What do sovereign bond investors know about the risks and costs of violent conflict? Do they rationally incorporate available information, or are they overly-optimistic – or pessimistic – about the economic effects of political violence? To answer these questions, we estimate event-studies using daily sovereign bond trading prices and information on violent armed conflicts over the past two decades. We show that bond prices fall by an average of 0.7 points after the onset of state-involved conflict. Using reduced-form parameters, we calibrate a bond pricing model which implies both under-reaction and investor learning: the share of the shock that is priced in rises from 14% initially to 75% after 15 days. Consistent with the model, effects are larger where priors are optimistic because of recent peacefulness. Prices also respond more to severe outbreaks of violence near the capital city, and to center-seeking conflicts where rebels threaten the state. The magnitudes of these heterogeneous responses imply accurate beliefs about the process of conflict damages. The results suggest that bondholders have a nuanced understanding of the underlying political and spatial dimensions of conflict.