NEXUS: A multi-modal framework for capturing financial news interactions in market forecasting

The incorporation of financial news into market prices is a subtle and multi-stage process, which complicates the assessment of how specific events in the news relate to subsequent market movements. We introduce NEXUS (News-Exchange Unified Signal Framework), a comprehensive architecture designed to capture both the intrinsic connections between financial news and market behavior, and the contextual relationships that exist among multiple news events. NEXUS fuses multi-modal inputs that combine numerical signals from markets with textual signals from news reports. We evaluate the approach on two large scale datasets that together span fifteen years of the S&P 500 and NASDAQ 100 and include more than 2 million news articles. Across these settings, NEXUS improves daily Sharpe ratios by 0.427 and 0.338 over advanced baselines for the two markets, indicating superior risk adjusted performance. Beyond accuracy, the results reveal three salient phenomena in the interaction between news and markets: the presence of delayed price assimilation, the persistence of informational effects over time, and the limitations of sentiment-based models in capturing deeper contextual cues. By explicitly modeling intra-day relational dependencies among news events and estimating their market-conditioned influence on prices, NEXUS provides a structured understanding of financial information propagation beyond aggregated multi-modal fusion.