Generative and Stochastic Modeling for Financial Risk Management: Hybrid Approaches for Forecasting and Fraud Detection

Recent advances in generative artificial intelligence are reshaping financial risk management by enabling more flexible modeling of uncertainty, complex dynamics, and rare events. This study investigates hybrid modeling frameworks that integrate classical stochastic models with modern machine learning and generative AI techniques for financial forecasting and fraud detection. We focus on financial time series characterized by nonlinearity, volatility clustering, and structural breaks, which challenge traditional econometric approaches. The proposed framework combines stochastic processes (including ARIMA-type and volatility-based models) with machine learning methods such as ensemble learning and neural networks, alongside emerging generative models and explainable AI (XAI) tools. Generative components are leveraged to enhance scenario generation and stress testing, allowing for improved representation of tail risks and extreme market conditions. Machine learning models are optimized through systematic hyperparameter tuning to improve predictive accuracy and robustness across different forecasting horizons. Empirical results on real-world financial datasets demonstrate that hybrid generative–stochastic approaches outperform standalone classical or machine learning models in terms of forecasting accuracy, stability, and adaptability. In the context of fraud detection, the integration of generative models enhances the detection of anomalous and rare patterns, while XAI techniques improve model interpretability and transparency—critical requirements for regulatory compliance and informed decision-making in financial institutions. Overall, this work highlights the transformative role of generative AI as a paradigm shift in finance, offering scalable and interpretable solutions for forecasting, risk assessment, and fraud detection. The findings offer practical insights for financial practitioners and contribute to the growing body of literature at the intersection of generative modeling, machine learning, and financial risk management.