This paper proposes a novel large dimensional regime switching factor model in which the regimes change according to the threshold principle: the shift is driven by a variable parameterised as a linear combination of the elements of a vector of covariates. The paper considers nonparametric principal components estimation of the latent factors and factor loadings; and concentrated least squares estimation of the remaining set of parameters. The asymptotic properties of the estimator are derived and the good finite sample performance is shown in a comprehensive Monte Carlo analysis. Finally, an extensive empirical application focuses on the links between financial markets and the real economy: the results show that nonlinear dynamics in the joint distribution of a large panel of U.S. macroeconomic variables may be anticipated by changes in the conditional distribution of stock returns as described by conditional volatility and higher order moments; and the regimes identified by the model are related to the business cycle reference dates provided by the NBER. The empirical results therefore suggest that the conditional distribution of stock returns is a potential leading indicator of economic activity.

 VIEW PAPER