Collaborative Research: Applications of CCP estimation to dynamic discrete choice models with unobserved heterogenity

许多重要的决定都涉及到权衡现在和未来。例如，留在学校的决定需要在今天产生成本，以在未来获得(通常是可观的)经济利益。同样，开始吸烟的决定在今天可能会带来(人们认为的)社会效益，但未来会付出巨大的健康代价。理解这些动态决策是如何做出的，对于制定有效的社会政策至关重要。然而，对这些决策过程进行正式建模可能非常复杂，这给严格的分析制造了障碍，并限制了前瞻性政策分析的范围。这个研究项目的目标是通过开发一种更简单但同样严格的分析动态决策过程的经验方法来减少这些障碍。本文开发的框架建立在现有的经验方法-条件选择概率(CCP)估计的基础上，该方法为分析动态离散选择问题提供了一种计算上容易处理的方法。CCP方法尚未广泛应用于实践，主要是因为人们认为它们具有不必要的限制性，要求研究人员观察代理人自己看到的世界的一切。这一研究项目证明了这种看法是不正确的，概括了可以使用CCP方法估计的模型类别，并提供了纳入未观察到的异质性的具体方法。此外，由于估计器的计算简单，这些未观察到的变量可以持续存在，而不是永久性的。例如，某些市场可能对特定类型的工人有很高的需求，但需求较高的市场可能会随着时间的推移而变化。由于问题的计算复杂性，过去没有对这些类型的问题进行估计。该项目通过分析工会如何影响超市行业的进入、退出和投资决策来说明解决方法的优势。该项目特别关注工会如何通过超市做出的动态决策影响产品市场竞争。该提案的更广泛影响是大大增加了可以从动态角度分析的问题的类别。此外，通过大幅减少从事这类研究所需的技术专长，该项目将向更多类别的研究人员和学科开放这一领域。例如，通过加入随时间变化的未观察变量，该算法特别适合于具有学习的动态游戏和模型中的应用。

Many important decisions involve balancing present against future. For example, the decision to stay in school requires incurring costs today to reap (often substantial) financial benefits in the future. Similarly, the decision to start smoking may have (perceived) social benefits today, but substantial health costs down the road. Understanding how such dynamic decisions are made is essential in formulating effective social policy. However, formally modeling these decision processes can be extremely complicated, creating a barrier to rigorous analysis and limiting the scope of forward looking policy analysis. The goal of this research project is to decrease these barriers by developing a much simpler, but equally rigorous empirical technique for analyzing dynamic decision processes.The framework developed here builds on an existing empirical methodology, Conditional Choice Probability (CCP) estimation, that provides a computationally tractable method for analyzing dynamic discrete choice problems. CCP methods have not yet been widely used in practice, mainly due to the perception that they are unnecessarily restrictive, requiring the researcher to observe everything about the world that the agents themselves see. This research project demonstrates that this perception is incorrect, generalizing the class of models that can be estimated using CCP methods and providing specific methods for incorporating unobserved heterogeneity. Furthermore, because of the computational simplicity of the estimator, these unobserved variables can persist without being permanent. For example, certain markets may have high demand for particular types of workers but the markets with high demand may change over time. These types of problems have not been estimated in the past because of the computational complexity of the problem. The project illustrates the advantages of the solution method by analyzing how unionization affects the entry, exit, and investment decisions in the supermarket industry. The project pays particular attention to how unionization affects product market competition through the dynamic decisions made by supermarkets.The broader impact of the proposal is to greatly increase the class of problems that can be analyzed from a dynamic perspective. Further, by significantly reducing the technical expertise necessary to engage in research of this type, this project will open up the field to a broader class of researchers and disciplines. For example, by incorporating unobserved variables that transition over time, the algorithm is particularly well suited to applications in dynamic games and models with learning.