I am doing out of sample forecast for my daily exchange rate data. Here I am using different models like ARIMA(0,1,3), ARCH(3), GARCH(1,1), EGARCH(1,1) and TARCH(1,1) with 3 distributions (normal,t and GED).

I want to compare among them using RMSE and MAE to pick up the best model but I noticed that RMSE of ARIMA(0,1,3) is the lowest while the RMSE and MAE of seven other models(ARCH-t, ARCH-ged, GARCH(1,1)-t, GARCH-ged, EGARCH(1,1)-t, TARCH(1,1)-t and TARCH(1,1)-ged) are the same.

So the question here which one should I pick up?


1 Answer 1


The selection of the model should obviously be based on data. Divide your data in three parts. Since you have a time series data, you should make the selection like this: $[0,T_1]$ is the model you will use to estimate your model, $[T_1,T_2]$ you may use to choose the best hiperparameters for your model and $[T_2,T_3]$ you may use to select the model. Note that $T_1<T_2<T_3$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.