VAR Forecast Statistics

[goffbl]

I'm getting strange results using the VAR Forecast dialog. I'm using EViews 9.5 and estimating a simple VAR for teaching purposes using d(Fed Funds), d(3 month Treasury), d(10 year Treasury), d(Baa rate) for 1954.1 2015.12 with underlying rates in percent. Taking the in-sample, static forecasts for Fed Funds rate over 2006.1 to 2007.12 as an example, the RMSE, MAE, and MAPE generated by the Forecast dialog are 5.4, 5.4, 99.9 even though the actual forecast series is nearly identical to the actual series. When I compute these manually, they are 0.13, 0.10, 2.10, which make sense. I examined the results with a different data set and generated similar "glithcy" results. If the data are entered in levels in the VAR, the forecast stats make sense, but when entered as d(*) and having Eviews integrate back to levels, something is going haywire.

[EViews Gareth]

Could you post the workfile?

[goffbl]

I've attached it

[EViews Gareth]

And instructions on what you're doing...

[goffbl]

I've been generating these results from the VAR dialog windows. The commands would be

smpl 55.1 2015.12
var rates.ls 1 3 d(reff) d(rtb3m) d(rtb10) d(raaa) d(rbaa)

smpl 06.1 07.12
var rates.fit(g, e)

[EViews Gareth]

Ugh, looks like we're calculating it between the tranformed forecast and the untransformed actuals. Bad. We'll fix. Thanks for letting us know.

[goffbl]

Thanks for looking into it