Nonparametric Methods for Economic Models

9410182 Matzkin大多数现有的经济模型测试都是首先进行的，规定了模型中的函数，直到某些参数。接下来，估计参数，并对模型所暗示的对估计参数值的限制进行测试。这一过程通常有两个缺点。首先，如果限制被拒绝，目前还不清楚这是因为错误的模型还是错误的参数规范。其次，如果这些限制没有被拒绝，目前还不清楚这是否意味着有任何有力的证据支持该模式，因为这些限制只是必要的，而不是充分的。这个项目通过开发识别、估计和测试非参数模型的新方法来解决这些问题。首先，该项目将确定如何使用几种类型的限制，如不等式限制和优化行为，来保证非参数联立方程系统中函数的识别。将考虑的系统包括那些具有不可观测的内生变量、离散变量和不可微函数的系统。这项研究的结果将有助于计量经济学家对估计参数非参数或参数非线性的方程系统感兴趣。其次，该项目开发了非参数限制，以表征经济模型中可观察变量的行为。这些结果将有助于利用可观察到的行为来证伪特定的理论。最后，该项目开发了使用加权平均方法和形状限制方法的方程组中非参数函数的估计器。

9410182 Matzkin Most existing tests of economic models proceed by first, specifying the functions in the model up to some parameters. Next, the parameters are estimated, and a test for the restrictions that the model implies on the values of the estimated parameters is performed. This procedure suffers typically from two drawbacks. First, if the restrictions are rejected, it is not clear whether this is due to a wrong model or a wrong parametric specification. And second, if the restrictions are not rejected, it is not clear whether this implies any strong evidence in favor of the model, since the restrictions were only necessary and not sufficient. This project addresses these problems by developing new methods for the identification, estimation, and testing of nonparametric models. First, the project will determine how several types of restrictions, such as inequality restrictions and optimizing behavior, can be used to guarantee the identification of functions in systems of nonparameteric simultaneous equations. The systems that will be considered include those with unobservable endogenous variables, discrete variables, and nondifferentiable functions. The results of this research will be useful to econometricians interested in estimating systems of equations that are either nonparametric or nonlinear in parameters. Second, the project develops nonparametric restrictions characterizing the behavior of observable variables in economic models. The results will be useful to falsify specific theories using observable behavior. Finally, the project develops estimators for nonparametric functions in systems of equations using weighted average methods and shape- restricted methods.