A Comparative Study of Financial Time Series Forecasting Using Machine Learning and Traditional Statistical Methods - An Application To Stock Market Data

Machine learning, as a subtopic of artificial intelligence, has powerfully been applied in multiple industries increasingly. With its ever-increasing prediction power combined with conventional statistical methods, the investment industry makes use of artificial intelligence in making a decision and making more robust inferences. Today, AI is certainly a key phenomenon in quantitative investment analysis. In the centre of quantitative investment analysis, financial asset price prediction using machine learning algorithms, has subjected to a great number of studies. Much as the stock market behaviour is uncertain and hard to model due to its inherent exposure to numerous factors, we aim to predict Hong Kong Exchanges and Clearing Limited stock prices based on historical prices taking advantage of state-of-art machine learning algorithms in this study. We implement three supervised learning algorithms that are Stochastic Gradient Descent based Linear Regression, Random Forest Regression, Support Vector Regression, and a deep learning algorithm, Multi-layer Perceptron Regression, as well as a traditional statistical model namely Autoregressive Integrated Moving Average and Facebook’s Prophet on Hong Kong Exchanges and Clearing Limited close prices. By adopting a suitable feature set, meticulous data pre-processing, and training optimization, our findings demonstrate that machine learning algorithms along with the automated Autoregressive Integrated Moving Average model provide promising results in predicting financial time series, more specifically stock market data. With that being said, Facebook’s Prophet model needs further improvement in predicting time series.