Suppose I have a non-stationary time series and have obtained forecasts using various methods such as ARIMA, ETS, Theta etc. I want to find a weighted combination of these forecasts.

I found in the forecastcomb package that one of its methods regress the target variables against its forecasts which they name as OLS.

If the series are non-stationary, can we apply OLS?

  • $\begingroup$ If you use OLS the standard error won't be right probably because of autocorrelation. OLS, without special modifications, is not used for time series for that reason. Also non-stationary data will generate incorrect R squared results - this is called spurious regression. In general its a bad idea to do OLS regression against non-stationary data, but maybe there is something special going on here. I would just use substantive judgement in weighting. $\endgroup$ – user54285 Feb 8 at 17:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.