EViews.com

This thread is about the add-in DMtest. This add-in carries out Diebold and Mariano's (1995) test for the equality of forecast accuracy of two forecasts. The add-in requests the user to input 2 equation objects, the forecast sample, and to select a loss function (squared or absolute). The output is a spool that includes a table of results which displays the Diebold-Mariano test statistic and a p-value, and graphs of the two forecasts.

Please let me know if there are any questions.

- Angela Vossmeyer

thanks alot for this Add-in , how to explain the results?.
best regards.

Please provide option for modified dmtest

What is "modified dmtest"?

Hi, I'm transforming my dependent variable, let say "x", by taking the first difference, so it is equal to d(x,1). When I run the add-in, an error message appears: "Syntax error in "IF @OBSSMPL <> D(X,1).@OBS THEN".

Is there a way to overcome that error or It is not possible to work with dependente variables using functions like d(x,1).

Kind regards,

Should be able to fix that - stay tuned for an update.

Hi Annibal,

Thanks for bringing this to attention. The file has been updated and uploaded so this shouldn't be an issue anymore.

Let me know if there's anything else,

-Angela

Hi ecofin, if the p-value is above your desired significance level, the results indicate that the researcher cannot reject the hypothesis of equal expected forecast error, or forecast accuracy. Hope this helps! - Angela

Please tell how to run the addin as program line command

hi, thanks for your reply
is that I understood H0: equal expected forecast error, alternative H1: not equal expected error.
I have simple suggestion to Add line vertical and horizontal in graph to show the point of forecast101 intercept forecat201, and estimate this point in out put.

Best regards.

Your case is somewhat specific. How do you know the two forecasts are going to have an intercept? How do you know that they are going to have just one? They could cross 15,000 times!

hi
good question, I don't know but I have estime two models x c ar(1) the second x c ar(1) ar(2) for box-jenkins aire line passenger (just select 2 models for an example), I know They could cross 15,000 times. it is my mistake to understand that there are some models that can intercept in one point.

Best regards. ecofin

hi
thanks for this Add-in ,is this Add-in the Diebold and Mariano’s asymptotic test?, and can you write in the future an Add-in with four COMPARING FORECAST ERRORS OF DIFFERENT MODELS like:
1-Diebold and Mariano’s asymptotic test;
2-Diebold and Mariano’s sign test;
3-Diebold and Mariano’s Wilcoxon sign-rank test;
4-Serially correlated loss differentials.
the guide reference in book Chapter 2 A Practical Guide to Forecasting Financial Market Volatility, Ser-Huang Poon, John Wiley, pp 24-28.

Kind regards. metrix

Hi Metrix,
Yes, this add-in is for the asymptotic test. I will look into these other suggestions.
Thanks,
Angela