A0261
Title: A two-stage maximum entropy approach for time series regression
Authors: Pedro Macedo - University of Aveiro (Portugal) [presenting]
Jorge Duarte - University Pungue (Mozambique)
Maria Costa - University of Aveiro (Portugal)
Mara Madaleno - University of Aveiro (Portugal)
Abstract: The maximum entropy bootstrap for time series is a powerful technique that creates a large number of replicates, as elements of an ensemble, for inference analysis. As an alternative to the use of some traditional estimators, generalized maximum entropy is proposed for the estimation of parameters in all the models generated by the maximum entropy bootstrap. A simulation study suggests that generalized maximum entropy is competitive (in a mean squared error loss sense) with some traditional estimators when the models are reasonably well-conditioned and is superior in ill-conditioned scenarios. Empirical applications on energy markets and climate change science are provided to illustrate the procedures where maximum entropy is used both in data replication (maximum entropy bootstrap) as well as in parameter estimation (generalized maximum entropy).
