I am trying to predict average weekly stock prices for time series data.

Steps I followed:

  1. I tested the data to check whether it was stationary or not using ADF and KPSS tests.
  2. Next, to make the data stationary - I normalized the data using StandardScaler, Removed the trend by taking the first difference, removed the increasing volatility, removed seasonality
  3. I then plotted ACF and PACF graphs to identify the MA and AR lags.
  4. I modelled the values for ARIMA model and it seems to give an awful performance

I was thinking to apply SARIMA model next, but I am not sure if I should have removed seasonality before since SARIMA takes seasonality into account.

I would really appreciate your thoughts on this.


1 Answer 1


As for SARIMA, it's definitely a good idea to take seasonality into account, but it may depend on the specifics of your data. Have you tried looking at the seasonal decomposition of your time series? That could help you determine whether you need to include seasonality in your model.

Also, keep in mind that sometimes even the best models can give poor performance. It might be worth exploring other methods of prediction, such as machine learning algorithms or hybrid models.

  • $\begingroup$ Following is the snapshot of the original data - i.stack.imgur.com/58Uqg.png Since, this data look multiplicative ; Following is the snapshot of the decomposition of the time series i.stack.imgur.com/kOkBT.png $\endgroup$
    – Kriti
    Apr 19, 2023 at 21:18
  • $\begingroup$ Since SARIMA already takes into account of seasonality and integration; can I pass the raw data to my model which is non stationary or I should make it stationary and then pass the transformed stationary data to the fit the model? @pigsalaciarat $\endgroup$
    – Kriti
    May 8, 2023 at 15:57

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.