EcoSta 2017: Start Registration
View Submission - EcoSta2017
A0326
Title: Model selection in the time-variant cointegration model Authors:  Roberto Leon-Gonzalez - GRIPS (Japan) [presenting]
Abstract: Calculating predictive likelihoods or marginal likelihoods in Time-Variant Cointegrating models can be very time consuming and challenging in large dimensions. We adapt the Stochastic Search Variable Selection (SSVS) algorithm in order to use the Gibbs sampling algorithm for model selection. However, we find that the SSVS algorithm can suffer from a high correlation between the model indicator and the other parameters. To solve this problem we design an algorithm that proposes a joint update of the model indicator and other model parameters such as the cointegrating space and the adjustment coefficients. We find that this joint update solves the problem and allows the algorithm to switch among models irrespective of initial values. We illustrate the methodology with an analysis of 10 electricity prices from Mexico.