Development of an AI-Driven Trading Bot for XAU Using Sentiment Analysis and LSTM Models

Abstract

Presented in this paper is the challenging task of building a profitable, risk management-focused automated trading system for the volatile gold market, which can be generalized to the forex market. It is argued that the failure of mono-modal trading models necessitates a multi-modal approach. The main system, based on a Long ShortTerm Memory (LSTM) network and augmented by technical and external market data, is managed by a deterministic, rule-based execution logic. This system demonstrated significant profitability in comprehensive 5-year back-testing (2020-2025), turning a $1,000 deposit into $75,277 ($74,277 profit). Following this, forward testing (April 1-22, 2025) achieved $8,638 from $1,000 in 22 days, and a successful one-month live test generated approximately $1,678 profit. As the primary predictive method is aggressive, a secondary model, or "brain" of the system, was developed as an optional submodule using a Deep Q-Network (DQN) Reinforcement Learning agent. This agent, while more risk-averse, achieved a consistent Sharpe Ratio of 3.40, demonstrating its potential as a complementary tool for risk monitoring. Ultimately, this study demonstrates the viability of dual architecture, with its characteristics dictating its operational mode.

Description

Citation

Perera, K.N.S.D.; Buddhika, P. (2025) Development of an AI-Driven Trading Bot for XAU Using Sentiment Analysis and LSTM Models, Technology Innovations & Digitalization Accompanying Creativity (TIDAC), NSBM Green University, Sri Lanka. P.328

Endorsement

Review

Supplemented By

Referenced By