To Be or Not to Be of Maxentropic Type? An Appraisal of Parametric Spectral Analysis in Econometrics

Abstract

The paradox that a parametric approach makes less assumptions than a nonparametric one can be seen as an obstacle for a wider use of frequency domain techniques in econometrics. This paper offers a throrough appraisal of the parametric model in the presence of notoriously short economic time series. Confronted with its nonparametric alternatives, pragmatic and theoretical aspects are addressed. Based on a duality framework, a proof of the theoretically desirable ME-property of the ML-estimated autoregressive model is provided. In the case of the true series following a fairly general stochastic process, it is numerically demonstrated that the parametric model in combination with a Bayesian order identification criterion offers the better representation of the true spectral density.

Keywords: Maximum entropy, parametric spectral analysis

JEL Classification: C20, C22