投资组合管理的研究从优化决策的角度来说，已经得到了很好的发展。然而，资产收益预测作为决策的基础，与投资组合优化贯穿一体的研究却未得到很好的探索。本项目拟研究1）Forward-Looking和Backward-Looking相结合的方法生成对资产收益的预测，考虑到Forward-Looking方法对应的逆问题具有多解的不确定性，将构建稳健投资组合优化问题；2）基于有用信息稀缺性和信息来源的多样想，建立基于混合分布模型的学习型预测方法并融入投资者观点，在此基础上构建稳健的投资组合问题。3）考虑崩溃风险指标下的非线性对冲组合模型和优化方法，为避免系统风险发生时造成巨大损失提供决策模型和方法。

In portfolio management, the optimization (decision) methodologies are well developed up to date. However, there are very limited literatures considering the integration of prediction into portfolio decision. The aims of this project are to investigate the following issues: 1) Combining Forward-Looking and Backward-Looking methods in portfolio selection. Notice that the inverse problems according to the Looking-Forward methods are always have multiple solutions, we try to construct robust portfolio selection models to overcome this type of uncertainty. 2) Portfolio selection based on leaning approaches via mixture model. The mixture model will be used to deal with the informations from different channels, including the Backward-Looking information, Forward-Looking information and investor's subjective views. 3) Nonlinear portfolio selection based on "crash down" risk measure. We will consider the portfolio selection problem involving nolinear assets such as options to hedge the crash down risk while systematic risk occuring.