نوع مقاله : مقاله علمی پژوهشی
نویسندگان
1 دانشکده علوم مهندسی، پردیس فنی، دانشگاه تهران
2 دانشکده علوم مهندسی، پردیس فنی، دانشگاه تهران، ایران
3 دانشکده حسابداری، پردیس مدیریت، دانشگاه تهران
4 دانشکده مدیریت، دانشگاه میبد
چکیده
کلیدواژه‌ها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Objective: This study aims to measure the efficiency of the Iranian capital market and investigate the ability of hybrid machine learning models to predict the direction of the Tehran Stock Exchange's total index. Also, evaluating the importance of factors affecting the index's predictability and explainability in machine learning models is another objective of this study.
Method: In order to analyze market efficiency at two weak and semi-strong levels, data related to the Tehran Stock Exchange's total index was used in the five-year period from 2019 to 2024. The proposed hybrid model includes the boosted gradient model (XGBoost), which has been improved by optimizing hyperparameters through the Genetic Algorithm (GA). The performance of this hybrid model has been statistically compared with other machine learning algorithms including boosted gradient, random forest, support vector machine, and logistic regression. Also, to increase the explainability of the model and analyze the importance of input variables in predicting the direction of the index, the SHAP method has been used.
Findings: The results showed that the XGBoost-GA model performed statistically better than other comparative models with an accuracy of 84% in predicting the direction of the Tehran Stock Exchange index. Comparing the results at different levels of market efficiency showed that at the semi-strong level, adding fundamental variables to the forecasting model improved the accuracy, which indicates the effect of fundamental information on predicting the direction of the index and, as a result, market inefficiency at this level. Also, at the weak level, the machine learning model based on technical data performed better than the random model, which is also an indication of market inefficiency at this level. In addition, the explainability analysis of the model using SHAP showed that the effect of variables on predicting the direction of the index varies depending on the type of input data. In the purely technical model, factors related to price behavior and short-term fluctuations such as the relative strength index (RSI), trading volume, and moving average divergence and convergence played a key role. In contrast, in the hybrid model that included fundamental and technical data, in addition to technical variables, factors such as real and legal liquidity inflows, gold prices, and financial indicators of companies such as return on assets (RoA) and return on equity (RoE) had a significant impact.
Conclusion: The results of the study show that the Iranian capital market is inefficient at both weak and semi-strong levels, and the predictability of the aggregate index is possible using technical and fundamental data. The analysis of the importance of input variables shows that some technical, fundamental, and macroeconomic indicators play a more important role in predicting market behavior, which can help make more informed investment decisions and better understand the behavior of machine learning models in predicting financial time series.
Keywords: Market efficiency theory, genetic algorithm, machine learning explainability, extreme gradient boosting, Tehran Stock Exchange aggregate index.
Send feedback
کلیدواژه‌ها [English]
)