This dissertation focuses on some recent developments in econometric analysis and on the use of econometric analysis as an instrument for proper economic policy evaluation. The first chapter of this thesis proposes a framework for modeling the nonlinear dynamic in time series, and a specification of this framework is applied to the macro series of U.S. GNP and unemployment. The second chapter develops an inference procedure that is useful in testing the significance of some forms of nonlinearity and solves the problem of inference in the presence of nuisance parameters that are not present under the null of the test. In the third chapter, some simulation methods recently introduced in econometric analysis, Markov Chain Monte Carlo methods, are discussed and some strategies to improve their performance are proposed. The final chapter of this thesis investigates the effect of the aggregation of heterogeneous agents on the estimation of structural parameters using aggregate macro data. The results of the last chapter suggests that a consistent procedure for policy evaluation should consider not only the effect of the policy action on the forward-looking behavior of the agents but also the effect of policy changes on the cross-section distribution of the economic agents.
