A two-stage maximum entropy approach for time series regression

Abstract

The maximum entropy bootstrap for time series is a technique that creates a large number of replicates, as elements of an ensemble, for inference purposes, which satisfies the ergodic and the central limit theorems. As an alternative to the use of traditional techniques, this work proposes generalized maximum entropy for the estimation of parameters in all the replicated models. An empirical application and a simulated example illustrate the advantages of this two-stage maximum entropy approach for time series regression modeling, where maximum entropy is used both in data replication and in parameter estimation.