Abstract
This paper examines the presence of nonlinearities in the Phillips curve. We allow for a flexible form of nonlinearity and estimate a threshold regression model with the number and location of thresholds determined directly from the data. Over the estimation period starting in the late 1960s, we document that the linear model cannot be rejected if we properly control for inflation expectations. More precisely, not controlling for consumer expectations may lead the econometrician to overestimate the degree of nonlinearity. Our results hold with aggregate data, regional data, and controlling for cost-push shocks directly or using instrumental variables. While on the whole the case for nonlinearities is negligible, we examine historical episodes during which nonlinearities may have played a more important role but again do not find a significant case for nonlinearities.