Stock Market Forecasting using Ensemble Learning and Statistical Indicators

Abstract

With a volume of 2 billion+ trades per day and a market capitalization of 2.56 trillion USD the national stock exchange (NSE), India is one of the largest stock exchanges in the world. Every day the value of stocks, commodities, bonds and futures fluctuate inducing volatility and forecasting these fluctuations to make money requires deep knowledge about the market and their historical data. Thus, a simple time series forecasting model is not enough to predict future movements as we need to know about the market sentiment, trend and industry fundamentals to bolster our stand of declaring a stock or commodity as bearish or bullish. In this research, using machine learning forecasting models like Attention integrated Long Short Term Memory (LSTM) Model and a Reinforcement Learning agent coupled with statistical indicators and trading strategies like Auto Regression Integrated Moving Average (ARIMA), Prophet, Momentum trading and Pairwise trading to quantify the trend and market sentiment an approach to predict movements is devised. Using this approach increases the accuracy of stand-alone algorithms and helps in generating a cumulative analysis of the stock on the basis of itself and its stock universe data.