The elusive Phillips curve

The empirical evidence that has accumulated over the past twenty years for major countries, and especially the United States, Canada, and the U.K., suggests three distinct “phases” for the Phillips curve. In the “early” phase the coefficient on the unemployment variable was correctly signed and statistically significant; in the “middle” phase, as further studies were made and the data period extended, the unemployment coefficient tended to become numerically smaller and often to be statistically insignificant; in the “late” phase, as yet more studies accumulated and the data period was extended into the seventies, the unemployment coefficient was sometimes perversely signed and also statistically significant. This paper offers a possible theoretical interpretation of this phenomenon. The theoretical analysis is supplemented with some empirical simulations.