Estimation with Simulation: Theory and Application

该项目开发了使用模拟来克服数值复杂的计量经济学模型的计算负担的估计方法。这是一条非常重要的研究路线，因为许多经济问题即使使用超级计算机也是难以解决的。许多应用经济学家必须有意识地设计模型，以避免难以计算的方面，尽管这个问题的经济学自然会导致这个方向。这个项目推广了模拟矩方法，用它来解决复杂的经验经济问题，并扩展了研究人员在他之前的NSF拨款下为线性指数模型开发的一类新的半参数估计量。将模拟矩方法推广到条件多项式概率模型以外的有限因变量模型，以及与矩函数近似有关的推理问题，如对数似然得分。在计算方面的研究解决了模拟矩估计的存在性和不可微性。研究的第二部分是模拟矩估计方法在具有非线性税收的劳动力供给分析中的实证应用。第三部分是指数模型的半参数估计。给出了一类线性指数模型估计量的渐近分布理论，并讨论了这类估计量与其他估计量之间的关系。

This project develops estimation methods that use simulations to overcome the computational burdens of numerically complex econometric models. This is a very important line of research because many economic problems are intractable even using supercomputers. Many applied economists must consciously design models to avoid computationally intractable aspects, even though the economics of the issue naturally lead in that direction. This project generalizes the simulated moments methods, uses it to solve complicated empirical economic problems and extends a new class of semi-parametric estimators for linear index models developed by the investigator under his previous NSF grant. The method of simulated moments is generalized to limited dependent variable models besides the conditional multinomial Probit model and to inference problems associated with approximating moment functions like the score of the log- likelihood. The research in computation addresses the existence and nondifferentiability of simulated moments estimators. The second part of the research is an empirical application of simulated moments estimation methods to the analysis of labor supply with nonlinear taxes. The third part of the research is about semi-parametric estimation of index models. Asymptotic distribution theory for a class of estimators for linear index models will be derived and the relationship between this family of estimators and others is explored.