An Infinite Hidden Markov Model with GARCH for Short-Term Interest Rates

This paper introduces a novel Bayesian time series model that combines the nonparametric features of an infinite hidden Markov model with the volatility persistence captured by the GARCH framework, to effectively model and forecast short-term interest rates. When applied to US 3-month Treasury bill rates, the GARCH-IHMM reveals both structural and persistent changes in volatility, thereby enhancing the accuracy of density forecasts compared to existing benchmark models. Out-of-sample evaluations demonstrate the superior performance of our model in density forecasts and in capturing volatility dynamics due to its adaptivity to different macroeconomic environments.