本研究以臺灣中型 100成分股中之電子類股為樣本，採用技術分析指標以及三大法人買賣超數量之資訊，以類神經網路之方式預測未來股價，再以由平均數－變異數 (mean-variance)法則決定資金配置權重；在樣本期間(2006~2010年)本研究之策略可達到 334.1%之累積報酬率，年化報酬率亦高達 34.13%；就 Jensen, Treynor及 Sharpe評鑑指標而言，本研究所採用之策略，均較其他同類型基金具有較大的風險溢酬。而不論在金融海嘯發生前至發生中(2006~2008年，空頭期)，或金融海嘯發生後(2009~2010年，多頭期)，本研究所採用之投資策略，確實比其他同類型基金具有較為優異的表現。

This study constructs investment portfolio by employing neural network methodologies, which are used to predict stock price with technical analysis indexes and institutional investors’ net trading volume. We use the electronic stocks in the FTSE TWSE Taiwan Mid-Cap 100 Index during 2006-2010 as a sample. Mean-Variance rules are employed to decide the weights of capital allocation. Our strategy can make a five-year cumulative return 334.1%, equivalent to annually 34.13%. Appraising by the three major methods (Jensen, Treynor and Sharpe), our strategy performs better than other similar type mutual funds in Taiwan market during the sample period, whether before or after the financial tsunami.