This dissertation studies financial fragility caused by coordination failure and discusses plausible regulations to alleviate coordination problems and enhance social welfare. It consists of two chapters. In the first chapter, "Capital Flows in the Financial System and Supply of Credit," I study how capital flows in the financial system affect the coordination problem among banks in supplying credit to the real economy. When credit contraction raises concerns about an economic recession, the economy can end up in a self-fulfilling credit freeze that banks abstain from lending for fear that others would withhold lending. I show that capital flows across banks can alleviate the self-fulfilling credit freeze problem because banks that are prone to supply credit can borrow from other banks to extend more credit to the real economy. However, when the interest rates for interbank capital flows are low, they signal dim economic prospects and deter credit supply by banks. As a result, the economy can get stuck in an equilibrium with low interest rates. In such equilibrium, freezing interbank capital flows and real credit crunch co-exist and reinforce each other through a vicious feedback loop. This is consistent with the observations in the shallow post-crisis recovery—low real interest rates, contraction in wholesale funding markets, low credit growth, and sluggish economic growth. My model suggests that regulations addressing counter-party risks can be a remedy to prevent capital flows in the financial system from freezing, which breaks the vicious feedback loop and stabilizes the real credit market. This paper develops a model to study how capital flows in the financial system affect the coordination problem among banks in supplying credit to the real economy. In the second chapter, "Intervention with Screening in Global Games," joint with Junyuan Zou, we propose a novel intervention program to reduce coordination failure. Compared with conventional government-guarantee type programs, such as demand deposit insurance, ours incur a lower cost of implementation and suffers less from moral hazard problems. The proposed program effectively screens agents based on their heterogeneous beliefs of the coordination results. In equilibrium, only a small mass of ``pivotal agents" self-select to participate in the program. However, the effect is amplified by strategic complementarities, and coordination failure can be significantly reduced. We demonstrate the generality of the proposed program with applications in panic-based bank runs, debt rollover problems, self-fulfilling market freezes, and underinvestment problems in the real economy.