Essays on bootstrap methods in econometrics

This thesis proposes novel implementations of the bootstrap in econometrics. Chapter 1 explores bootstrap methods for test statistics showing an asymptotic bias which is difficult, or impossible, to estimate, proposing modifications of standard bootstrap methods delivering valid inference, totally bypassing bias estimation. Chapter 2 develops enhanced inference techniques for nonparametric regression and regression-discontinuity designs, introducing novel bootstrap approaches for debiasing with greater efficiency than the current state-of-the-art. Chapter 3 tackles the problem of invalidity of “standard” bootstrap methods in a predictive regression setup, when the predictability parameter may lie on the boundary of the parameter space, proposing a modified approach restoring bootstrap validity. Chapter 4 investigates the flattening of the Phillips Curve, presenting a time-varying structural estimation framework to disentangle underlying drivers of macroeconomic shifts. Finally, Chapter 5 contributes to robust inference on stochastic time-varying coefficients, proposing new confidence intervals which are robust to “large” – and more efficient – bandwidths. Collectively, these contributions advance theoretical and practical econometric tools for addressing complex real-world economic problems.