A comparative study of the performance of Stock trading strategies based on LGBM and CatBoost algorithms.

Today, investment in the stock market requires novel and efficient methods along with effective trading strategies for more accurate prediction of stock price future movements. This paper compares the performance of implementing LGBM and CatBoost trading strategies on a portfolio, which is formed, based on fundamental analysis and future study. First with use of future study and expert’s opinion, stock market scenarios designed and a portfolio consist of 6 fundamental stocks is built. In next step for each selected stocks a model is developed by means of LGMB and CatBoost algorithms and related stocks data from 2014 to 2019 to predict stock price movement. Model inputs includes, technical indicators, stocks trading data and some market and fundamental index. Bayesian hyper parameter was used to optimize the model’s key parameters. Results show that models optimized with Bayesian hyper parameter are more accurate than models, which optimized with grid search and implementing short-term trading strategies based on gradient boosting machine (LGBM) prediction signals cause better performance in comparison with CatBoost based strategies and Tehran Stock Exchange Index.