近年来伴随着金融市场广度与深度的不断拓展,频发的金融风险对世界经济及金融市场造成了巨大损失(如美国次贷危机),学者和投资者越来越关注规避小概率巨额风险的最优投资决策及有潜力风险资产遴选方法的研究.文章就此开展了如下研究:首先以损失超过VaR部分的 条件期望 CVaR作为投资者愿意承担风险的上限,改进投资预算约束为非紧约束,提出了基于巨额损失波动性的投资组合模型. 数值试验验证了模型具有良好的收敛性,即使在生成较少数量的情景下也能快速收敛;当投资者对最低期望收益率要求不高时,不必全额投入预算资金就能满足投资者对预期收益的要求;随着投资者对最低期望收益率要求的提高,更多预算资金被投入可能带来更高收益的风险资产,资金预算约束逐渐趋于紧约束;模型给出的最优投资决策在样本外各滚动窗口测试中均实现了较高收益,但发生巨额损失的波动程度却显著降低,达到了控制小概率极端风险的目的.其次,结合常规基本面分析法和聚类分析技术,提出了风险资产的遴选方法.该方法适用于跨市场跨行业不同品种间风险资产的筛选,可兼顾同一类别内资产的同质性及不同类别资产间的异质性,以此达到分散化解风险的目的.实证研究表明,该方法遴选出的``少量''风险资产在各项评价指标上具有明显的优势,聚类技术的引入大大降低了投资者选择资产的难度.}

In recent years the global financial markets expanded extensively and deeply. The financial risks occurred frequently accompanied by heavy loss upon the world economy and financial markets (such as American Sub-prime Mortgage Crisis). Researchers and investors bury themselves in the optimal decision to avoid the excessive loss with small probability and the methods to pick out the promising risk assets. In view of these issues, a portfolio selection model based on the volatility of excessive loss is proposed, in which CVaR is used as the upper bound to control the risk level, and the budget constraint is not tight as usual. Numerical tests validate the model's convergency even when there is not too many scenarios. The budget capital is not necessarily fully poured into the market when a lower return rate is required by investors. When a higher return rate is required, more capital will be invested into the risky assets with higher return and the budget constraint tends to be tight. The optimal portfolio gains higher return throughout the rolling windows of out-of-sample tests, but with significant lower volatility. Meanwhile, an approach to pick out the promising risky assets is investigated by incorporating the fundamental analysis and clustering technology. This approach is appropriate for the selection of risky assets among a variety of markets and industries, which eliminates the investment risks by accounting for the homogeneity in the same cluster and heterogeneity among different clusters. The promising risky assets selected by this method dominate in each evaluation index. The introduction of clustering technology substantially cuts down the efforts for assets selection.