<![CDATA[General Background: Stock price prediction is a complex problem due to the non-linear, stochastic, and volatile characteristics of financial markets. Specific Background: Advanced deep learning approaches such as Long Short-Term Memory (LSTM) and Transformer architectures have been applied to capture sequential patterns and global dependencies in time-series financial data. Knowledge Gap: However, existing approaches often lack integration between accurate forecasting and quantitative risk measurement within a unified framework. Aims: This study proposes a Hybrid Transformer–LSTM model integrated with Monte Carlo simulation to provide both precise stock price prediction and risk evaluation. Results: Using historical daily stock price data of BMRI from March 2013 to March 2025 and incorporating technical indicators such as RSI and moving averages, the model achieved a Mean Absolute Percentage Error of 4.13% and a Mean Absolute Error of 246.35 Rupiah. Monte Carlo-based Value at Risk at a 99% confidence level estimated a potential maximum loss of 5.35%. Novelty: The study combines sequential learning, attention mechanisms, and probabilistic simulation in a single framework linking prediction accuracy with risk quantification. Implications: The proposed approach provides a comprehensive analytical basis for supporting investment decision-making through reliable forecasting and measurable downside risk estimation. Highlights : Combined deep learning architecture produces low forecasting error on long-term historical data Probabilistic simulation quantifies maximum potential loss under high confidence level Integrated framework links predictive modeling with measurable investment risk Keywords: Hybrid Transformer LSTM, Stock Price Prediction, Monte Carlo Value at Risk]]>
