How Well Does the New Keynesian Sticky-Price Model Fit the Data?

The New Keynesian sticky-price model has become increasingly popular for monetary-policy analysis. However, there have been conflicting results on the empirical performance of the model. In this paper, I attempt to reconcile these conflicting claims by examining various specifications of the model within the context of a single framework. I find that the New Keynesian model does not fit the U.S. data well; in particular, the model requires additional lags of inflation not implied by the model under rational expectations. These additional lags have the interpretation that some fraction of the population uses a simple univariate rule for forecasting inflation. The views expressed in this paper are those of the author and should not be construed as those of any member of the Board of Governors of the Federal Reserve system or any other member of its staff. Earlier versions of this paper were presented in seminars and workshops at the January 2000 Econometric Society meetings in Boston; the European Central Bank; and the Federal Reserve Board. I am grateful to seminar participants for helpful comments. The New Keynesian sticky-price model has received increasing attention in recent years. For example, King and Wolman (1999), Levin, Wieland, and Williams (1999), and Rotemberg and Woodford (1997, 1999) incorporate New Keynesian sticky-price assumptions in models that they use for policy evaluation. At the same time, however, questions have arisen about the empirical validity of the model. Fuhrer and Moore (1995) have argued that the standard New Keynesian model with sticky prices and rational expectations does not fit U.S. post-war data. Fuhrer (1997b) and Roberts (1998) have shown that modifying the model so that the it includes lags of inflation not predicted by the standard model with rational expectations allows it to fit the data well. These additional lags have been justified by various arguments. Roberts (1997, 1998) argues that the additional inflation lags can be interpreted as meaning that some agents use simple autoregressive rules of thumb to forecast inflation. By contrast, Brayton, et al. (1997) and Fuhrer and Moore (1995) argue that the additional lags reflect characteristics of the structure of the economy. While Fuhrer (1997b) and Roberts (1998) claim empirical support for the role for additional lags in the New Keynesian Phillips curve, there is not, to date, a consensus concerning the empirical performance of the New Keynesian sticky-price model under rational expectations. In particular, Rotemberg and Woodford (1997, 1999) claim empirical support for the original New Keynesian sticky-price model under rational expectations. They argue that allowing for a serially correlated error term permits the model to fit U.S. data well over the post-1979 period with purely rational expectations. These competing views matter, because policy implications differ markedly depending on whether the fully or partially rational versions of the New Keynesian model are adopted. Rotemberg and Woodford (1997) show that in their model with fully rational expectations, an optimized central-bank reaction function ought to have a very large coefficient on lagged interest