利用情緒指標及長短期記憶模型預測台灣加權股價指數

Predicting Stock Prices with Sentiment Indicator and the LSTM : Evidence form Taiwan Stock Market

張士琳(Shih Lin, Chang)

指導教授 ： 王致怡

國立臺北商業大學/財經學院/財務金融系研究所/碩士(2018年)

摘要

本研究利用LSTM神經網路，結合價量指標及情緒指標預測台灣加權股價指數隔日收盤價漲跌，研究樣本包含自2003年4月至2018年4月的日資料。輸入變數除基本價量指標外，尚包括ARMS、券資比、市場週轉率、以及機構法人買賣超與金額、買賣權成交量及未平倉量等情緒指標。本研究變數採用不同函數轉換方法及不同分類樣本，並將樣本區分成訓練集及測試集樣本，採移動視窗法，以滾動窗口的方法檢驗樣本內及樣本外準確度及預測績效。實證結果顯示，加入情緒變數後，有助提升夏普比率與準確度。其中以綜合考量買賣權未平倉及成交量為影響隔日大盤漲跌之重要因子，預測台灣加權指數隔天收盤價準確率近五成，樣本內外績效均較僅使用價量指標明顯提升。本研究將有助提供效率市場及行為財務實證的參考。

關鍵字

長短期記憶模型（LSTM） ； 深度學習 ； 情緒指標

並列摘要

This paper aims to use the Long Short Term Memory (LSTM) model to predict the next day’s direction of daily Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX). The sample covers the daily data for 15 years from April 2003 to April 2018. The input variables used in this study contains different conbinations from the price, volume and market sentiment variables, including the original data and transformed data. By applying a sliding window approach, this study evaluates the model starts with looking at its in-sample and out-of-the-sample accuracy and forcast performance across different data sets. The empirical results show that adding sentiment factors will improve the accuracy and forecast performance of predicting next day’s direction of TAIEX. In addition, this paper show that options-based sentiment indicators are the important factor influencing the forecast performance. This study will provide further insighs into market efficiency and behavioral finance fields.

並列關鍵字

Long Short-Term Memory (LSTM) ； Deep learning ； Sentiment indicator

參考文獻

一、 國外文獻

Baker, M., & Stein, J. C. (2004). Market liquidity as a sentiment indicator. Journal of Financial Markets, 7(3), 271-299. doi:10.1016/j.finmar.2003.11.005

Barber, B. M., & Odean, T. (2001). Boys will be boys: Gender, overconfidence, and common stock investment. The quarterly journal of economics, 116(1), 261-292.

Barberis, N., Shleifer, A., & Vishny, R. (1998). A model of investor sentiment1. Journal of financial economics, 49(3), 307-343.

Bergerson, K., & Wunsch, D. C. (1991). A commodity trading model based on a neural network-expert system hybrid. Paper presented at the Neural Networks, 1991., IJCNN-91-Seattle International Joint Conference on.