动态离散选择模型的贝叶斯估计与变量选择

Estimation of dynamic discrete choice models is computationally demanding and is subject to the curse of dimensionality. In this project, we propose a new Bayesian Markov chain Monte Carlo method that can simultaneously estimate and conduct variable selection on dynamic discrete choice models with a large number of state variables. The method makes it possible for researchers to use dynamic discrete choice models to model high dimensional data where the researchers are not sure which of the many available variables are of relevance to the decision making of the agents in their models. Although the project focuses on dynamic discrete choice models, we also show that the algorithm can be easily extended to the variable selection and estimation of dynamic games of imperfect competition. We demonstrate the performance of our algorithms by estimating a dynamic model of individual migration and a dynamic game of shopping center competition. The proposed method has wide applications in industrial organization, labor economics, finance, as well as other fields of economics.

本课题提出一个新的贝叶斯马科夫蒙特卡洛方法用以同时对动态离散选择模型进行估计和变量选择。该方法可以让研究者在不确定哪些变量与个体选择相关的情况下，在初选阶段保留所有可能相关的变量，让估计方法在估计模型参数的同时对模型进行降维。通过该方法，用动态离散选择模型模拟高维数据成为一种现实可能。同时，该方法不仅可以用于动态离散模型，而且在一定拓展之后，可用于动态不完全竞争模型。本课题用两个实验展示该方法的有效性：一个为个体动态迁徙模型，另一个为购物中心动态竞争模型。在科研意义上，本课题首次将变量选择和高维数据分析方法引入了对动态结构模型估计的研究。课题提出的方法在产业组织，劳动经济，金融等各个领域都有广阔的应用前景，并且在一定意义上开启了用动态结构模型处理大数据的研究方向。