Is the assumption of constant factor loadings too strong in practice?

Abstract

The assumption of constant loadings when estimating factor models is implicit in all empirical applications used in macroeconomics. We test this assumption explicitly in a general smooth transition loadings setting using a well-studied macroeconomic dataset on the U.S. economy. Our proposed testing approach has reasonable finite sample properties relative to alternatives that allow for an abrupt change in the factor loadings. In an empirical application, we find evidence in support of non-constancy in factor loadings and show that it is mainly concentrated in certain sections of the economy (such as financial variables). Overall, our findings provide additional support toward developing factor models, which explicitly account for non-constant factor loadings.