Robust Inference on Counterfactuals

点击翻译按钮获取中文摘要

Large scale and detailed econometric models are used to evaluate the effects of economic and social policies before these policies are implemented. This approach is needed because experimenting with alternative policies is too costly or not possible. In developing data-ready models for such policy evaluation, model-builders often make several restrictive assumptions without the guidance of economic theory. However, policy predictions can be very sensitive to such assumptions. Researchers and policy makers need methods to help understand to what extent policy predictions are driven by economically grounded assumptions or on restrictive assumptions. The proposed research project will develop econometric tools that allow researchers to judge whether policy predictions are based on economically grounded theory or on restrictive assumptions. Researchers studying education, labor, public health, antitrust, trade, development, and environmental policies, among others, will use the tools developed in this project. This research project will therefore aid in formulating policies to improve social outcomes including welfare, the efficient provision of goods and services, and economic growth and competitiveness. The research will also contribute at a technical level to STEM education including operations research and statistics. The goal of the proposed research is to develop robust, computationally tractable estimation and inference methods for counterfactuals in structural econometric models. The methods will be robust in the sense that they will characterize the set of counterfactuals predicted by the model when some distributional or functional form assumptions are relaxed but other features of the model more strongly grounded in economic theory are maintained. To provide computational tractability, the research will build on recent advances in convex programming methods developed in the field of distributionally robust optimization (DRO) in operations research. The research will advance techniques and inference procedures developed in DRO to accommodate important distinguishing features of structural econometric models, including endogeneity, fixed-point constraints, and unobserved heterogeneity. The inference procedures developed during this research will also contribute broadly to the literature on partially identified semi-parametric econometric models. The methods will be specialized to leading models of demand for differentiated products, workhorse dynamic models of single-agent behavior, and static and dynamic models of strategic interactions between multiple agents.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.