Search code examples
forecasting

TSLM and Stationary Data

I’m trying to use TSLM to create a model, do both time series need to be stationary before we use the TSLM, i.e do you need to do differencing to stationarize the TS. According to Rob Hyndman notes the TSLM should take care of that, is that correct?

Solution

  • For any TSLM model, the residuals should be white noise (and therefore also stationary). So just check that the residuals look like white noise (use checkresiduals()).

    If they are not white noise, you should try a dynamic regression model instead, using auto.arima() for example.

    If you do use auto.arima(), and the residuals are non-stationary, differencing will be applied before estimation.