Brief: I am running Arima model of store data and trying to predict store growth . Inputs : 2018 - 2021 ( Monthly data) store data and forecasting 2022-2026 using one store variable only.

Model : Arima ( (p,d,q) seasonal=(P,D,Q) m =12) Using grid search for multiple p,d,q and P,D,Q values running the different scenario

Running the model, future forecast values have a constant growth , please can someone guide why year on year growth is remaining constant. It becomes difficult to forecast . Should I add something different to above arima model so that I can get a different growth rate for future values.

Growth Rate

  • $\begingroup$ What ARIMA model do you end up with? Does it include a drift term? $\endgroup$ Feb 8 at 10:23
  • $\begingroup$ Arima(tsdata,order = c(p_val,d_val,q_val), seasonal = list(order = c(P_val,D_val,Q_val),period = 12), include.drift = TRUE) @StephanKolassa $\endgroup$ Feb 8 at 11:43
  • $\begingroup$ I am going with lowest Mape and AIC value , but for every simulation the growth is remaining constant across years, value will differ but growth is constant . $\endgroup$ Feb 8 at 11:52

1 Answer 1


Your question is extremely similar to many questions about "flat forecasts" from ARIMA models. Please take a look through those earlier threads.

Essentially, ARIMA with drift fits a linear trend through your data and models deviations from that trend using AR and MA terms. ARMA processes often converge quickly to the overall mean, or to zero if there is no nonzero mean. In your case, the ARIMA forecast converges quickly to the linear trend.

Note also that ARIMA - and any other method as well - attempts to separate the signal from the noise. In this case, the signal is the trend, and any remaining ARMA dynamics. The noise is unforecastable (that's why it is noise), so it is not forecasted.


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.