Financial crises in developing countries

This dissertation provides both a theoretical and empirical look at financial crises in developing countries. The first chapter examines the effects that capital inflows have on the financial system in the context of a demand deposit banking model. In this environment, an adverse-selection problem arises where short-term capital has the incentive to enter the domestic banking system while long-term capital chooses to stay out. Then, short-term capital flows limit the risk-sharing function of banks. As short-term inflows increase, a threshold is reached beyond which it becomes optimal to restrict capital inflows. In addition, if the quantity of inflows is unknown, then banking crises occur as short-term inflows become large. In this case, the bank’s insurance function is lost and assets have to be suboptimally liquidated. In spite of this, restricting capital inflows may not be optimal at all times, since the cost of doing so may be greater than the detriment of allowing them in. The second chapter considers policy design in a banking environment where both fundamental runs (that stress macroeconomic variables, such as negative technology shocks, as the cause of bank runs) and sunspot runs (where self-fulfilling expectations generate equilibria where agents panic and run on banks) are possible. Under this environment, policies of narrow banking and suspension of convertibility will not be optimal. In contrast, a lender of last resort mechanism, where a central bank lends currency to banks in the event of a run, achieves the optimal outcome by preventing costly liquidation of investments and optimally distributing risk when there are runs. While the second chapter models both types of runs under one environment, the third chapter uses a multinomial logit model that differentiates both types of runs to study the factors associated with the emergence of financial crises. By doing this, important characteristics particular to each type of run come to light which are not accounted for by standard binomial logit specifications. We find evidence indicating that the two types of crises are indeed different, and are explained by different variables. Finally, by accounting for both types of crises, our results provide better support to existing self-fulfilling theoretical models.