摘要：

This paper considers model averaging in spectral density estimation. As an alternative to model selection, we use a Mallows criterion to select the data-driven weights and construct the spectral density function by averaging the autoregressive coefficients from all potential autoregressive models. We extend the consistency of the autoregressive spectral estimator in Berk (1974) to the autoregressive spectral averaging estimator. Simulation studies show that the proposed averaging estimator achieves a lower bias than other existing model selection and model averaging methods, and the bias of the averaging estimator approaches zero as the sample size increases. As an empirical illustration, the proposed method is used to construct the Diebold-Mariano test (Diebold and Mariano, 1995) and examine the relative performance of the two exchange rate forecasts.