A Shock to Remember: Dynamic Stock-Market Responses to Armed Conflict

This study examines how stock markets dynamically adjust to armed-conflict announcements, moving beyond the conventional event-study focus on terminal cumulative abnormal returns. Using a nine-event archive covering conflict-related announcements between 2003 and 2022, we analyze 81 event–sector–measure trajectories for U.S.-listed firms in the aerospace and defense, oil and gas, and alternative energy sectors. The 2022 Ukraine event has observations through the [0,20] window but no [0,25] endpoint; accordingly, analyses requiring the 25-day endpoint use eight events and 72 series, whereas analyses that can accommodate the unequal trajectory lengths use all nine events and 81 series. Each trajectory captures cumulative abnormal performance across successive post-event windows, representing the timing, persistence, and direction of market adjustment. Dynamic time warping combined with average-linkage hierarchical clustering identifies three adjustment patterns: persistent amplification, transitory responses, and directional reversal. Persistent amplification characterizes 53% of trajectories, while 21% exhibit front-loaded responses that subsequently attenuate toward zero and 26% reverse direction over the post-event period. These classes are defined relative to the direction of the initial response and are sensitive to this normalization. The analysis further examines differences between conflict initiation and termination and across sectors. Given the limited number of independent conflict episodes, inference is conducted at the conflict level using exact permutation procedures and Bayesian hierarchical models with conflict-level random effects. In the eight-event endpoint sample, reversal is more frequent after conflict initiation than after termination (48.9% versus 14.8% of series), and the hierarchical model gives this difference a posterior probability of 0.974. The exact conflict-level test gives p = 0.071; no primary result survives a correction for multiple testing, and simulated power with eight conflicts is approximately one-half under the stated assumptions. The finding is therefore consistent in direction but statistically fragile. Reaction magnitude does not differ systematically by phase, and the timing and persistence of market responses vary considerably. These findings demonstrate that terminal event-study measures can mask economically distinct adjustment processes and highlight the value of modeling abnormal returns as dynamic trajectories.