Optimization-Conscious Econometrics Summer School

About

The aim of the Summer school is to equip graduate students with tools to carry forefront research at the intersection of optimization and econometrics.

The OCE Summer school will be held on the University of Chicago campus June 4-7, 2023, and students are expected to likewise attend the OCE Conference on June 9 and 10, 2023.

Students may apply for funding to support accommodations and travels. Please contact event organizer Guillaume Pouliot with any questions at guillaumepouliot@uchicago.edu.

Application Requirements

The Summer School is designed for graduate students focused in econometrics who want to strengthen their background in optimization. Qualified applicants should have completed their second year field courses in econometrics.

Students would likewise attend the BFI OCE II Conference, on June 9-10.

Registration

Fill out this form to apply for funding to attend the OCE Summer School June 4-8. Students are expected to cover the costs for the additional accommodation associated with staying for the  BFI OCE Conference II.

As part of the application form, student applicants will be required to send their transcript and a short statement of purpose explaining how they expect the school to help in their future research.

Application Form

Speakers

Guillaume A. Pouliot

Guillaume A. Pouliot

Assistant Professor at Chicago Harris

Guillaume Pouliot is an Assistant Professor at Chicago Harris. His research focuses on developing statistical methods for nonstandard problems in public policy and economics, the extension of machine learning methods for applications in public policy, and problems at the interface of econometrics and optimization.

Pouliot received his PhD from Harvard University. Previously, he received his B.A. (Honors) in economics as well as his M.S. (concurrent) in statistics from the University of Chicago.

Pierre E. Jacob

Pierre E. Jacob

Professor of Statistics, ESSEC Business School

Pierre E. Jacob is a professor of statistics at ESSEC Business School, in Paris, France.

His research pertains to statistical inference and time series analysis. Jacobs develops Monte Carlo methods to compare models or to estimate latent variables. At ESSEC he teaches courses on forecasting and statistics in general.

Jacob received his PhD from Université Paris Dauphine.  Previous to joining ESSEC, Jacob was Associate Professor at the statistics department at Harvard University.

Alfred Galichon

Alfred Galichon

Director, NYU Paris; Professor of Economics, NYU Arts & Science; Professor of Mathematics, NYU Courant Institute; Affiliated Professor of Data Science, NYU Center for Data Science

Alfred Galichon is a professor of economics and of mathematics at New York University, an affiliated faculty of NYU’s Center for Data Science, and the director of NYU Paris. He also serves as the principal investigator of the ERC-funded EQUIPRICE project at Sciences Po, Paris.

His research interests span widely across theoretical, computational and empirical economic questions and include econometrics, microeconomic theory, and data science. He is one of the pioneers of the use of optimal transport theory in econometrics, and the author of a monograph on the topic, Optimal Transport Methods in Economics (Princeton, 2016).

Galichon holds a Ph.D. in economics from Harvard University (2007), and an engineering degree from Ecole Polytechnique (X97) and one from Ecole des Mines de Paris (Corps des Mines, 2002). Among his numerous awards, he is an elected Fellow of the Society for the Advancement of Economic Theory, a Fellow of the Econometric Society, a “Young leader” of the French-American foundation, and a recipient of the Edmond Malinvaud prize.

Jean-Bernard Lasserre

Jean-Bernard Lasserre

Senior Scientist at the Centre National de la Recherche Scientifique (CNRS) at LAAS in Toulouse, France; Member of the Institute of Mathematics at the University of Toulouse

Jean B. Lasserre is a Senior Scientist at the Centre National de la Recherche Scientifique (CNRS), at LAAS in Toulouse, France, and is also member of the Institute of Mathematics at the University of Toulouse. He graduated from ENSIMAG (Grenoble) and earned his PhD and Doctorat d’Etat from the University of Toulouse. He spent two one-year visits in the EECS department at UC Berkeley. A SIAM Fellow, he is the recipient of the 2015 John von Neumann Theory Prize, the 2015 Khachiyan Prize, and the 2009 Lagrange Prize in Continuous Optimization, and was the 2014 Laureate of an ERC Advanced Grant from the European Research Council (ERC), and an Invited Speaker at the ICM 2018 in Rio de Janeiro. His research interests are in Applied Mathematics, Probability and Optimization.

 

Curriculum

Guillaume Pouliot

Quantile Regression
Modern Introduction to Quantile Regression
We parallel the best linear predictor development of OLS à la Chamberlain, but for quantile regression. We consider interpretation, standard inference, and their pitfalls.
Quantile Regression Through the Lens of Linear Programming
We consider quantile regression (QR) as a linear program (LP). We use QR as an example to introduce key concepts in LP duality. We derive the key QR identity as a direct consequence of LP duality.
Regression Rankscore Inference
We consider the default approach to large sample inference for quantile regression, the nature of its pivotality, and its connection –or lack thereof– with permutation tests.
Instrumental Variable Quantile Regression
We consider computation, inference, and weak IV robust subvector/simultaneously valid inference for the inverse quantile regression estimator, as well as an introduction to applications of mixed integer linear programming in econometrics.

Pierre Jacob

Markov Chain Monte Carlo and Couplings
Introduction to Markov Chain Monte Carlo (MCMC)
Introduction to Markov chain Monte Carlo (MCMC) and its relevance in data analysis, for example in parameter inference, hypothesis testing and model choice. Connections are made between MCMC algorithms and iterative optimization methods.
MCMC Theory
Convergence to stationarity and rates, law of large numbers and central limit theorem for MCMC, with an emphasis on the usefulness of couplings and Poisson equations as proof techniques.
Implementable Couplings
How to use couplings in practice to diagnose convergence of Markov chains, to parallelize computation completely or to estimate (without any bias) the asymptotic variance in the central limit theorem for Markov chains.
Designing Couplings
How to construct algorithms that generate pairs of Markov chains that meet exactly after a random number of steps. This will cover the most popular MCMC algorithms as well as more recent sampling algorithms designed for distributions supported on manifolds.

Alfred Galichon

Optimal Transport
Network Flow Problems
We introduce network flow problems and their linear programming treatment as well as optimization algorithms specialized to this problem.
The Optimal Assignment Model
We cover the optimal assignment model and its entropic regularization.
More on the Optimal Assignment Model
We cover the semi-discrete case as well as the one-dimensional case and its connections with notions of quantile
and ranks.
Inverse Optimal Transport
We discuss the estimation of matching models and the theory of inverse optimal transport

Jean-Bernard Lasserre

The Moment-SOS Hierarchy and the Christoffel Function for Data Analysis
Course 1: Moments and positive polynomials
In this course, we briefly describe the theory of moments and positive polynomials which provides the rationale behind the Moment-sum-of-squares hierarchy.
Course 2: LP- and SOS-based positivity certificates for optimization
We explore linear programming and sum-of-squares approaches to producing positivity certificates for optimization.
Course 3: Some applications of the Moment-SOS hierarchy
We explore the intersection of the theory of moments and positive polynomials with several areas (optimization, real algebraic geometry, functional analysis). We also explore the almost endless list of important applications (optimization, computational algebra and geometry, probability & statistics, signal processing, control, optimal control and nonlinear PDEs, quantum information, to cite a few).
Course 4: The Christoffel function and its links with the Moment-SOS hierarchy
While being an old tool from the theory of approximation and orthogonal polynomials, we claim that the Christoffel function provides a simple and easy-to-use tool for some problems in data analysis, approximation and interpolation of discontinuous functions. Moreover, we will reveal its links (some quite surprising) with optimization, sum-of-squares, certificates of positivity, equilibrium measures of compact sets, and more.