Descriptif
This course provides a working introduction to Causal Machine Learning methods. These
algorithms arise from the merge of classical causal inference techniques in econometrics
with machine learning (ML) techniques in statistics. The resulting inference algorithms
depart from classical ML - which is designed with a predictive aim - and provide an
answer to policymakers' questions (like estimation of the impact of pricing, advertisment,
employment programs, taxes, etc...). Economists, policymakers and social scientists want
to learn the causal relation between economic variables instead of the correlation between
them and want to estimate the counterfactuals. By using the formal theory of causality,
like the potential outcome framework of Rubin, and structural modeling with endogeneity
we will explain state-of-the-art Causal Machine Learning techniques.
Objectifs pédagogiques
- Understanding of the challenge related to the economic question and the econometric problem
- Design of a Machine Learning/AI-based procedure specific to the problem under consideration
- Understanding of the difference between methods for causal analysis and policy evaluation versus prediction.
- Implementation in R of the methodologies analyzed and application to real datasets.
effectifs minimal / maximal:
/25Diplôme(s) concerné(s)
- Programmes d'échange internationaux
- Titre d’Ingénieur diplômé de l’École polytechnique
- MScT-Data and Economics for Public Policy (DEPP)
Parcours de rattachement
Pour les étudiants du diplôme Programmes d'échange internationaux
Connaissance de base d'économètrie et statistique
Pour les étudiants du diplôme Titre d’Ingénieur diplômé de l’École polytechnique
Connaissance de base d'économètrie et statistique
Pour les étudiants du diplôme MScT-Data and Economics for Public Policy (DEPP)
Pour les étudiants du diplôme Programmes d'échange internationaux
Pour les étudiants du diplôme Titre d’Ingénieur diplômé de l’École polytechnique
Programme détaillé
Topics that will be treated in the course (with some examples of applications) :
1. Review of basic concepts :
(a) statistical Learning,
(b) predictive inference via penalized regression,
(c) predictive inference via modern Machine Learning : Random trees, random forests, Neural Nets/ Deep Learning,
(d) endogeneity and instrumental variables in econometrics.
2. Inference based on Double Lasso methods, Partialling-out and Neyman Orthogonality, Debiased Lasso. (Application to Growth data)
3. Causal Inference via Structural modeling, Double Machine Learning (DML) and Generic DML. (Application on Wage Gap Analysis)
4. Causal Inference via Randomized experiments : Average Treatment Effect (ATE), selection bias, heterogeneity.
5. Causal Inference via Unconfoundedness, conditional ATE. (Application on Government Spending on Welfare).
6. Causal Trees and Causal Forests for nonparametric inference for Conditional ATE.
7. Generalized Random Forests.
8. Advanced Topics (if time) : neural network-based Natural Language Processing algorithms, Double Machine Learning in the Difference-in-Difference framework.
