Abstract

CABRERA GONZALEZ, Gustavo  and  LEON ARIAS, Adrián de. Markovian modeling to identify and to forecast the dynamics of the industrial production index from 1980 to 2018. EconoQuantum [online]. 2019, vol.16, n.2, pp.23-41.  Epub Mar 17, 2020. ISSN 2007-9869.

In this article, by Markov switching modeling we study the identification of unknown states and forecasting of the monthly industrial production index of Mexico from 1980 to 2018. Given that the data-sample is subject to strong economic and financial fluctuations, from a battery of auto-regressive models (linear and Markov switching parameters), the specification that best fits to data through the Bayes factor is chosen. The model selection of the monthly growth rates index leads to parameters (mean and volatility) change over time. A forecast exercise is carried out on the Markovian model of best fit to data. To measure the accuracy on the estimation, its efficiency is compared with the linear auto-regressive models on the same data. Results provide evidence that the mean of the forecasting errors (in-sample and out-sample) are lower than those of the linear auto-regressive model. The Bayesian methodology applied allows to estimate endogenously and accurately infer, despite of identification problems of Markov switching parameters, small number of observations in regimes, atypical data, number of regimes, and uncertainty in parameters subject to switch.

Keywords : Industrial production index; Markov switching; Bayesian analysis; forecasting.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )