In the spotlight: Bayesian threshold autoregressive models
The bayesmh command can now be used to fit a wide variety of Bayesian time-series models. In Stata 17, bayesmh allows you to include time-series operators, such as L.x for the lag of x or D.x for the first difference of x, for independent variables in linear, nonlinear, and multiple-equation models; this support for time-series operators opens the door to many additional models.
As an example, Bayesian estimation of threshold autoregressive models is now available. These models are useful for analyzing time series with structural breaks caused by regime switching. Influential applications include econometric modeling of business cycles. Bayesian estimation is particularly helpful for using previously available information, selecting a model, and achieving more reliable estimation in small-sample-size cases.
In my blog post Bayesian threshold autoregressive models, I demonstrate first how to fit Bayesian AR(1) models and then how to fit threshold autoregressive models of varying complexity. I also show how to select the best model.
Interested in other unique ways to use bayesmh? Check out my previous blog Comparing transmissibility of Omicron lineages for an application of a Bayesian multilevel multinomial model.
— by Nikolay Balov
Associate Director, Bayesian Statistics