Scribd
Upload
11 views

Advanced Topics in Data Science

Uploaded by

hoang
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
11 views

Advanced Topics in Data Science

Uploaded by

hoang
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Advanced Topics in Data Science

2022-2023 Academic Year


Master of Research in Economics, Finance and Management

1. Description of the subject

● Advanced Topics in Data Science Code: 32282


● Total credits: 6 ECTS Workload: 150 hours
Term: 1st
● Type of subject: Elective
● Department of Economics and Business
● Teaching team: David Rossell
Advanced Topics in Data Science

2. Teaching guide

● Introduction
Statistical and Machine learning techniques are having a deep effect on many
disciplines, including Economics. Due to the increasing number of applications using
data science with techniques popular in Economics many World-leading institutions
incorporated data science in their PhD programs. As examples see the set of lectures
in the NBER Summer Institute by Guido Imbens and Susan Athey (in particular lecture
3, www.nber.org/econometrics_minicourse_2015/), the Big Data and machine
learning syllabus at the Harvard Economics PhD program
(https://locator.tlt.harvard.edu/course/colgsas-156429), the Chicago Booth PhD
courses on Bayesian inference, Big Data and Machine Learning, or the Bocconi PhD
courses in text analysis, Econometrics of networks and causal analysis.

Data analysis methods are evolving to cope with increasingly challenging problems,
a case in point being high-dimensional situations where one considers a large
number of parameters or models. For instance, one may have a regression or factor
model where the number of covariates far exceeds the sample size, or may want to
simplify the interpretation of complex data via clustering, text or latent variable
analysis, or may want to predict an outcome using flexible algorithms. Engaging
effectively in such research, either from a methodological or applied perspective,
requires one to understand, and when needed modify or extend, such methodology.
Just as importantly, they need to communicate such ideas effectively to a potentially
non-expert audience.

The goal of this course is to introduce students to some foundations behind these
methods, with a certain emphasis on the Bayesian framework, penalized likelihood
and latent variable methods (e.g. as used in text analysis) methods, expose them to
and discuss research literature, and practice the skills needed for applying and
presenting novel research. The learning outcomes are an improved familiarity with
selected research topics in Statistics that are relevant for Data Science, at a level
sufficient to critically appraise, modify and apply novel methods, and improved oral
and written presentation skills. The course also intends to provide students with
applied data analysis skills useful for their MRes thesis and subsequent research work.

Pre-requisites: the course is designed for MRes students who are familiar with basic
Statistical inference, specifically linear regression and maximum likelihood estimation.
Basic R programming skills are beneficial, though examples and links to learning
resources will be provided for students who are not familiar with R.

● Contents

1. Foundations. We briefly review classical results from maximum likelihood


estimation and computational methods such as the bootstrap, and we introduce
the basic Bayesian paradigm for statistical inference, its use for model selection,

2
Advanced Topics in Data Science

parameter estimation and prediction, and standard computational tools such as


Gibbs or Metropolis-Hastings.
2. Foundations of variable selection in high-dimensional regression. We review the
fundamental penalized likelihood and Bayesian frameworks for linear regression
models with a large number of variables, and for treatment effect estimation in
such settings. We shall discuss the relative merits of current strategies such as
LASSO, adaptive LASSO and related penalties to help decouple variable selection
from prediction and various Bayesian strategies to achieve good performance in
high dimensions. We will discuss theoretical and practical considerations and
computational strategies.
3. Beyond linear regression. We will extend the earlier strategies to settings where one
considers certain forms of causal inference, simple time series models,
generalized linear models, flexible models for count data, random forests and
Bayesian additive regression trees, or capturing non-linear relationships.
4. Mixture and flexible models. We shall move towards more flexible models enabled
by the use of latent variables. We will place some attention to mixture models and
some non-parametric methods. We shall discuss applications to mixture-of-
regressions models to account for unobserved confounders that may bias
inference, to hidden Markov models for time series and to text data analysis.

References. The books below provide a good introduction to a substantial part of


the topics covered in this course (and many others), however we shall complement
them with a number of additional selected research manuscripts.

● Andrew Gelman, John B. Carlin, Hals S. Stern, David B. Dunson, Aki Vehtari, Donald B.
Rubin. Bayesian Data Analysis (3rd edition). CRC Press, 2013.

● Trevor Hastie, Robert Tishirani, Martin Wainwright. Statistical learning with sparsity.
The LASSO and its generalizations. CRC press.

● Sara van de Geer, Peter Bühlman. Statistics for high-dimensional data: methods,
theory and applications. Springer, 2001.

● Nils Lid Hjort, Chris Holmes, Peter Müller, Stephen G. Walker. Bayesian non-
parametrics. Cambridge University Press, 2010.

● Sylvia Frühwirth-Schnatter. Finite mixture and Markov switching models. Springer,


2006.

Some papers on data science and economics:

● Chernozhukov, V., A. Belloni, C. Hansen (2014), "High-Dimensional Methods and

3
Advanced Topics in Data Science

Inference on Treatment and Structural Effects in Economics", J. Economic


Perspectives

● Chernozhukov, V., D. Chen, A. Belloni, C. Hansen (2012), "Sparse Models and Methods
for Instrumental Regression, with an Application to Eminent Domain",
Econometrica

● Einav, L. y J. Levin (2014), “Economics in the age of big data,” Science, 346 (6210).

● Varian, H. (2014), “Big data: new tricks for econometrics,” Journal of Economic
Perspectives, 28 (2), 3-28.

● Teaching methodology

The course will be delivered in a combination of regular lectures, computer-based


seminars where students get hands-on experience with the taught data analysis
methods, and presentations by students (on published manuscripts chosen by the
students and on the final project).

● Assessment and Grading System

Students will be asked to orally present 2 research papers of their choice (40%
of the final mark), some selected exercises from the seminar sessions (10% of
final mark) and a written report (50% of the final mark). This project will be
decided by the students but must be pre-approved by the lecturer, and should
involve the application, critical assessment or extension of the research
methods seen in class. The content can be theoretical, empirical, a practical
application or a combination of the former.

You might also like