本研究從台灣前60大市值比上市公司中，挑出49家公司為樣本，蒐集2006-2018間的資料，採用技術指標作為變數，以卷積神經網路預測為選股策略，選取投資組合成分股， 再利用「平均－變異數」分析配置權重，並根據不同風險趨避程度，建構不同投組。結果卷積神經網路的投資策略，在訓練樣本期間(2010～2016年)內的績效表現相當好，但應用在樣本外期間(2008～2009年，2017～2018年)則表現不佳。若使用此種交易策略與簡單動能策略比較，則動能策略建構的投資組合能在訓練樣本外期間表現的較佳。

This Research selects 49 companies from the top 60 companies in Taiwan as a sample, collects stock data from 2006 to 2018. Choose technical indicators as variables, and use convolutional neural network prediction as a stock selection strategy to form a portfolio. In the selected stocks, the “Mean-Variance Analysis” is used to allocate the asset weights, and different investment groups are constructed according to different risk aversion levels. The result of this study shows that: the investment strategy of the convolutional neural network is quite good during the training period (2010~2016) of data. However, the strategy make negative return during the out-of-sample period (2008-2009, 2017~2018). With this performance, compare to a simple momentum strategy, the momentum portfolio can perform better during the out-of-sample period.

[1] 王春峰、屠新曙、厉斌(2002)，效用函数意义下投资组合有效选择问题的研究，中国管理科学，第10卷第2期，4月，頁15-19。
[2] 李顯儀、吳幸姬(2009)，技術分析資訊對共同基金從眾行為的影響，臺大管理論叢，第20卷第1期，12月，頁227-260。
[3] 陳嘉惠、高郁惠、劉玉珍(2002)，投資人偏好與資產配置。臺灣管理學刊，第1卷第2期，2月，頁213-232。
[4] 詹錦宏、吳莉禎(2011)，動能投資策略於台灣股票市場之應用—含金融海嘯之影響，會計學報，第3卷第2期，5月，頁1-22。
[5] Allen, F. & R. Karjalainen (1999). ‘‘Using Genetic Algorithms to Find Technical Trading Rules,’’ Journal of Financial Economics, 51, 245-271.
[6] Bai, S., J. Kolter & V. Koltun (2018). ‘‘An empirical evaluation of generic convolutional and recurrent networks for sequence modeling,’’ Retrieved from https://arxiv.org/abs/1803.01271.
[7] Bodie, Z., A. Kane & A. Marcus (1999). Investments, 4th ed. McGraw-Hill Companies, 178-193.
[8] Cesarone, F., A. Scozzari & F. Tardella (2010). ‘‘Portfolio selection problems in practice: a comparison between linear and quadratic optimization models,’’ Retrieved from https://arxiv.org/abs/1105.3594
[9] De Bondt, W. & R. Thaler (1985). ‘‘Does the Stock Market Overreact?,’’ Journal of Finance, 40, 793-805.
[10] Jegadeesh, N. & S. Titman (1993). ‘‘Returns to Buying Winners and Selling Losers: Implications for Market Efficiency,’’ Journal of Finance, 48, 65-91.
[11] LeCun, Y., L. Bottou, Y. Bengio & P. Haffner (1998). ‘‘Gradient-based learning applied to document recognition,’’ Proc. IEEE, 86, 2278-2324.
[12] Lo, A. W. and A. C. MacKinlay (1990). ‘‘When Are Contrarian Profits Due to Stock Market-Overreaction,’’ Review of Financial Studies, 3, 175-208.
[13] Markowitz, H. (1952). ‘‘Portfolio Selection,’’ Journal of Finance 7, 77-91.
[14] Thawornwong, S., D.Enke & C. Dagli (2003). ‘‘Neural Networks as a Decision Maker for Stock Trading: A Technical Analysis Approach,’’ International Journal of Smart Engineering System Design, 5(4), 313-325.
[15] Vejendla, A. & D. Enke (2013). ‘‘Evaluation of GARCH, RNN and FNN Models for Forecasting Volatility in the Financial Markets,’’ Journal of Financial Risk Management, 10(1), 41-49.
[16] White, H. (1988). ‘‘Economic prediction using neural networks: the case of IBM daily stock returns,’’ Proc. IEEE int. conf. on neural networks, 2, 451-458.
[17] Wood, D. & B. Dasgupta (1996). ‘‘Classifying trend movements in the MSCI U.S.A. capital market index－A comparison of regression, arima and neural network methods,’’ Computers & Operations Research, 23 , 611-622.