Stochastic Simulations for Inference in Nonlinear Errors-in-Variables Models

本项目将研究同时非线性变量误差模型中的推理--识别、估计、预测和规格检验。研究了测量误差模型中可辨识性与非线性之间的联系，并分析了适当的充分可辨识性条件。将研究测量误差对非线性模型中检验假设、估计和验证的标准程序的影响。建立了非线性变量误差模型估计的参数和半参数程序。改进的蒙特卡罗和基于残差的随机模拟被用来处理测量误差问题，不仅适用于预测，而且适用于估计和规格测试。这些过程的大样本渐近理论将得到发展。特别注意测量误差产生过程的处理。测量误差的标准高斯白鼻子模型是分析的起点。然后继续分析更少限制和更合适的测量误差模型，这些模型是非线性和非高斯的。具有有限因变量的同步系统被用来更具体地检验新技术的适用性。

This project will study inference in simultaneous nonlinear error-in-variables models - identification, estimation, prediction and specification testing. The linkages between identifiability and nonlinearities in the measurement error model will be studied and appropriate sufficient identifiability conditions analyzed. The consequences of measurement errors on standard procedures for testing hypotheses, estimation and validation in nonlinear models will be studied. Parametric and semiparametric procedures are established for the estimation of nonlinear error-in-variables models. Modifications of Monte Carlo and residual-based stochastic simulations are developed to handle measurement error problems and adapted to not only prediction but also estimation and specification testing. The large-sample asymptotic theory for these procedures will be developed. Particular attention is given to the treatment of the measurement error generating process. The standard Gaussian white nose model for measurement errors is the starting point for the analysis. Analysis then continues to less restrictive and more appropriate measurement error models which are nonlinear and non-Gaussian. Simultaneous systems with limited dependent variables are used for a more specific examination of the applicability of the new techniques.