Estimating the natural rate of unemployment: Developing models for forecasting inflation
Although the natural rate of unemployment has been intensely researched over the past half century, the challenge remains to develop a model of the natural rate that provides reliable, timely forecasts that can be used by policymakers. In this paper we develop and evaluate several models of the natural rate of unemployment for the purpose of determining whether and to what extent any are suitable for forecasting. Because the natural rate of unemployment is an artificial construct it is not feasible to directly evaluate forecasts of the natural rate by conducting the usual comparison of actual and forecasted values. Our approach is to use our various forecasts of the natural rate of unemployment to forecast near and medium term inflationary pressures. Using this approach we are able to compare our forecasts of the inflation rate using the natural rate with forecasts of the inflation rate using various other methods.
We consider a variety of statistical models for estimating the natural rate of unemployment including Kalman filter, Bayesian methods, ARMA, and structural vector autoregressions. Our sample is 1984 – 2007 which incorporates the period between the onset of the Great Moderation and the start of the Great Recession. Each model is used to produce forecasts for inflation ranging from one to eight quarters ahead. We use standard evaluation methods (modified Diebold Mariano tests) to determine the relative usefulness of the various measures of the natural rate in forecasting inflation.