OCE - Semester Course

Course material for ECON2148 : Optimization-Conscious Econometrics

Autumn 2020, Harvard University

Tuesday-Thursday, 3:00-4:15 am. NEW TIME!

This course studies the core optimization concepts underlying econometric estimation and inference. The objective is to both develop a deep understanding of how estimators are computed, and to get a better theoretical and geometrical understanding of classical econometric estimators through the prism of optimization theory.

Each optimization concept or method is studied using a well established econometric estimator as the working example: linear programming is taught through the example of quantile regression, duality is taught via nonparametric inference, numerical linear algebra is taught via partial identification questions in OLS, integer programming is taught as a solution method for instrumental variables quantile regression, and so on.

Syllabus

Lecture 0: Markov Chain Monte Carlo Methods

Lecture 1: Quantile Regression and Linear Programming

Lecture 2: Nonparametric Robust Inference and Duality -- updated Feb 19, 2021

Lecture 3: Optimal Transport and Network Flow Problems -- updated Feb 19, 2021

Lecture 4: Regularized Optimal Transport, Sinkhorn's Theorem, and Iterative Projections -- updated Fab 19, 2021

Lecture 5: Integer Programming and Instrumental Variables Quantile Regression

Lecture 6: Numerical Algebra and Partial Identification in Ordinary Least-Squares with Singular Design Matrix -- preliminary

Lecture 7: Lagrange Multiplier Theory and the Degrees of Freedom of Constrained Estimators

Lecture 8: Numerical Integration and Uniform Inference

Problem Set 1

Problem Set 2