Greatly enjoy exploring data in Orange Data Mining!

I have daily average temperature data for several years. I can plot the periodogram, and do a seasonal decompose. Is there a way to forecast the temperature using SARIMA? I tried using ARIMA with moving average order (q) = 70 (the max possible). It is still spinning. Shouldn't it be set to 365 for when looking at yearly data?

Appreciate if you can share some advice Cheers!


1 Answer 1


It is generally not recommended to set the moving average order (q) to a very high value like 70. This is because large values of q can make the model very complex and may lead to overfitting. Additionally, it may cause the model to take a very long time to fit or fail to converge.

When working with seasonal time series data, using a seasonal ARIMA (SARIMA) model is often more appropriate instead of a standard ARIMA model. SARIMA models are designed specifically for time series data with seasonal patterns.

To determine the appropriate parameters for a SARIMA model, you can use a combination of visual inspection of the time series data, the autocorrelation and partial autocorrelation plots, and model selection criteria such as AIC or BIC.

In terms of setting the moving average order (q) for yearly data, it would depend on the seasonality of the data. If the data exhibits a yearly seasonality (i.e., a repeating pattern that occurs once per year), then you would typically set the seasonal moving average order (Q) to 1 and the seasonal period (m) to 12. However, if there is no clear seasonality in the data, then you may not need to use a SARIMA model at all.

  • $\begingroup$ When doing a periodogram in Orange Data Mining, it shows 365 days seasonality for a daily temperature, with a score of 50% . The line plot for 5 years also looks yearly seasonal. The ARIMA has these parameters: autoregression order p , apparently max 99. Differencing degree d , apparently max 2. Moving average order q, apparently max 70 . ARIMA with p=1 or 12, d=0 or 1, q=0 or 1 did not produce good results. Is ARIMA widget suitable for this purpose? If yes, what would be the good starting parameters? $\endgroup$
    – Gamer 007
    Mar 14, 2023 at 8:31
  • $\begingroup$ The starting parameters can be observed from the PACF plot, usually we start with lower values then we increase the to higher lagged values, keep an eye on the BIC (the lower the better) and R^2 (the higher the better) metrics they help you determine the model with the best predictive power and better performance, keep in mind the best model can be the simplest one. AFAIK, Orange (at this time), doesn't provide a widget for SARIMA, I'm not sure if it's possible to use the Seasonal Adjusment widget to decompose the series then feed the seasonal component to ARIMA model along other components. $\endgroup$
    – Ibrahim.H
    Jun 11, 2023 at 21:05

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.