An Empirical Application of a Random Level Shifts Model with Time-Varying Probability and Mean Reversion to the Volatility of Latin-American Forex Markets Returns [Una aplicación empírica de un modelo de cambios de nivel aleatorios con probabilidades cambiantes y reversión a la media a la volatilidad de los retornos cambiarios en América Latina]

Following Xu and Perron (2014), this paper uses daily data for six Forex Latin American markets. Four models of the family of the Random Level Shift (RLS) model are estimated: a basic model where probabilities of level shift are driven by a Bernouilli variable but probability is constant; a model where varying probabilities are allowed and introduced via past extreme returns; a model with mean reversion mechanism; and a model incorporating these two features. Our results prove three striking features: first, the four RLS models fit well the data, with almost all the estimates highly significant; second, the long memory property disappears completely from the ACF, including the GARCH effects; and third, the forecasting performance is much better for the RLS models against an overall of four competitor models: GARCH, FIGARCH and two ARFIMA models.