Lunch-Seminar CIREQ-McGill 2019-2020
joint with the Department of Economics, McGill University
Leacock 429 (McGill University, 855 Sherbrooke West)
RÉSUMÉ
Data with a large number of variables relative to the sample size are increasingly common in empirical economics. We concentrate on causal parameter estimates in settings where we want to control for many confounders and/or in the presence of many instruments. A highly popularized Machine Learning technique by Belloni, Chernozhukov and co-authors is the use of Lasso (Least Absolute Shrinkage and Selection Operator). The use of dimension reduction and principal components to regularize (Carrasco and co-authors and Galbraith and Zinde-Walsh (2019)) can provide advantages in many settings (e.g. dependence). We aim to evaluate these approaches.
