Verifica empirica ed analisi critica del modello mistura bivariata

Abstract

Bivariate mixture models, introduced by Tauchen-Pitts, try to esplain the relationship between return volatility and trading volume in financial markets, by stochastic changes in a single latent variable, representing the number of information arrivals. In this article, dynamic bivariate mixture models that allow for autocorrelation in the latent variable are represented by nonlinear state space systems, nonlinearity being due to the measurement equation. The parametres are estimated by simulated maximum likelihood and the latent variable by simulated non linear filter, both being based on the same rejection sampling algorithm. The results, based on italian daily stock market data, reveal that dynamic bivariate mixture models neither can explain the autocorrelation, nor can account for the persistence in return variance.