2

I am trying to predict how much revenue a store will generate in next month based on revenues of previous months. I was doing simple regression for forecasting before, but I have recently read about stationary and non-stationary data, and how most analysis are being done on stationary series.

Should I also convert my data to stationary before analysis? Or is trend in data important in my case?

Ach113
  • 123
  • 3
  • 1
    It depends on what method you are using. For ARIMA yest. For exponential smoothing now. They operate very differently and make different assumptions. You need to consider the assumptions of the method you are using. – user54285 Aug 06 '20 at 21:21
  • Not a duplicate, but see: https://stats.stackexchange.com/questions/19715/why-does-a-time-series-have-to-be-stationary – kjetil b halvorsen Aug 06 '20 at 21:31
  • @user54285 as of now I am just using linear regression, I want to use different approach that may be more suitable for this task, but I cant think of any – Ach113 Aug 07 '20 at 04:24
  • I would not use linear regression for time series data. If you just want to predict future results than Exponential Smoothing or ARIMA would be a start (the later is a lot harder and makes more assumptions). If you have regressors you might want to start with regression with ARIMA error. The biggest issue with regressors is spurious regression, but the solution for that such as cointegration are not for the faint of heart. Its what I am trying to learn now. – user54285 Aug 07 '20 at 17:17

0 Answers0