【作者】 张金利；

【导师】 唐万生；

【作者基本信息】 天津大学 ， 管理科学与工程， 2007， 硕士

【摘要】 金融市场存在着大量的不确定性,表现为随机性和模糊性,在风险资产的投资中则表现为未来收益率的不确定性。在投资问题中存在人的主观判断,因此不确定性往往表现为模糊性。另外,典型的投资组合模型大多是静态的,然而投资者的投资行为往往是动态的。为此,本文研究模糊动态投资组合的破产风险控制模型和基于VaR值的模糊动态投资组合模型,并给出了相应的求解算法,具体内容如下:首先,投资者的风险偏好不是一成不变的,在不同的时期投资者一般有不同的投资行为,即有不同的风险偏好水平。本文假定投资者只有在每一阶段末的资产都能够满足消费的情况下才能存活,决策的目标函数为每一阶段末的资产金额不低于预期消费水平的可信性之和,建立了模糊动态投资组合的破产风险控制模型。其次,通过对风险值VaR的分析,提出用VaR来度量投资者的风险偏好水平:投资者在不同的时期可以有不同的VaR,即风险偏好是不同的。目标函数为各个阶段末的实际收益率不低于预期收益率的可信性之和,建立了基于VaR的模糊动态投资组合模型。对本文提出的两种动态投资组合模型,运用结合模糊模拟、遗传算法、神经网络和动态规划理论的混合智能算法进行求解,并分别给出相应的计算实例验证算法的有效性。更多还原

【Abstract】 Stochastic and fuzzy are two main aspects of the uncertainty in the financial market. The uncertainty of securities’ return rate is the case. The financial market is characterized by fuzziness because of an investor’s subjective judgment. Therefore, the models of portfolio selection with fuzzy criterion are investigated, their hybrid intelligent algorithms are also designed. The concrete contents are as follows:Generally speaking, risk preference is not fixed but depends on the context of choice. This paper assumes that the investor survives only if the wealth is large enough to meet the consumption requirement in every time period over the finite horizon. The criterion function is the sum of the credibility that the current wealth is not lower than the given consumption level. So the model of fuzzy dynamic portfolio selection for survial is put forward.After analysing the characters of the VaR value, VaR value is used to measure the investor’s risk preference. The investor may have the different VaR value in different time period, which indicates that the investor’s risk preference is changing. The criterion function is the sum of the credibility that the return rate of portfolio is not lower than a given expected one. So the fuzzy dynamic portfolio selection model based on the VaR value is put forward.In order to solve the two models proposed above, a hybrid intelligent algorithm that the fuzzy simulation based genetic algorithm and artificial neural network are integrated with the dynamic programming is given to solve it. Illustrative cases are given respectively to demonstrate the efficiency of the proposed method.更多还原