Scholarlink

The stock market has experienced significant growth and transformation in recent years, which has created an optimal environment for the development of innovative modelling techniques. Rational Macro Agent-Based Modelling (RMABM) is a method that integrates the rational decision-making concepts of economic theory into agent-based modelling. However, the current models frequently misrepresent real-world dynamics because they only include restricted rational and bounded-rational actors. The primary objective of this research is to construct a stock exchange platform that employs RMABM to simulate dynamic trading environments in which fictitious equities are traded by rational and bounded-rational actors. The work involves the training of these agents using historical Bombay Stock Exchange (BSE) stock values, utilising Double Deep Q-Network (DDQN) algorithms and transformers for decision-making. In order to achieve this, the news agents that employ Financial Bidirectional Encoder Representations from Transformers (FinBERT’s) sentiment analysis capabilities has been implemented to analyse company-specific news using Large Language Model Meta AI (LLaMA 3.1). BSE’s historical stock price data is employed to conduct training. The evaluation criteria include the accuracy of predictions, cumulative reward graph values and the loss curves.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Realistic Macroscopic Players for Predictive Analysis to Balance Stock Market Dynamics Using FinBERT and DDQN

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • M. Srivani,
  • Magesh Boopathi,
  • Aashiq

摘要

The stock market has experienced significant growth and transformation in recent years, which has created an optimal environment for the development of innovative modelling techniques. Rational Macro Agent-Based Modelling (RMABM) is a method that integrates the rational decision-making concepts of economic theory into agent-based modelling. However, the current models frequently misrepresent real-world dynamics because they only include restricted rational and bounded-rational actors. The primary objective of this research is to construct a stock exchange platform that employs RMABM to simulate dynamic trading environments in which fictitious equities are traded by rational and bounded-rational actors. The work involves the training of these agents using historical Bombay Stock Exchange (BSE) stock values, utilising Double Deep Q-Network (DDQN) algorithms and transformers for decision-making. In order to achieve this, the news agents that employ Financial Bidirectional Encoder Representations from Transformers (FinBERT’s) sentiment analysis capabilities has been implemented to analyse company-specific news using Large Language Model Meta AI (LLaMA 3.1). BSE’s historical stock price data is employed to conduct training. The evaluation criteria include the accuracy of predictions, cumulative reward graph values and the loss curves.