TGM AL MAMUN
Empirical Research Guide · Markov-Switching Models

Which Markov-switching model fits your data?

The right Markov-switching model depends on whether your series are stationary, whether the regime is observed or hidden, and how your data are structured. Test a real dataset, upload your own, or enter your test results, and follow the decision tree to the model your paper needs.

5 real example datasetsTests run in your browserPython · statsmodels

Choose a dataset

Five real datasets. Each one leads to a different outcome, including the case where Markov switching is the wrong tool.

Method guides

Routes to a finished paper

Each guide takes one real dataset through pre-estimation tests, estimation, diagnostics, robustness and sample text for the paper, with Python code and real output.

How the tool decides. Markov-switching models need stationary series, so trending levels are sent back for transformation. If an observed variable above or below a threshold defines the regime, the threshold VAR (Bayesian Guide) is the better tool. Otherwise the structure decides: one series (MS-AR), a regression (MS regression), several series (MS-VAR) or a panel (panel MS-VAR). The page also reports the evidence for regimes: the likelihood-ratio statistic of a two-regime model against one regime, compared with the Garcia critical value of about 14, the AIC and BIC of both models, and the share of periods in the smaller regime, which flags a collapsed regime. These are screening numbers from a simple two-regime fit with no autoregressive terms and a switching variance, so they can differ from the full test in each guide (the Hamilton GNP guide, with AR(4) and a switching mean only, does not reject linearity at the Garcia value). The guide for each model covers its own test.