Capital flows, exchange rate policy and growth

The first chapter deals with the fact that capitals do not flow from rich to poor, but to middle-income countries. It uses data on 158 countries and 28 years to document this pattern. The author proposes an explanation that is based on the probability of default, where income level matters. This also accounts for the fact that poorer countries are perceived to be riskier. The model suggests that human capital should work as a capital attractor in middle-income countries but not in poor countries. The author uses non-parametric and parametric methods to test this prediction.; The second chapter proposes a definition of de facto fixed exchange rates. Using this definition (and an existing definition) the author shows that: (i) most fixed exchange rates are short lived; (ii) there is a positive association between pegs duration and devaluation size; and (iii) there has been a downward trend on the duration of pegs. Chapter three proposes a model that rationalizes the existence of short lived pegs as an optimal policy. Sticky inflation introduces a trade-off between low(er) inflation and real exchange rate misalignment. The author shows that if runs are costless then governments are forced to choose between a strong commitment and a flexible exchange rate. But, if there are some costs to run against the currency, there is space for optimal soft pegs and capital controls may be an optimal policy.; Chapter four investigates the effects of high-devaluations on growth. The author uses a large dataset and three methodologies: unconditional comparison, fixed effects regression and matching estimators. Matching estimators allow to control for bias due to selection on observables. The author combines matching estimators with diff-in-diffs to control also for selection on unobservables. The main results are that: (i) high-devaluation years are also years of lower growth; (ii) GDP tends to recover, with some evidence of full convergence after three years of the event; (iii) not controlling for selection on observables could lead us to conclude that the slowdown starts previous to the event and is not caused by the event; but (iv) matching estimators still show evidence of a contractionary effect of high-devaluations.