A Study on Impact of News Data on Share Market

Abstract

The stock exchange is a highly temporal market where the emergence of new information newlineinfluences the prices of the stocks. The source of this information is numeric and nonnumeric newlinedata. The efficient market hypothesis states that the share price is always fair newlineand reflects all the publicly available information in the market. Thus, the strategy of the newlineinvestors is to use their heuristic knowledge to detect the hidden signals from the data newlineand decide their entry and exit positions, which will lead to profit. Using numeric data for newlinetrading in the share market is a well-studied problem. However, non-numeric data consists newlineof hard-to-quantify information, which influences the investors and, in turn, influences the newlineshare market. The use of non-numeric data for trading is an under-explored area. This newlinethesis investigates the effect of non-numeric data, specifically news data, for trading in newlinethe share market. Further, this thesis explores the problem of share trading from an AIbased newlineperspective. We divide the problem of share trading using news data into two subproblems: newlineshare market prediction and algorithmic trading. In this thesis, we perform all newlineour experiments in the Indian stock market. We also create a benchmark dataset consisting newlineof financial news articles related to the Indian stock market, historical day-wise prices of newlinebenchmark indices, and minute-wise prices of the NIFTY 50 benchmark index. We use newlinethis dataset for all our experiments. newlineIn share market prediction, we examine the use of news data to predict the close price newlinemovement. In this study, we examine the relationship between news articles and the direction newlineof close price movement of the stock market. We explore the effectiveness of different newlinestatistical and deep learning-based classifiers that use price data or news data as input to newlinepredict the direction of close price movement. We also explore the effectiveness of using newlinedifferent text representation schemes to represent news data and examine their effect onprediction performance. newline