A comparative study on machine learning based trading simulation in Stock Exchange of Thailand (SET)

Financial return of stock trading is a difficult task which the investors have to know and notice various factors related to stock price to handling profit and loss in their portfolio. The objective of this research is to build machine learning and compares the portfolio value from predicting stock price and buy/sell signal comparing two baseline strategies through simulation test. To observe the performance of each machine learning will be related with the profit of trading that the investor can get any advantage before make decision and investing in stock exchange. This research conducts three machine learning techniques which are XGBoost, Long Short Term Memory (LSTM), and Deep Q-Network Learning (DQN). The result of this study shows that in first baseline strategy, LSTM appears the most profit and XGBoost appears the most profit in second baseline strategy.

股票交易的收益决策是一项极具挑战性的任务，投资者需充分掌握并考量所有与股价相关的要素，以此管控投资组合中的盈亏状况。本研究旨在构建机器学习模型，通过模拟测试对比两种基准策略下，基于股价预测与买卖信号生成的投资组合价值。本研究的核心目标之一是观测各机器学习模型的表现与交易收益的关联，助力投资者在做出交易决策、参与证券市场投资前获取竞争优势。本研究采用三种机器学习模型：极限梯度提升树（XGBoost）、长短期记忆网络（Long Short Term Memory，简称LSTM）以及深度Q网络学习（Deep Q-Network Learning，简称DQN）。研究结果表明：在第一种基准策略下，长短期记忆网络的收益表现最优；在第二种基准策略下，极限梯度提升树的收益表现最优。