Economics 570: Big Data Econometrics

Fall 2022

Instructor: Tim Armstrong

Office: Kaprielian (KAP) Hall 310A
Email: timothy.armstrong@usc.edu
Office Hours: Wed 10am-11:50am at KAP 310A or via zoom

TA: TBA
Office Hours: TBA

Course Time and Location: Tuesday, 4pm-7:20pm,

Seeley G. Mudd Building (SGM) 101 and online
Course Webpage: Blackboard

Course Description

This course provides an introduction to machine learning and related methods for big data
from the perspective of economics. Students will be introduced to modern estimation meth-
ods for high-dimensional data, which will be illustrated through applications to causal infer-
ence and prediction problems in economics, business, and related fields. Students will gain
experience working with these methods through programming assignments. The course will
be focused on methodology and its practical application and will culminate in an empirical
project in which students apply course concepts to real-world data. By the end of the course,
students should be able to do the following.

• Apply machine learning methods to estimate causal effects in experimental and obser-
vational settings and solve prediction problems.

• Understand the uses and limitations of machine learning for answering economic ques-
tions.

• Implement machine learning algorithms using R.

Optional Textbooks

Much of the lecture material draws from the first textbook below.

BDS Taddy, M. (2019) Business Data Science: Combining Machine Learning and Economics
to Optimize, Automate, and Accelerate Business Decisions. McGraw Hill.

1
 ISL Gareth, J.; Witten, D.; Hastie, T.; and Tibshirani, R. (2017) An Introduction to Sta-
tistical Learning. Springer.

MM Angrist, J. and Pischke, J. (2014) Mastering ’Metrics: The Path from Cause to Effect.
Princeton University Press.

Prerequisites

Econometrics at the level of ECON 513 is required, along with calculus at the level of MATH
226 and linear algebra at the level of MATH 225.

Software

This course will use R, a free statistical programming language. While you do not need
to have prior knowledge of R to take this course, you will be expected to learn R on your
own time not only through lectures and homework, but also by searching independently for
relevant commands, tutorials and documentation using Google, StackOverflow, the R help()
command, etc. Programming experience in another language should suffice, so long as you
are willing to learn R on your own time.

Evaluation

Grades will be based on problem assignments (60%) and a final paper (40%).

Homework

Regular programming exercises will be assigned to reinforce the concepts taught in class,
as well as offer an opportunity for students to code and implement the algorithms covered.
Students will work in groups of 4-5, which will be assigned at the beginning of the course,
and then reassigned about halfway through the course. While each student is expected to
contribute, only one problem set will be turned in per group. R must be used as the language
for programming exercises. Problem sets will be turned in through Blackboard, along with
code and any graphs generated. Assessment will be based on whether the right approaches
were used and whether the right solutions were obtained.

Final Project

2
Students are expected to work in groups of 4-5 students each and apply course concepts
to a real-world data set to tackle an empirical problem that interests them. Students are
expected to collect their own data as part of the project. Suggestions for starting points for
publicly available data will be posted on Blackboard. Each group will submit a writeup, as
well as code to reproduce the analysis. Groups will give short presentations of their work
in class during the last few lectures. Assessment will be based on how appropriately the
quantitative tools were applied. Due date for this project is TBA.

Homework Policies

Problem sets will be discussed in class a week or more after the original due date. The
absolute deadline for turning in a problem set is when solutions are made available: home-
work turned in after this second deadline will receive zero points. 15% will be taken off for
problem sets that are turned in late, but before the problem set is discussed in class.

Course Outline

1. Introduction (0.5 weeks; BDS Intro; ISL Ch. 2).

(a) Prediction vs. causal inference.

(b) Intro to R.

2. Sampling (1 week; BDS Ch. 1; ISL Ch. 5.2).

(a) CLT and standard errors.

(b) Bootstrap.

3. Regression (1.5 weeks; BDS Ch. 2; ISL Ch. 3).

(a) Linear conditional mean model.

(b) Logistic regression.

4. Fundamental concepts in ML (2.5 weeks; BDS Ch. 3; ISL Ch. 2, 5.1, 6).

(a) Cross-validation.
(b) Regularization.
(c) Bias-variance trade-off.

5. Classification (1.5 weeks; BDS Ch. 4; ISL Ch. 4.3).

(a) Logistic lasso.

(b) Multinomial logit.

3
 6. Causal inference (2.5 weeks; BDS Ch. 5, 6; MM Ch. 1, 2).

(a) Randomized control trials.

(b) High-dimensional controls.

7. Woodlands (1 week; BDS Ch. 9; ISL Ch. 8).

(a) Classification and regression trees.

(b) Random forests.

8. Unsupervised learning (2 weeks; BDS Ch. 7; ISL Ch. 10).

(a) Principal components analysis.

(b) Partial least squares.

9. Natural language processing (1.5 weeks; BDS Ch. 8).

(a) Tokenization and bag-of-words.

(b) Topic models.

10. Deep learning (1 week; BDS Ch. 10).

(a) Architecture of neural networks.

(b) Stochastic gradient descent.