AØKK08386U Seminar: Econometrics, Machine Learning and Big Data.

Volume 2019/2020
Education

MSc programme in Economics

The seminar is primarily for students at the MSc of Economics

Content

The development of machine learning methods and the use of big data have been exploding over the past decade, and sometimes presented as an alternative to more traditional approaches in econometrics that have been (and still are) widely used in the social sciences. Historically, the two fields have been competing. However, the recent literature has emphasized that they are complementary and have strong synergies. Machine learning can leverage econometrics to design new methods with improved flexibility to address new complex problems. Moreover, the collection of Big Data creates new challenges for traditional econometric methods in terms of computation and identification. Machine learning may allow researchers to tackle some of these challenges.

 

The goal of this seminar is to bridge the gaps between the two approaches, and to explore new avenues of research that can crossbreed econometric and machine learning methods in a productive way. This idea has recently been flourishing and has produced a number of relevant insights for economists (Mullainathan, Spiess (2017); Athey (2018)). The set of tools stemming from machine learning can broadly be categorized according to whether they enhance current tools for policy evaluation and causal effects (e.g., use of instrument variables, matching, etc.), or whether they are general procedures estimating models in high-dimensions.

 

For policy evaluation, for example, the new tools make the following improvement: i) automatic selection of variables for use in regression models (Belloni et al. (2014)); have data driven choices of heterogeneous treatment effects (Athey, Wager (2018)).

 

The econometric methods that are relevant for this seminar includes the classical, frequentist approach or Bayesian econometrics. The student will have the choice between the two approaches (or possibly decide to use both).

 

Participants of the seminar will get a chance to work on a number of different projects that are either theoretical or applied, or both. Examples of seminar projects include the replication of old studies that do not make use of these new methods, to check the robustness of their findings against more flexible approaches; collecting new data and applying the methods on these data; extending the new methods in relevant ways.

Learning Outcome

In addition to the learning outcome specified in the Curriculum the student is after completing the seminar expected to be able to:

 

Knowledge:

  • Have reviewed the relevant literature related to the topic chosen, and understand the state of the art as well as the limitations of the current approaches.
  • Have a grasp of the econometric and machine learning methods relevant for seminar topics.

 

Skills:

  • Combine econometric methods with machine learning methods appropriately, to enhance the former with the latter.  
  • Derive an algorithm theoretically, and to code it to produce a computer program.
  • Write a computer program from scratch in Python or R that implements the method. This process includes: learn  how  to  (better)  write  code,  to  debug  it,  to  test  it  (e.g., using unit  root  testing),  to document it, to share it (package creation) and to apply it to real data.
  • Collect relevant data and apply the methods.

 

Competencies:

  • Master the methods borrowed from econometrics and machine learning to create a new approach relevant for the estimation of a particular program, from the theoretical foundations of the approaches, to their implementation in practice.
  • Athey, S. (2018). The impact of machine learning on economics. In The Economics of Artificial Intelligence: An Agenda. University of Chicago Press.
  • Belloni, A., Chen, D., Chernozhukov, V., & Hansen, C. (2012). Sparse models and methods for optimal instruments with an application to eminent domain. Econometrica, 80(6), 2369-2429.
  • Mullainathan, S., & Spiess, J. (2017). Machine learning: an applied econometric approach. Journal of Economic Perspectives, 31(2), 87-106.
  • Wager, S., & Athey, S. (2018). Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association, 113(523), 1228-1242.

 

BSc in Economics or similar
It is highly recommended to have followed at least two of the following courses before the start of the seminar (ideally, at least one from econometrics and one from data science):
- Advanced Microeconometrics (AØKA08084U)
- Seminar in Advanced Microeconometrics (AØKK08106U)
- Dynamic Programming - Theory, Computation, and Empirical Applications (AØKK08207U)
- Topics in Social Data Science (AØKK08371U)
- Summer School in Social Data Science (AØKK08216U)
- Summer School in Bayesian Econometrics (AØKK08359U)
- Seminar in Bayesian Econometrics (AØKK08333U)
Similar courses followed at another institution are also valid.

