Statistical and Machine Learning Methods for Stock Price Prediction

Abstract

Financial markets are very dynamic and thus difficult to predict the price of stocks. There are financial theories like the efficient market hypothesis, which indicate that the future prices of stocks are hard to predict. Nonetheless, empirical research also suggests that there can be some patterns and inefficiencies in financial markets. In this paper, the author reviews and analyzes the typical methods that researchers apply to stock price predictions, such as traditional statistical methods, machine learning methods, and deep learning methods. The paper is aimed at comparing the various methods used to analyze financial time series data, including ARIMA and GARCH models, and machine learning algorithms, including the support vector machine, random forest, and neural network. Besides, the latest trends regarding the implementation of deep learning and the utilization of alternative data are explained to give a more comprehensive outlook on the contemporary prediction frameworks. According to the existing literature, machine learning and deep learning models are likely to be more successful in attaining the nonlinear relationship and the temporal dependence in stock prices than traditional statistical models. Nevertheless, their performance is very dependent on the quality of data, selecting features, and the market conditions.