Exogenous, endogenous, and observable switching models of industrial production in the United Kingdom

We forecast the real industrial production (IP) growth in the United Kingdom (UK) and real stock market returns in the UK and United States (US). We use the exogenous-switching vector autoregressive (VAR) model (EXS–VAR), the novel endogenous-switching VAR model (ENS-VAR), and the novel observable-switching VAR model (OS-VAR). In EXS-VAR, the transition probabilities are constant, i.e. it is a Markov-switching (MS) VAR model. ENS–VAR’s transition probabilities are dynamic and driven by observable explanatory variables. OS-VAR is a regime-switching score-driven model where a score-driven filter drives the predictive probabilities in an information-theoretically optimal way. The in-sample data window is from April 1963 to December 2023. The in-sample forecasting period is from January 2014 to December 2023. We compare the statistical and forecasting performances of the regime-switching VAR models and the classical VAR model. The results show that the ENS–VAR is superior to the competing VAR specifications and provides the most accurate predictions of real UK IP growth.