Researchers have extensively tried machine learning algorithms in news classification and related quantitative finance domains in the past. Stock market investors look forward to being able to predict stock prices successfully not only to get the best returns but also to minimize the risk of losses with a forecast of stock prices and movement of the stock exchange depending upon the type of news. In this paper, we hypothesize that any news that comes to the market can broadly be classified into two types: Class A- News that has an effect such that it leads to a rise in the stock prices of the reference stock and a fall in the stock prices of its competitor stocks, or vice versa, and Class B- News that results in a simultaneous surge or decline in the stock prices of the reference stocks and its competitor stocks alike. This study is an effort to mathematically validate this hypothesis. 
This domain hasn’t been explored, and through our work, we try to demonstrate the capability of the existence of a pattern in the market, which could then be used for building automated trading strategies. We also adopt a unique approach to model the data as a supervised machine learning problem and by solving, on obtaining an accuracy of 66.5% we prove that such patterns exist and further suggest research inputs on ideas derived from this. 
 

Rakshika Gupta

Papers

Quantifying Stock News Relevance in Indian Markets

Panel Data Investigation of Determinants of Profitability: Evidence from Indian Corporate Sector