A comparison of statistical and machine learning models for stock price prediction

A huge proportion of money around the world is held by the stock markets. It is one of the most pivotal aspects of the financial institutions and experts. Predicting the movements of stock markets can improve decision making for traders. In this paper, data science techniques are employed to predict the 'closing prices' of stocks. Performance of Facebook Prophet (Fb-P), auto-regressive integrated moving average (ARIMA), simple exponential smoothing (SES), vector autoregression (VAR), random forest (RF), linear regression (LR), k-nearest neighbours (k-NN), decision tree (DT), extreme gradient boost (XGB), and long short-term memory (LSTM) are examined on stock price data of companies listed on the Bombay Stock Exchange (BSE). The proposed work concludes that for extended time frames, LR model performs the best followed by LSTM and ARIMA when compared on the same metrics.