Machine Learning in economics

The course provides an introduction to modern machine learning techniques, with a

focus on applications in economics. The course can be broken down in three parts. First,
we introduce some of the most popular machine learning techniques and discuss their use
and advantage in prediction exercises (Part 1). Then, we discuss two recent applications
of machine learning in (micro-)econom(etr)ics, where the goal lies with estimating causal
effects. The first application concerns estimation of average treatment effects in the pres-
ence of many control variables (Part 2). The second application concerns estimation of
heterogeneity in average treatment effects (Part 3).
The statistical/econometric theory of each part is covered in a lecture and a corre-
sponding lab session (in the subsequent week) covers the practical side of it, going through
implementation details in R. The evaluation is based on homework assignments and a final
project.

Instructors: Hannah Bull [lab sessions—office hours: W after 16h by appointment only]
& Philipp Ketz [lectures—office hours: W 17h-18h and by appointment]

Assignments

The website for the lab sessions and assigments is here: https://ml-in-econ.appspot.
com/. There will be three assignments, one for each part. The assignments will be “prac-
tical”, i.e., they will require coding and consist in applying the methods you have learned
to new datasets. You will receive one assignment at each of the three lab sessions and you
will have two weeks to work on them, i.e., they will be due by midnight on 15/2, 1/3, and
15/3. Assignments are individual. Submission is via a link provided on the website.

Final project and grading

The final project consists of a short replication study of a paper of your choice in
groups of up to three. In addition to replicating the results in the original paper, you will
supplement the results by applying one of the (machine learning) techniques that you have
learned in this course—the paper will have to be chosen accordingly. The project should
result in a write-up of ∼3 pages (excluding graphs and references). Submission is via a
link on the website and the due date is 12/4 by midnight. You will be required to submit
both your R code and the final report.
The final grade is a weighted average of the final project (40%) and the three home-
works (60%; equally weighted).

1
 Detailed course outline (including references)

Digital copies of all references mentioned below are available on the course website.

Part 1 covers the following topics—the chapter references refer to James, Witten,
Hastie, and Tibshirani (2013):

• “Statistical learning” (Chapter 2.1)

• Model fit (Chapter 2.2.1)

• Bias–Variance Trade–Off (Chapter 2.2.2)

• Cross–Validation (Chapter 5.1.1-4)

• Regularized regression (Chapter 6.2)

• Tree-based methods (Chapter 8)

An additional reference that “explains” machine learning to economists is Mullainathan

and Spiess (2017). (See also Varian (2014), which talks a bit more about “Big data.”) In
addition to some of the above topics, Mullainathan and Spiess (2017) also touch on the
issue addressed in Part 2. See Hastie, Tibshirani, and Friedman (2009)—the “bible” on
machine learning—for additional information on the above methods, including mathemat-
ical details.

Part 2 considers the problem of estimating average treatment effects in the presence of
many regressors. The first (and technical) reference is Belloni, Chernozhukov, and Hansen
(2014b). An easily accessible explanation of their contribution is given in Belloni, Cher-
nozhukov, and Hansen (2014a) (which also describes a closely related key contribution to
the literature (Belloni, Chen, Chernozhukov, and Hansen, 2012)). We will focus on the
double selection estimator of Belloni, Chernozhukov, and Hansen (2014b). Furthermore,
we discuss the closely related double/debiased machine learning (estimator) proposed in
Chernozhukov, Chetverikov, Demirer, Duflo, Hansen, Newey, and Robins (2018).

Part 3 considers the problem of estimating/discovering treatment effect heterogeneity.

Technically, this amounts to (nonparametrically) estimating the conditional average treat-
ment effect (CATE). We consider “causal trees” (Athey and Imbens, 2016) and “causal
forests” (Wager and Athey, 2018). A simple exposition of their methods is given in Davis
and Heller (2017). An alternative approach that we will also consider is proposed in Cher-
nozhukov, Demirer, Duflo, and Fernandez-Val (2018).

2
References
Athey, S. and G. Imbens (2016). Recursive partitioning for heterogeneous causal effects.
Proceedings of the National Academy of Sciences 113 (27), 7353–7360.

Belloni, A., D. Chen, V. Chernozhukov, and C. Hansen (2012). Sparse models and methods
for optimal instruments with an application to eminent domain. Econometrica 80 (6),
2369–2429.

Belloni, A., V. Chernozhukov, and C. Hansen (2014a). High-dimensional methods and

inference on structural and treatment effects. Journal of Economic Perspectives 28 (2),
29–50.

Belloni, A., V. Chernozhukov, and C. Hansen (2014b). Inference on treatment effects after
selection among high-dimensional controls. The Review of Economic Studies 81 (2),
608–650.

Chernozhukov, V., D. Chetverikov, M. Demirer, E. Duflo, C. Hansen, W. Newey, and

J. Robins (2018). Double/debiased machine learning for treatment and structural pa-
rameters. The Econometrics Journal 21 (1), C1–C68.

Chernozhukov, V., M. Demirer, E. Duflo, and I. Fernandez-Val (2018). Generic machine

learning inference on heterogenous treatment effects in randomized experiments. Work-
ing paper.

Davis, J. and S. B. Heller (2017). Using Causal Forests to Predict Treatment Heterogeneity:
An Application to Summer Jobs. American Economic Review 107 (5), 546–50.

Hastie, T., R. Tibshirani, and J. Friedman (2009). The elements of statistical learning.
Springer.

James, G., D. Witten, T. Hastie, and R. Tibshirani (2013). An introduction to statistical

learning. Springer.

Mullainathan, S. and J. Spiess (2017). Machine learning: an applied econometric approach.

Journal of Economic Perspectives 31 (2), 87–106.

Varian, H. R. (2014). Big data: New tricks for econometrics. Journal of Economic Per-
spectives 28 (2), 3–28.

Wager, S. and S. Athey (2018). Estimation and Inference of Heterogeneous Treatment

Effects using Random Forests. Journal of the American Statistical Association 113 (523),
1228–1242.