Implementing a non-Gaussian quantitative model for improving the accuracy of risk modeling and for efficient portfolio construction

AbstractPurpose- The research aims to propose a scalable multivariate non-Gaussian model for VaR and volatility estimation and analyze its efficiency(in appropriately taking into account fat tails, and asymmetry for VaR estimation) as compared to traditional Gaussian approaches used for VaR estimation and portfolio construction. Financial distributions generally tend to portray asymmetry, fat tails or a mixture of distributions, which are captured by the discussed model that involves incorporating the Pareto distribution, skewed- t distributionThe study also intends to demonstrate that this results in portfolios that are having more utility. (Utility: How much excess return was generated for each unit of risk taken by the portfolio and also generating a portfolio with higher economic growth)alternative approaches including mean-variance. The utility can be leveraged by efficiently selecting(maximizing the utility defined above) the portfolio assets and their weights. This research also investigates the correlation across stock market indices of 6 Asia-Pacific countries and uses VaR and variance as the risk measures for the analysis.Findings-The authors successfully concluded that the model proposed in the paper estimates VaR more accurately than the traditional model. Also, our proposed model tends to outperform the literature model in portfolio construction by producing asset weights more efficiently.Design/Methodology/Approach- This research uses a GSEV strategy that comprises a univariate EGARCH approach for calculating stochastic volatility and the leverage effect. The Generalized Pareto distribution captures asymmetry and heavy tails in the GARCH residuals (GPD). The skewed-t copula is used to describe asymmetric tail dependence. Originality/value- The value added is to depict how the non-Gaussian model proposed by the authors includes the exponential GARCH (EGARCH) technique, Generalized Pareto distribution, skewed-t copula, and mixed distributions consideration gives VaR prediction and portfolio asset weights values more accurately than the traditional model