Topics in semiparametric and nonparametric models with thresholds

Chapter 1. We develop a nonparametric model with unknown threshold, under random sampling, which extends the previous theory developed by Tong (1982), Chen (1993) and Hansen (2000). We allow shifts in nonparametric curves according to a threshold variable relative to an unknown benchmark. An estimator is built for the threshold and the nonparametric term. Consistency of both the estimator for the threshold parameter and the estimator of the nonlinear function is proven. Monte Carlo simulations show that under presence of a threshold variable, Nadayara-Watson estimator exhibits a bias even as data size increases, while in our estimator such bias is not depicted.;Chapter 2. We propose to model capital flows by a Markov Switching Model and provide a theoretical framework for this econometric Model. Calvo and Reinhart (1991) differentiated the impact of capital flows by its source: global (push effect) or domestic (pull effect). These effects are often blurry but become evident under financial turmoil. Previous research defined capital flows' subjectively. Therefore, difference in findings can be attributed to difference in the definition used. We move from this approach by implementing a data based Markov Switching model to identify turmoil episodes and then estimate the effects of the determinants in those selected episodes through a dynamic panel data regression. Our results suggest that push and pull effects are paramount to Capital Flows in contrast to previous finding were only global effects mattered.;Chapter 3. We develop a new semiparametric model with unknown threshold under random sampling. The threshold, the coefficients and the nonparametric functions in our model are estimated simultaneously. Consistency of the estimator of the threshold parameter, of the model coefficients and of the nonlinear function is proven using empirical process theory. Monte Carlo simulations showed that under presence of a threshold variable, Robinson's semiparametric regression exhibits a bias that persists even as data size increases, while our estimator corrects this bias. Finally, we used our estimator on an application on capital flows. We found that an expansionary monetary policy is related to a positive effect on capital inflows, and this relation is stronger under episodes of turmoil.