An Empirical Study On Portfolio Model Based On Conditional VaR Value

As a measure of tail-distribution of investment return, conditional value-at-risk (CVaR) has been accepted in industry by both investors and administrators as an important risk measure. Financial institutes have been required by Basel Accord III to report CVaR for the sake of risk supervision. Model using CVaR as objective function has gradually taken place of Markowitz’s (1952) model using variance as objective function in terms of practical importance. This paper conducts an empirical study of Chinese A-share market and simulates situation of non-crisis market and market in crisis, compares stability of optimal portfolio derived from data-driven model based on CVaR with data-driven model based on variance, and states whether investors are likely to under-estimate risk if they use model based on CVaR. Finally, we conclude that no matter whether the market is in crisis or not, model based on CVaR are more vulnerable to estimation error of historical data than model based on variance, and this impact will be enlarged by optimization, leading to instability of portfolio weight. Moreover, investors will be misled by this model to under-estimated true risk. So, we don’t recommend investors in A-share market to adopt model based on CVaR to construct portfolio weights.