International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 10 | Oct 2025
p-ISSN: 2395-0072
www.irjet.net
Financial Market forecasting using Reinforcement Learning : A Survey Rushikesh S. Mankape1, Prof. Pramila M. Chawan2 1M.Tech Student, Dept of Computer Engineering and IT, VJTI College, Mumbai, Maharashtra, India.
2Associate Professor, Dept of Computer Engineering and IT, VJTI College, Mumbai, Maharashtra,India.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Financial markets represent some of the most
constantly evolve, allowing traders to adapt strategies to unpredictable market behavior.
complex and dynamic environments, where asset prices fluctuate in response to economic, political and behavioral factors. Traditional quantitative trading models often rely on static statistical assumptions, making them less effective during sudden market shifts or highly volatile periods. Quantitative Artificial Intelligence (Quantitative AI) merges mathematical modeling and artificial intelligence to develop self-learning systems that can trade autonomously in real time. This paper surveys recent developments in quantitative AI, with an emphasis on deep reinforcement learning (DRL) algorithms such as deep Q-networks (DQN), policy gradient (PG), and advantage actor-critic (A2C). The goal of these models is to optimize trading strategies, minimize draw-down, and achieve high risk-adjusted returns. We also propose a set of PPO, A2C and DDPG algorithms to enhance the robustness and adaptability of the automated trading system. Challenges such as interpretability, generalization, and data paucity are discussed, along with directions for future financial AI research .
The purpose of this study is to explore how DRL algorithms can be effectively used to optimize financial trading strategies. We review key developments in quantitative AI, analyze several DRL models, and propose a unified approach that integrates multiple learning agents for improved performance.
2. BACKGROUND Quantitative finance refers to the use of mathematical models, statistical techniques, and computational methods to analyze financial markets. Traditionally, models such as linear regression, auto-regressive integrated moving average (ARIMA), and GARCH were used to predict stock prices or volatility. However, these models assume data consistency, which is unrealistic in real-world business environments. As the volume and complexity of market data increased, machine learning emerged as a powerful alternative.
Key Words: Quantitative Finance, Deep Reinforcement Learning, DQN, A2C, Policy Gradient, Automated Trading, Financial Markets, Artificial Intelligence
Machine learning (ML) introduced the ability to recognize nonlinear patterns in price data. Models such as Random Forest, Support Vector Machine (SVM), and neural networks were used to classify market conditions and predict price movements. Nevertheless, these methods cannot account for sequential decision making and longterm reward adaptation.
1.INTRODUCTION Financial markets are one of the most data-intensive and dynamic environments where asset prices are influenced by countless variables. Traditional quantitative trading strategies, including moving averages, time-series momentum, and regression-based forecasting, operate using predefined rules and assumptions about data consistency. These methods struggle when the market experiences sudden changes such as crashes, political instability, or high-frequency trading spikes. As a result, traders and researchers have shifted toward the integration of Artificial Intelligence (AI) into quantitative trading systems to enable adaptability and automation.
Deep reinforcement learning (DRL) overcomes these limitations by combining reinforcement learning principles with deep neural networks. A DRL agent interacts with the market environment, taking actions such as buying, selling or holding based on historical data and receiving rewards based on portfolio performance. Through trial and error, the agent learns optimal policies to maximize long-term returns. Popular DRL algorithms include deep queue-network (DQN), policy gradient (PG), and advantage actor-critic (A2C).
Artificial intelligence enables computers to learn from data and improve their performance without explicit programming. Specifically, Deep Reinforcement Learning (DRL) allows algorithms to make decisions through experience, rewarding good trading actions and punishing bad ones. Quantitative AI combines AI and finance to create intelligent systems capable of understanding complex market patterns and making real-time trading decisions. Unlike traditional models, DRL-based systems
© 2025, IRJET
|
Impact Factor value: 8.315
3. LITERATURE SURVEY In recent years, Reinforcement Learning (RL) has emerged as a transformative tool for algorithmic trading by enabling computational models to develop self-improving trading strategies through direct interaction with market data. Unlike conventional quantitative techniques that
|
ISO 9001:2008 Certified Journal
|
Page 732