Policymakers frequently use guarantees to mitigate panic‐based runs in the financial system. We analyze a binary‐action coordination game under the global games framework and propose a novel intervention program that screens investors based on their heterogeneous beliefs about the system's stability. The program only attracts investors who are at the margin of running, and their participation boosts all investors' confidence in the financial system. Compared with government guarantee programs, our proposed program is as effective at mitigating runs but features two advantages: it costs less to implement and it is robust to moral hazard.

We show that larger trades incur lower trading costs in government bond markets (“size discount”), but costs increase in trade size after controlling for client identity (“size penalty”). The size discount is driven by the cross-client variation of larger traders obtaining better prices, consistent with theories of trading with imperfect competition. The size penalty, driven by the within-client variation, is larger for corporate bonds, during major macroeconomic surprises and during COVID-19. These differences are larger among more sophisticated clients, consistent with information-based theories.