I am trying to predict stock price of a company, the data is non stationary.

Steps I followed -

  1. Analyze the raw data
  2. Determine whether the raw time series data is stationary or not using ADF and KPSS
  3. Applied first differencing and seasonal differencing to make the data stationary
  4. Determine the MA and AR lags using the stationary data by plotting ACF, PACF plots

My question is should I pass raw data (non-stationary, from Step 1) to time series model like SARIMA, ARIMA and SARIMAX and use the stationary data(Step 3) to determine MA and AR lag coefficients for the model


I should pass the stationary data(Step 3) to the time series model like SARIMA, ARIMA, SARIMAX, etc. and use the MA and AR lag coefficients for the model. And then to determine the predicted original time series , I should undo all the transformations that I did in Step 3 to make the time series data stationary.

Thank you for your help

  • $\begingroup$ Depending on the type of non stationary data, (S)ARIMA(X) models are able to handle the data as is (see also this question on the stats stackexchange). $\endgroup$
    – Oxbowerce
    Commented May 8, 2023 at 16:55
  • $\begingroup$ Thanks. I see, so it means that if a model can handle non stationary data then the time series can be used as is? In what cases would we be using the converted stationary time series data to fit the model? @Oxbowerc $\endgroup$
    – Kriti
    Commented May 8, 2023 at 17:00

1 Answer 1


In general, you're going to use the data as is to fit the model, but use the data analysis to choose/validate/understand your parameters (p,d,q,P,D,Q, etc). For the most part, it's advisable to do a grid search on your parameters anyway to get the best fitting model, but having some intuitive understanding of where to set the grid search limits will always help.


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.