Inside and Outside Bounds: Threshold Estimates of the Phillips Curve

There have been several instances over the past 40 years when large movements in the unemployment rate have elicited little response in the inflation rate. Such instances, while casting doubt on the tradeoff implied by the linear Phillips curve, are also associated with large inflation forecasting errors. In principle, these movements are consistent with a Phillips curve relationship; they just require the curve to shift in the same direction as the unemployment rate. Econometric representations of the Phillips relationship usually incorporate factors that can cause the Phillips curve to shift over time. However, the literature has not yet provided a test of whether such factors are sufficient to explain the episodes of horizontal movement. In this paper, the authors test the explanatory power of a double threshold specification of the Phillips relationship against a simple linear specification, and compare dynamic and static out of sample forecasts of inflation across linear and double threshold specifications of the Phillips curve. The authors find that traditional shifters in the relationships are insufficient for characterizing the periods of horizontal movement, and that a double threshold specification makes significant improvements in the static and dynamic out of sample inflation forecasting performance of the Phillips curv