The article explores how adding either a VaR or a CVaR constraint affects portfolio selection when security returns have a discrete distribution. The researchers found that portfolios on the VaR-constrained boundary show a certain separation pattern, while portfolios on the CVaR-constrained boundary exhibit a different separation pattern. They also discovered that the CVaR-constrained optimal portfolio has a smaller CVaR compared to the VaR-constrained optimal portfolio, suggesting that a CVaR constraint is more effective in reducing large losses in the mean-variance model.