| Summary: | 碩士 === 靜宜大學 === 會計學系研究所 === 91 === In the ever-changing market of securities, all researches and trading rules are subject to significant inaccuracy because of the dynamic environment. So every single analytic theory should be self-adjustable in order to accommodate the dynamic nature of investment.
The objective of this study is to disentangle the complicated relationship between technical indices and fluctuating stock prices so that changes in stock prices can be effectively forecasted, by utilizing technical indices in credit trading and the technique of neuro-fuzzy.
The study results show that, based on in-sample data, neuro-fuzzy is better than regression and Buy and Hold Strategy in terms of annual rate of return, no matter trading cost is taken into account or not. When apply this to the out-sample data, which is based on 30 stocks, neuro-fuzzy has better rate of return than both regression and Buy and Hold Strategy in 24 stocks, if not considering trading cost. When trading cost is included in the comparison of models, neuro-fuzzy is better than regression in 21 stocks and better than Buy and Hold Strategy in 23 stocks. From the above it can be concluded that neuro-fuzzy yields higher rate of return.
In prediction of the direction of stock price fluctuations, neuro-fuzzy, with more than 80% hit rate, is more accurate than regression when predicting negative change of stock prices, no matter in in-sample or out-sample data. Although neuro-fuzzy falls behind regression in the prediction of positive changes of stock prices, neuro-fuzzy, with 75% hit rate in total more accurate than regression in predicting the overall change of stock prices.
In term of the mean squared error, regression and neuro-fuzzy are almost the same, for the on in sample data. Nevertheless, based on out-sample data, neuro-fuzzy has a bigger mean squared error than regression.
|
|---|
