Extensions to SCAR Models: Modeling the Dependence Structure in Equity Returns

Despite their drawbacks at higher dimensions, SCAR models are currently being used at large and they perform quite well as compared to their competitors. This paper investigates possible extensions on SCAR models and aims to ultimately find a extension which outperforms the original specification. We empirically apply SCAR to real-world data using daily returns from two popular indices, the Dow Jones Industrial Average and NASDAQ. First, we relax the assumption of Gaussian marginal distributions errors. The Student's t-distribution better describes the stylized facts of asset returns such as the leptokurtosis and we find out that the performance of SCAR significantly increases in this scenario. Another extension we consider is to combine the best-performing copulas into a mixed copula model which captures the dynamics of stock returns better. The mixed copula SCAR only outperformed the original specification during times of significant market occurrences where the combined copula takes into account greater dependence for losses. This research provides the current literature with two extensions of the SCAR model which predominantly outperforms the original model.