Analytical Study Of Machine Learning Models For Stock Trading In Malaysian Market

Nowadays, Machine Learning (ML) can serve as one of the solutions to accelerate the process of decision-making in forecasting daily stock market price movements. Nonetheless, inadequate number of research and lack of extensive data analysis using various ML models had limit the investors to appreciate the efficiency and capability of these models. Previous studies usually concentrate on the forecasting stock index or selecting a few stocks with restricted features. Therefore, this study focused to contribute on evaluating different algorithm models such as traditional ML and deep learning models with big stock data of multiple parameters from selected companies in Bursa Malaysia. The three traditional ML selected includes Logistic Regression (LR), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGB), while another three deep learning models selected are Deep Belief Network (DBN), Multilayer Perception (MLP), and Stacked Auto-Encoder (SAE). By setting the ML algorithms and their parameter along with using Walk-Forward Analysis (WFA) method, the algorithm design of trading signal was evaluated based on two groups of evaluation indicators, namely directional and performance. Comparative analysis of evaluation indicators for all trading algorithms has been assessed and discussed. For stock trading in Malaysian stock market particularly, the experimental results of this study demonstrate that deep learning models have better performance in directional evaluation indicator compared to traditional ML in time series forecasting. However, traditional ML models are more efficient than deep learning in performance evaluation indicators in terms of profitability and risk assessment.