Empirical research of Taiwanese stock market''s rate of return -- adoption of higher moments and genetic algorithms

• Main Authors: Liu Chao-Hsien, 劉昭賢
• Other Authors: Diana Hwei-An Tsai
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/26012595100080384184

碩士 === 國立中山大學 === 經濟研究所 === 86 === The development of the CAPM involves placing a restriction on the investors'' utility function or on the distribution of returns. That restriction being that either investors possess a quadratic utility function or securities have normally distributed returns. Consequently, investors choose between securities on the basis of mean and the variance of those securities. Higher order moments either do not exist or can be ignored. But when higher order moments is a feature of the distribution of returns, higher * moments valuation models are necessary. Empirical studies have found ex post common stock returns to be consistently positively skewed and many authors have been successful in using skewness as an explainer of returns. In this paper, we first augment market model with statistical information describing the skewness of the distribution of returns on the asset, and apply the resulting empirical model to measure the conventional beta measure of risk for a number of TSE''s stock. Based on this augmented model, our findings are that (a) the skew-augmented asset pricing model out-performs market model, and (b) a beta measure of risk based on the skew-augmented asset pricing model shows interesting difference in risk before and after 90''s deregulatory change in Taiwanese stock market. Secondarily, we augment market model with statistical information describing the skewness and kurtosis of the distribution of returns on the asset, then partition this four-order-moments-augmented asset pricing model according to the relative performance of individual asset returns vis-a-vis the market substantial clarification results : the explanatory power of this "Four State Decision Rule Model"(4SDR) better than market model. Traditional APM have only used the historical data to explained the source of assets'' return, so the main contribution of APM was only for reference. Here we adopt a method of "Artificial Intelligence"(AI), called "Genetic Algorithms", to forecast index return of TSE. We use fundamental variables(macro-economic data) to train our system according the rule of "survival of fittest", then select the investment strategy having the best fitness (we select accumulative return as our fitness). After our simulation, we find that GAs can correctly create benefitable investment strategies and GA''s investment strategies can offer a higher accumulative return about three times of "buy and hold" strategy during our sampling period.