Conditional VaR 모형을 사용한 포트폴리오 최적화에 관한 연구 : 평균-분산 모형과 비교

Abstract

Recently, Conditional Value-at-Risk(CVaR) is considered to be more coherent risk measure than VaR in that CVaR has properties of sub-additivity and convexity that VaR doesn t have. This paper empirically tests four-asset portfolio optimization with CVaR constraints and Mean-Variance(MV) constraints, then compares two models using subsampling method and bootstrapping simulation. The optimal risky portfolio maximizes the risk-adjusted return, which is the excess return divided by the standard deviation(in MV) or CVaR. As the result, when we compare four-asset(government bond, corporate bond, large capital stock and small capital stock) portfolios by CVaR and MV approach, the optimal risky portfolio shows asset weight difference. However, this asset weight difference shows no significant difference between MV and CVaR models in that the upper and lower 5% points of asset weight difference distribution contain 0 between them. CVaR can be the better alternative to MV and VaR as a standard risk measure, because CVaR satisfies all the factors to be coherent risk measure. However, we can not conclude that CVaR would be superior to MV if there were no significant difference in terms of the optimal risky asset weight as the result shows.