Abstract:

Statistical regularities of financial asset returns, such as the leptokurtic behavior, volatility clustering and the asymmetry gain/losses, have often lead to discard the normality assumption in favor of more complex models. In this work, we consider a parametric estimation procedure, which minimizes an empirical version of the Havrda- Charvat-Tsallis (HCT) entropy. The procedure is used to estimate location and scale parameters of financial asset returns. The resulting estimator tunes the trade-off between robustness and efficiency by a single constant q. Asymptotic derivations show that the method is considerably accurate for multivariate problems with negligible losses of efficiency compared to maximum likelihood. Theoretical properties, ease of implementability and empirical results on simulated and financial data make it a valid alternative to classic parametric methods and minimum divergence methods based on kernel smoothing.