.
Am I using the right command ?
thanks
max

[TomM]

Dear All,
I am estimating a SURE model with inst option. My variable is in first
difference. I need to get forecast for the variable in levels, but the model
is estimated in first difference for unit root issues. Is there a way to convert the forecasted values (both in sample and out of
sample) from first differences to levels ?

The easiest solution is to define identity equations relating the levels to the first differences and include those in your FORECAST command. This way, you get forecasts for both the differences and the levels.

The FORECAST section of the RATS 6 Reference Manual has a similar example using identities to get the "anti log". For a linear relationship though, you can use EQUATION rather than FRML.

For example, if DY is the first difference of Y, you would define something like this:

equation(identity) y_ident y
# dy y{1}

and include Y_IDENT in your FORECAST.

Is the command FORECAST good also for in-sample forecasts ?

Certainly. It sounds like you are making a mistake in how you are saving the forecasts. I would start by including the PRINT option on your FORECAST statement--this will show you the actual and forecasted values, and the results will not depend on how (or whether) you are saving the forecats.

Assuming that is working as expected, check how you are saving the forecasts into series, and how you are comparing those series to your original variables.

If you still have problems, I would suggest following up with a question to our support department at support@estima.com. Be sure to include your program, data, and RATS serial number.