應用資料包絡法及遺傳演化類神經網路模型建構最適投資策略─以台灣股票型共同基金為例
| Autor: | Tsui-hua Huang, 黃翠華 |
|---|---|
| Rok vydání: | 2007 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 95 In this paper we adopt performance appraisal tools of DEA (Data Envelopment Analysis), Sharpe index, Treynor index, and the average ROR (rate of return) of month, etc, to select stock portfolios. Then, we establish portfolios with equal weights from these methods. Next, this research not only applies multiple linear regression, but also selects methods with computation intelligence (i.e. Neural Network, Genetic Algorithms and GABPN, etc) to construct prediction models for ROR of the mutual funds. Lastly, we establish various investment strategies (Multiple Regression, Neural Network, Genetic Algorithms, GABPN, Follow-The-Trend, Anti-Trend, and Buy-And-Hold) for mutual funds. Then this research is followed with further analyses and comparisons in performance of various investment strategies. This research adopts materials extracted from Taiwan Economic Journal (TEJ). There are a total of 165 stock funds samplings taken within the year of 2005. Top five in performance for the stock fund samples are selected as investment portfolios. Various prediction models are established. The independent variables selected are the RORs of the investment date which is five days prior. The ROR for that date is selected as dependent variables. The year of 2005 is adopted as the training period. From January 2006 to March 2006 is selected as prediction period. The empirical study findings are as follow: 1. As for stock selection strategy, this thesis applies DEA, the average ROR of month, Sharpe index, Treynor index, etc, to construct mutual funds that are highly relevant with respective performances observed. And it also exhibits significant relevancies through further statistical tests. 2. As for the prediction accuracy, root-mean-square error: The portfolio which is established through DEA, exhibits the lowest errors. Through statistical tests, the findings show there are significant differences existed. 3. As for the accuracy, hit-and-miss ratio: Portfolios established by following Treynor index and average ROR of month, in addition to those established by Genetic Algorithms, have the best performances. Nevertheless, there is no significant difference observed through the statistical tests. 4. As for the performance of investment funds: the ROR for those mutual fund portfolios selected via Treynor index is conspicuously higher than portfolios selected by other methods. Nonetheless, the ROR for investment portfolio selected by GABPN is better than portfolios selected by other methods. There are significant differences observed through the statistical tests. 5. As for overall performance (the average ROR of month, hit-and-miss ratio, and Root-mean-square error), the methods possessing learning mechanism like intelligence computation, perform better that those traditional methods without learning mechanism. 6. As for the investment strategies, the accumulated ROR is significant correlated to hit-and-miss ratios, and not significant correlated to RMSE however. This signifies that, in order to elevate the accumulated ROR, hit-and-miss ratios can also be used as the target to be relied upon. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|