Essays on the natural rate of unemployment and the Phillips curve.

摘要

The dissertation is devoted to various aspects of estimating and evaluating estimates of the natural rate of unemployment (NAIRU). Firstly, the dissertation shows that the uncertainty about the NAIRU may be reduced considerably because substantial information on the path of the NAIRU can be extracted from the relation between the unemployment rate and unemployment duration. Namely, it is demonstrated that permanent changes in the unemployment rate may be identified as the ones that do not result in subsequent unemployment duration changes. This observation helps to improve the NAIRU estimates substantially, as is indicated by their better forecasting power for inflation and better consistency between real-time and ex-post estimates. It is also shown that the approach produces similar results when applied to estimating natural output.; Secondly, the dissertation demonstrates that assessing the quality of NAIRU estimates by the root mean square of inflation forecasts based on these estimates may be misleading. As is demonstrated, the absolute size of the gross inflation forecast error does not characterize the NAIRU mismeasurement adequately because NAIRU mismeasurement explains a rather small fraction of the forecast error but, at the same time, is highly persistent. The contribution of the NAIRU mismeasurement into the forecast error may be extracted using the Kalman filter. The dissertation proposes a simple unified approach for evaluating its consequences to monetary policy performance.; The final part of the dissertation points out that the Phillips curve specifications that lead to especially wide confidence intervals for the NAIRU, namely, the ones with relatively small number of inflation lags, are characterized by considerable negative correlation between the error term and its higher-order lags. It is demonstrated that ignoring serial correlation of this kind by assuming that the errors are i.i.d. results in overestimating the degree of imprecision of NAIRU estimates. This explains the well-known fact that Phillips curves with larger number of lags produce tighter confidence intervals for the NAIRU. Moreover, this suggests that the right Phillips curve specification must contain a substantial number of inflation lags (at least 12 lags for quarterly U.S. data, 1955:1 to 2003:1) and, also, is characterized by relatively tight NAIRU confidence intervals.