Machine learning research in quantitative finance

Abstract

Generating consistent returns through efficiently trading in highly liquid secondary financial markets has been a highly sought after goal since the advent of the first global bourse, the Amsterdam Stock Exchange, in 1602 by the Dutch East India Company. As part of our research project, we attempt to pursue this challenging goal through machine learning research in quantitative finance. In pursuit of our objective, we develop a proprietary non-linear machine learning based pattern recognition algorithm for automatically generating high to medium frequency trading signals based on the statistical analysis of historical market data. The prime motivation behind such automated signals for trading is that the information based on fundamental analysis is already incorporated in the price of a particular asset. The asset classes included in our analysis are major commodities and stocks indices. To verify the true potential of our research output, we implement our trading strategy in the live market during the Global Oil Crisis 2015by executing 34 trades on the Multi Commodity Exchange of India Limited (MCX) during the Global Oil Crisis from July 3 to July 23, 2015. In such a highly uncertain market scenario, our research-oriented strategy generates a real net cumulative ROI of 135.14% per contract margin with a hit rate of 100%. It outperformed the market by taking primarily long positions in a real market crash, thereby exhibiting a low correlation with the market, and hence, a low ?. Our research seems to have realized its objective of building a portfolio with low ?, and hence yielding a potential to statistically deliver consistent returns in the future.