Multi-agent reinforcement learning approach for hedging portfolio problem.

Area of Science:

• * Computational Finance
• * Artificial Intelligence
• * Financial Market Analysis

Background:

• * Hedging stock portfolios is challenging due to hedge costs and limited derivative options, particularly in emerging markets like Vietnam.
• * Cross-hedging is often necessary in Vietnam due to the absence of stock-specific put options.
• * Existing VN30 stock index futures contracts are the primary tool for hedging strategies.

Purpose of the Study:

• * To explore the feasibility of developing an automated hedging strategy using cooperative multi-agent reinforcement learning.
• * To construct a hedging strategy without requiring advanced financial domain knowledge.
• * To mitigate risks and reduce losses for stock portfolios in the Vietnamese market.

Main Methods:

• * Development of a stock market simulator incorporating transaction fees, taxes, and settlement dates.
• * Utilization of daily stock returns as input data for the reinforcement learning agent.
• * Training cooperative multi-agent reinforcement learning agents on historical stock data.

Main Results:

• * The trained agent learned effective trading and hedging policies, leading to profit generation and loss reduction.
• * The hedging strategy demonstrated success in protecting portfolios, even during systematic market collapses.
• * The system proved capable of managing portfolio risk in volatile market conditions.

Conclusions:

• * Cooperative multi-agent reinforcement learning offers a viable approach for automated hedging strategy development.
• * This method can significantly enhance investment performance and risk management for Vietnamese stock market investors.
• * The AI-driven strategy provides a robust solution for controlling losses in unpredictable market environments.