XXVI Edition

14-15-16 December 2017"

Hidden Markov models with multivariate leptokurtic-normal components for robust modeling of daily returns series

Maruotti Antonello, Libera Università Maria Ss. Assunta

Hidden Markov models (HMMs) are often applied in finance to capture prices' changes and the stylized facts of financial returns. This paper presents an extension of the conditional multivariate normal distributions, widely used in the hidden Markov model literature, with the aim of handling the excess kurtosis of financial returns. In particular, we propose the elliptical multivariate leptokurtic-normal distribution as conditional distribution. This distribution has a closed form representation; compared to the Gaussian distribution it possess one additional parameter which permits the modeling the excess kurtosis. The possibility of considering kurtosis allows for a better fit to both the distributional and temporal properties of daily returns. For this model, we outline an EM algorithm for the maximum likelihood estimation which exploits recursions developed within the HMM literature. As an illustration, we provide an example based on the analysis of a bivariate time series of stock market returns.

Area: Other

Keywords: Hidden Markov model; EM algorithm; Elliptical distributions; Kurtosis; Multivariate time series; Daily returns

Paper file

University Network