Modeling financial dependence with support vector regression

Article type: Research Article

Authors: Wang, Yongqiao

Affiliations: School of Finance, Zhejiang Gongshang University, Hangzhou 310018, Zhejiang, China. Tel.: +86 571 2887 7720; Fax: +86 571 2887 7705; E-mail: wangyq@zjgsu.edu.cn

Abstract: Copula has become the standard tool in financial dependence modeling. This paper proposes a novel nonparametric bivariate copula estimation method which smooths empirical copula with shape-restricted least squares support vector regression. This method exploits a priori shape knowledge of copula function: boundary and 2-increasing, by supplementing the classical support vector regression with shape-related constraints on an equidistant grid of the support [0,1]2. Its training can be reformulated to a convex quadratic program, which is computationally tractable. Experiments on both an artificial data set and financial time series clearly show that it has good finite sample property and can achieve significantly better performance than parametric methods and kernel smoother.

Keywords: Support vector regression, copula, nonparametric regression, dependence, shape-restriction

DOI: 10.3233/IDA-130576

Journal: Intelligent Data Analysis, vol. 17, no. 2, pp. 233-249, 2013