SPARC: A New Semiparametric Correlation Model

We propose a new semi-parametric model for correlation dynamics in asset returns that nests the popular constant conditional correlation model of Bollerslev (1990). It is commonly found that correlations in financial markets are not constant over time. Models for dynamic correlations have been proposed recently, e.g. by Engle (2002). In particular, there is a growing literature reporting asymmetries in correlations, where correlations increase in bear markets but remain stable in bull markets, see e.g., Ang and Chen (2002), Butler and Joaquin (2002), Campbell et al. (2002), Cappiello et al. (2003) and Longin and Solnik (2001). Such asymmetries would have ma jor consequences for risk management, as benefits from diversification seem to melt down when they are mostly needed. Our model can detect asymmetries in correlations with respect to a common factor, for example the lagged market return or volatility. Unlike many competing multivariate GARCH models (see Bauwens et al., 2005 for a review), it does not suffer from the curse of dimensionality.