| Summary: | 碩士 === 輔仁大學 === 統計資訊學系應用統計碩士班 === 107 === Stocks are essential in financial instruments. Their importance is not only about investment, but also the basic way of operation in today’s social economy. Therefore, if we can efficiently analyze and predict stock price trends, investors will have the opportunity to profit.
Most of the research predicts stock based on daily data. Thus, this study explores whether high-frequency stock price data combined with recurrent neural network can have better prediction ability, and takes Taiwan50 ETF as a research object; the dataset uses daily data and high-frequency data spanning from August 13, 2015 to October 7, 2016. Besides, this study also adds in technical indicators as input variables, establishes high-frequency data with LSTM, BiLSTM, and GRU as experimental groups, and uses high-frequency data with RNN as the control group. The results show that the LSTM error value is the smallest, the MSE is 1.0142, the maximum error model is BiLSTM, and the MSE is 21.9834. In the simulated trading results, it is known that the profitability of using high-frequency data is better than daily data.
|
|---|
