Estimation Methods for LDV Models with Unrestricted Correlations in the Unobservables to Analyze Three Classes of Economic Problems

从经济理论推导出的有限因变量（LDV）模型，通常不会对不可观测物之间的相关性施加理论上合理的限制。然而，目前对具有自由相关不可观测量的LDV模型进行经典估计通常是不可行的，因为它需要计算多维积分。特别令人感兴趣的是在面板LDV模型和具有测量误差的LDV模型中出现的此类问题。本项目开发并实现了数值积分与实现的仿真方法，以促进以前难以解决的计算问题的解决。这是通过三个方向的发展来实现的：首先，将Hartley（1958）的EM算法与仿真方法相结合，以克服所研究估计器中的不连续问题。其次，提供了直接从条件密度进行模拟的技术。这些技术对于模拟矩法的实现非常重要（McFadden, 1988）。第三，设计了在未知参数向量上连续的仿真估计器。这些计量经济学方法被用来分析三类经济问题。首先，使用面板动态LDV模型研究了最不发达国家外债危机发生率和持续时间以及IMF条件限制计划的决定因素。通过估计初值的条件分布，为长期存在的“初值”问题提供了新的解决方案。其次，运用集体谈判模型对罢工的发生进行研究。这些问题是通过不完全测量的定性和连续变量的LDV模型进行计量经济学分析的。最后，通过计量经济学相似的方法对隐性串通行为的博弈论模型进行了检验。这些实用的工作模型将说明所采用的新方法的实用性和可操作性。

Limited dependent variables (LDV) models derived from economic theory frequently do not impose theoretically justifiable restrictions on correlations among the unobservables. At present, however, classical estimation of LDV models with freely correlated unobservables is generally infeasible because it requires the evaluation of multi-dimensional integrals. Of particular interest are such problems that arise in panel LDV models and LDV models with measurement errors. This project develops and implements simulation methods for numerical integration and implementation to facilitate solution of the previously intractable computational problems. This is achieved by developments along three directions: First, the EM algorithm of Hartley (1958) is combined with simulation methods to overcome discontinuity problems in estimators studied. Second, techniques are provided to simulate directly from conditional densities. These techniques are important for the implementation of the Method of Simulated Moments (McFadden, 1988). Third, new simulation estimators are devised that are continuous in the unknown parameter vector. These econometric methods are employed to analyze three classes of economic problems. First, determinants of incidence and duration of external debt crises and IMF conditionality programs in LDC's are studied using panel dynamic LDV models. New solutions to the long-standing problem of "initial values" are provided by estimating the conditional distribution of the initial values. Second, collective bargaining models are used to study the occurrence of strikes. These issues are analyzed econometrically by LDV models with imperfectly measured qualitative and continuous variables. Finally, game-theoretic models of tacitly collusive behavior are tested through econometrically similar methods. These working practical models will illustrate the utility and tractability of the new methods employed.