Intl Conference on Economics, Finance & Business, Prague

THE PREDICTABILITY OF STOCK RETURNS BASED ON PRINCE AND VOLUME RELATIONSHIP CAPTURING BY CANDLESTICK CHART IMAGE: USING CONVOLUTIONAL NEURAL NETWORK TO ANALYZE IMAGE RECOGNITION

MINGCHANG WANG

Abstract:

This study tries to prove the predictability of stock return based on the price-and-volume relationship information capturing by candlestick chart image from using the CNN model. Using the trading data of listed firms in Taiwan Stoch Exchange from 2010 to 2017, CNN model is used in feature classification about future stock return to filter the information content of price-and-volume relationship stemming from the candlestick chart image data transforming from four sets of one-dimensional time series data such as the opening price, the highest price, the lowest price, and the closing price, and this paper examines the significant existence of abnormal return of price-and-volume portfolios established from the feature of candlestick chart image. Besides, comparing to prior mentioned price-and-volume portfolios established by candlestick chart image, this paper also examines the abnormal return of price-and-volume portfolios established from past actual volume information. According to our empirical evidences, the research finds that the establishing portfolios from price information of candlestick chart image simultaneously include the volume information of candlestick chart image or past actual volume are able to generate outperform abnormal return, representing that trading volume is an important factor in both the predicted value and the actual value for portfolios.

Keywords: Deep Learning, Convolutional Neural Network, Trading Volume, Price Volume, Candlestick Chart