Three essays on asset allocation and asset pricing.
详细信息
文摘
This dissertation investigates several classical yet unsettled issues in asset allocation and asset pricing. In the first chapter, I evaluate optimized asset allocation models proposed in academia and industry against the 1/N rule equal-weight portfolio). Analytically, I derive the sufficient conditions under which optimized models achieve mean-variance efficiency. Empirically, using Ledoit-Wolf robust test, I find that the long-only constrained models do not outperform the 1/N rule consistently and significantly, but the unconstrained James-Stein estimator and global minimum variance portfolios do. Therefore it is valuable to reduce estimation errors in expected return. Additionally, I show that CAPM is able to accommodate the low-volatility "anomaly" which is behind the superior performance of minimum variance portfolio. The second chapter advocates adopting Markov regime switching models in the application of tactical asset allocation. Analytically, the regime-switching model is shown to be capable in generating the stylized patterns of asset returns including fat tail, skewness and volatility clustering. Empirically, through an in-sample study of stock-bond tactical allocation problem under mean-variance framework, I find that an univariate Gaussian regime-switching model outperforms a well-fitted ARMA-GARCH model in terms of mean return and information ratio, although the hypothesis of equal regime-specific mean cannot be statistically rejected. In addition, I demonstrate that this Gaussian setting enables closed-form solutions to complex portfolio construction frameworks that involve skewness and conditional value at risk. In the last chapter, I apply Vector Expected Utility, which is newly proposed by Siniscalchi 2011) to model ambiguity preference, to solve the equity premium/risk-free rate puzzle. I first show that Vector Expected Utility is conceptually better since it can accommodate both thought experiments in Ellsberg 1961) and Machina 2009). Secondly, to solve the puzzle, I find that the coefficient of ambiguity aversion should be around 1.3 when the other two preference parameters--time discount factor and risk aversion--are set to their reasonable levels. This calibrated value is robust since it can match the mean returns during 1889-1978 and also rationalize the Hansen-Jagannathan bound using postwar U.S. data, and can be tested in other studies.