The complexity of inflation and a Latent Variable Phillips Curve

Structural Phillips curve models have historically struggled at forecasting inflation out of sample. This warrants a rethink of some of their foundational assumptions. The current paper proposes a different take on the Phillips curve theory in which individual firms are no longer confined to using the same strategy for setting prices, while the demand for their products and services is no longer bound by the new Keynesian laws of monopolistic competition. Within this framework, the true drivers of inflation ultimately manifest themselves as both industry-specific and fundamentally unobserved. For this feature, the new model is called a “Latent Variable Phillips Curve” (LVPC). In the empirical section, out-of-sample inflation forecasts from 12,896 combinatorially generated Phillips curve factor specifications are used to answer two key questions raised by the LVPC model: First, does a given factor combination generate more accurate forecasts when the individual factors are treated as proxies of an unobserved inflationary force? Second, is this unobserved force universal across economic segments, implying some degree of homogeneity in price-setting behaviour at the aggregate level? The answer to the first question is overwhelmingly “yes”. The answer to the second question is overwhelmingly “no”.