CVaR Optimization and Multivariate Joint Density Modeling

One attractive candidate as standard risk metric is Conditional Value at Risk (CVaR), which has a lot of advantages compared with Value at Risk (VaR). In this paper, I study CVaR as an objective function in a series of optimization problems. I compare the CPU time of the linear programming approach proposed in [14] with that of the fast gradient descent method [5] when increasing the number of scenarios and assets, respectively. The discrepancy between the two methods are also discussed. I also fit the market data to a set of joint densities and compare the market efficient frontiers and the returns ex-post. Yi-Fei Chen, [email protected]