分析資訊不對稱對K線圖形預測股票投資組合報酬率之影響

Using Candlestick Chart to Predict Return Rates of the Stocks Portfolio: The Impact of Information Asymmetry

本研究透過卷積神經網路辨識K線圖形並建立模型，預測股票投資組合報酬率在資訊不對稱下是否具有顯著的超額報酬。研究樣本為台灣在2010年1月4日到2017年12月29日上市交易公司共八年之資料。本研究對股價資料進行影像特徵擷取並預測股票報酬率，且預測買賣價差與市場深度用以捕捉資訊不對稱。其中以報酬率與資訊不對稱因子（買賣價差和市場深度）形成投資組合，以檢定投資組合在考量風險因子後，是否具有顯著超額報酬。實證結果顯示在資訊不對稱預測值低時，預測高報酬率投資組合才具有顯著超額報酬。此外，從穩健性測試發現前期資訊不對稱值高時，則下一期資訊不對稱值會反轉變小，則預測下期高報酬率投資組合具有顯著超額報酬。

This study employs the convolutional neural network to identify the candlestick chart, construct a prediction model, and then forecast whether the stocks portfolio obtains significant excess returns under the information asymmetric. The samples are the companies that were listed on Taiwan Stock Exchange from January 4, 2010, to December 29, 2017. We predict the stocks’returns by identifying patterns of stock prices, bid-ask spread, and market depth to capture the information asymmetry. Furthermore, we use the return rates and the information asymmetry factors to construct the portfolios and verify whether the portfolios can receive significant excess returns when considering the risk factors. The empirical results show that while low predicted information asymmetry, the predicted high return rates portfolios can receive significant excess returns. Moreover, we conducted the robustness test and figured out that high information asymmetry in the earlier stage will result in lower information asymmetry value in the following stage, and the predicted high return rates portfolios can receive significant excess returns.