以資料包絡法及類神經網路模型建構投資組合預測模式-以台股50為例

The Application Portfolio Model Of Data Envelopment Analysis And Artificial Neural Network Model- Taiwan Stock Market 50 As An Example

黃立忠(Li-Chung Huang)

指導教授 ： 謝俊宏

國立臺中科技大學/商管學群/事業經營研究所/碩士(2009年)

https://doi.org/10.6826/NUTC.2009.00001

摘要

隨著經濟的深入發展和人民生活水準的不斷提高，投資理財已逐步成為決定和影響人們生活的重要方面，在眾多金融商品中，股市的資訊最容易取得，雖然風險較高，但也深受投資人青睞。傳統股票評價其預測的準確度往往取決於投資者在技術面及基本面的判斷。在本研究中，藉由有效的評估方法，如探討資料包絡法如何建構「台股50」個股投資組合，對於效率好轉公司之股票採買進策略，對於效率轉差公司之股票採賣空策略，此研究結果亦可提供投資大眾及管理者績效評估的參考。
接著本研究亦採取計算智慧的方法-類神經網路建構台股50績效預測模式。本研究資料擷取自台灣公開資訊觀測站，樣本為2009年3月之台股50各成分股，以2006年第四季至2008年第三季的歷史資料做為測試樣本，並預測各季報酬率。實驗結果證明經由資料包絡法及倒傳遞類神經網路所建構之上市公司股價季報酬率預測模式與實際季報酬率皆為漲時，可買進其公司股票，當持有投資組合達3個月，一直持有到21個月時，其累積報酬率均大於台股50平均報酬率，也大於銀行3年期定期利率，因此本研究投資組合確實可提供投資人進行投資時之參考依據。

關鍵字

資料包絡法 ； 倒傳遞類神經網路

並列摘要

With the deepening of economic development and people's rising living standards, manage finances and has become an important impact on people's lives, in many financial products, market information most easily achieved, although higher risk, but also favored by investors.Traditional stock assessment of the accuracy of their forecasts tend to depend on the degree of investors and the fundamentals in the technical judgments. In this study, through effective assessment methods, such as data envelopment analysis to explore how to construct "Taiwan 50" stock portfolio for better efficiency of mining companies to buy shares strategies for efficiency and deterioration of quality of the company's stock short-selling strategy adopted, The results of this study can also provide the investing public and regulators to assess the performance of the reference.
Then this study also adopted the method of calculation of wisdom - by type of mental construct network performance of the Taiwan stock market forecasting model 50. In this study, data acquisition from Taiwan MOPS, samples in March 2009 for the stock of the stocks of 50 to the fourth quarter 2006 to the third quarter 2008 data as a history of the test samples, and the forecast of the quarter rate of return. Experimental results show that by the DEA and mental types of back-propagation network constructed by the listed company's share price quarter rate of return forecasting models and the actual rate of return quarter were up, they can buy their company's shares, when the holders of the investment portfolio 3 months, has been holding to 21 months, the cumulative rate of return than the average stock market return of 50, nor more than 3-year bank-deposit interest rate, so does the portfolio of this study can provide investors to carry out investment based on the reference time.

並列關鍵字

data envelopment analysis ； back-propagation neural network

參考文獻

8. 王美慧與簡妤玲(2005)，運用類神經網路於證券業網路下單服務品質之研究，顧客滿意學刊，第一卷第一期，P1-P30。

1. 王淑芬(2004)，「投資學」，華泰文化事業股份有限公司。

23.陳國玄(2004)，人工神經網路與統計方法應用於台灣上市電子類

12.李惠研(2003)，類神經網路與迴歸模式在台股指數期貨預測之研

25.蔡秋田、王媛慧(2004)，廠商生產效率與股票報酬之關聯－資料