I don’t know if I can post my doubts about a theoretical issue. I don’t really know what’s the advantage of using Bayesian approach instead of Classical approach. I explain my issue, i used a “log-INGARCH” model to fit new COVID-19 daily positives but my prof asked me to do it using the Bayesian approach, i was able to do it but i can’t really find the reason why doing that. The only thing that comes in my mind is to obtain a smooth fit, but nothing other than that. Would explain me the possible reason of such a choice? Would you recomend me any reading on the topic?
Generally (there’s nothing special about the timeseries case) the benefits to a Bayesian approach is the ability to incorporate prior information in an explicit way, which can make for more robust inference for a given combination of model complexity and data availability. I transitioned to Bayes after encountering models where search algorithms for maximum likelihood were very fragile. Bayes helps by eliminating ridiculous areas of the parameter space, plus having the “easier” goal of exploring the typical set rather than finding the infinitessimal point of maximum likelihood. (“easier” is quoted because it’s still not easy, and I’m possibly speaking beyond my expertise in that assertion; certainly my observation is that the inferences I obtain are more robust).
Also, there’s the whole “a p-value is not what you think it is” thing.
Well, one thing I like of Bayesian methods in time series is that you can naturally build a robust predictive distribution to forecast. And in a classical approach the idea to have a t-student predictive distribution is not that natural.
Miggon and Fonseca have alot about it, and using garch models, when I find the articles I will put them here
Thanks i’d really like to read those articles
Well I found this one for now,
I really love her work, check it, and tell me your feedback. I like the volatility process as well :)