This seminar will have a strong programming component. It will rely on Python and/or the R programming language. The student should have some good programming skills in one of these two languages, or be willing to learn one of them before the start of the semester (a list of online tutorials will be provided for this purpose).
Activities:
- Kick-off meeting
- Finding literatur and defining the project
- Writing process of the seminar paper
- Presentation of own project and paper
- Giving constructive feedback to another student´s paper
- Actively participating in discussions at the presentations and other meetings.

At the seminar the student is trained independently to
- identify and clarify a problem,
- seek and select relevant literatur,
- write a academic paper,
- present and discuss own paper with the other students at the seminar.

The aim of the presentations is, that the student uses the presentation as an opportunity to practice oral skills and to receive feedback. The presentations is not a part of the exam and will not be assessed.

There is no weekly teaching/lecturing and the student cannot expect guidance from the teacher. If the teacher gives a few introduction lectures or gives the opportunity for guidance, this as well as other expectations are clarified at the kickoff meeting.

Process:
It is strongly recommended that you think about and search for a topic before the semester begins, as there is only a few weeks from the kick-off meeting to the submission of the project description/ agreement paper.

Before the presentations, your nearly finished version of the seminar project paper must be uploaded in Absalon, as the opponents and the other seminar participants have to read and comment on the paper. It is important that you upload a paper that is so finalized as possible due to the fact that the value of feedback and comments at the presentation is strongly associated with the skill level of the seminar paper.

After the presentations, you can with a few corrections improve the seminar paper by including the feedback and comments emerged during the presentations. It is NOT intended that you rewrite or begin the writing of the full project AFTER the presentation has taken place.
Schedule:
Autumn 2019:
• Kick-off meeting: September 2, 2019 at 13-15 (General introduction to the seminar, discussion of possible topics. Students are required to have studied reading list and use it to come up with topics for a seminar paper.)
• Deadline commitmentpaper: September 20 (Each group must submit a mandatory one-page proposal for their seminar paper. Before that, they can arrange a meeting with one of the supervisors to discuss topics and related literature.)
• Midterm poster session to obtain/provide feedback on ongoing projects: Week 41. Exact dates is agreed on at the kick-off meeting (Each student will be assigned at least one poster to discuss.)
• Deadline of pre-paper uploaded to Absalon: one week before presentations
• Presentations/Workshops: Week 46 or 47. Exact dates is agreed on at the kick-off meeting

All information regarding the seminar is communicated through Absalon including venue. So it is very important that you by yourself logon to Absalon and read the information already when you are registered at the seminar.
  • Category
  • Hours
  • Project work
  • 186
  • Seminar
  • 20
  • Total
  • 206
Credit
7,5 ECTS
Type of assessment
Written examination
A seminar paper in English that meets the formal requirements for written papers stated in the curriculum of the Master programme and at KUNet for seminars.
____
Exam registration requirements

Attendance in all  activities at the seminar as stated in the formal requrements in the Curriculum  and at the KUnet for Seminars (UK)  and  Seminars (DK)  is required to participate in the exam. Including to attend the poster session as a presenter and as a discussant

___

Aid
All aids allowed

for the seminar paper.

The teacher defines the aids that must be used for the presentations.

___

Marking scale
7-point grading scale
Censorship form
External censorship
____
Exam period

Exam information:

Autumn 2019:

Deadline for submitting the final seminar paper: December 2, 2019 before 10 AM

 

Exam information:

The seminar paper must be uploaded to the Digital Exam. More information will be available from the middle of the semester.

 

For enrolled students more information about examination, rules, aids etc. is available at the intranet for Master (UK) and Master (DK ).

____

Re-exam

Reexam information:

The reexam is a written seminar paper as stated in the Curriculum.

Deadline and more information is available at Seminars(UK) and Seminars(DK).

More information about reexam etc is available at Master(UK)andMaster(DK).

Criteria for exam assesment

Students are assessed on the extent to which they master the learning outcome for the seminar and can make use of the knowledge, skills and competencies listed in the learning outcomes in the Curriculum of the Master programme.

 

To receive the top grade, the student must with no or only a few minor weaknesses be able to demonstrate an excellent performance displaying a high level of command of all aspects of the relevant material.