[pystatsmodels] Fitting a SARIMAX model

[Georgios Boumis]

Hello everyone,

I have a time series for 751hours and trying to fit a model so I can match
the last 48 hours' predictions with the observed values. The acf and pacf
of this time series is as in the attached photos. I have made a boxcox
transformation to my data to get a better residuals analysis and to improve
AIC. I have been "playing around" with the seasonal model
(1,0,3)x(0,0,3,24) but I'm getting nowhere near the observed values where I
run my results.predict. How can I identify a better model for my data?

[Chad Fulton]

Post by Georgios Boumis
Hello everyone,
I have a time series for 751hours and trying to fit a model so I can match
the last 48 hours' predictions with the observed values. The acf and pacf
of this time series is as in the attached photos. I have made a boxcox
transformation to my data to get a better residuals analysis and to improve
AIC. I have been "playing around" with the seasonal model
(1,0,3)x(0,0,3,24) but I'm getting nowhere near the observed values where I
run my results.predict. How can I identify a better model for my data?

This can be a tough problem. The method results.plot_diagnostics() can
produce some diagnostic figures based on the residuals that may be useful.

Alternatively you might try a grid search for model selection - we have a
partially finished GSOC that can help with this, or else you might try the
pyramid library.

Best,
Chad