Hello - I just pip installed statsforecast on my new virtual machine and am running the ARIMAX model. I am comparing the output to output from my old virtual machine that had pip installed the statsforecast package a few months ago. Using exactly the same input data for the model, the output is very different between the 2 virtual machines (SARIMA parameter selection is different, exogenous variable coefficients have flipped signs). This is causing 2 different forecasts whilst using the same input data and statsforecast functions. Was there a recent update in the model selection criteria used for the ARIMA function? I am trying to understand what may be driving these discrepancies in the forecast. Thank you.

Yes, there have been several changes. If you need the same output use the same version in both machines