I have been looking at ARIMA/SARIMA models and some of the Bayesian Structural Time-Series models lately. The formulation of the two models does not seem that different but the fitting method of Kalman filter versus MCMC seems like the main difference.
I was just trying to understand what the performance differences are between standard ARIMA/SARIMA models and Bayesian Structural Time Series models (BSTS)? Is the loss lower for BSTS versus ARIMA, or are there specific cases where BSTS works better than ARIMA? Is there a particular lag structure that works better for one model or the other.
In my case, I have a much more complicated multi-seasonal component, so I was wondering if that might work better under one model or the other. But any evaluation of the differences between the models, or even references that discuss the difference would be nice.
