股票市場是一般民眾常接觸的金融市場之一，為使用少許資金來規避非系統性風險，指數股票型基金(Exchange Traded Fund, ETF)是個很好的選擇，ETF所收取的管理費與交易稅較低，且透明度較高，大多為追蹤指數或大盤。因此本研究使用類神經網路作為ETF之相關研究，嘗試將用類神經網路去預測ETF的當日的合理報酬，比較類神經網路是否能比傳統的Fama-French三因子模型得到更好的結果，再加入相關研究中常見的類神經網路模型，比較哪個結果誤差較低，其中包括ANN、LSTM、GRU、CNN、堆疊式LSTM、堆疊式GRU、CNN-LSTM和CNN-GRU，八個類神經網路模型。樣本期間為2010年到2019年，訓練期與預測期為5:1，使用python架構類神經網路模型，再加入其他因子於模型中，看是否能降低誤差，結果為LSTM與三因子模型為最好。
Stock market is one of the financial markets that the general public often comes into contact with. In order to use little capital to avoid non-systematic risks, index stock funds (Exchange Traded Fund, ETF) are a good choice. The management fees and transaction taxes charged by ETF Low and clear, mostly tracking index or market. Therefore, this study uses a neural network as the related research of ETF and tries to use the neural network to predict the reasonable return of the ETF on the day.
Compare whether the neural network can get better than the traditional Fama-French three-factor model Results and compare the common neural network models in related research which result is better, including ANN, LSTM, GRU, CNN, stacked LSTM, stacked GRU, CNN-LSTM and CNN-GRU, eight types of neural Net-work model. The sample period is from 2010 to 2019, the training period and the pre-diction period are 5:1. The python architecture neural network model is used, and other factors are added to the model to see if the error can be reduced. results are the long and short-term memory model and Fama French three-factor model has the best factor.
