Large Language Models for Financial Risk Forecasting with Multimodal and Multi-Source Data: RiskLabs

Abstract

Artificial intelligence (AI), especially large language models (LLMs), is receiving growing attention in finance. Prior work has focused largely on financial text summarization, question answering, and stock movement prediction, while the use of LLMs for financial risk forecasting remains limited. We propose RiskLabs, a framework that uses LLMs to support financial risk analysis and prediction. RiskLabs integrates multimodal financial inputs, including earnings conference call transcripts and audio, market time-series signals, and background news. Empirical results show that RiskLabs is effective for forecasting market volatility and risk measures. Comparative experiments quantify the contribution of different data sources and clarify the role of LLMs in the forecasting pipeline. We also discuss the main challenges of LLM-based financial risk forecasting and the opportunities created by multimodal integration.