MKTG 584 B – Dynamic Structural Models

Fall 2018

Time & Place:

TBD

Instructor:

Hema Yoganarasimhan

Office: Paccar 481

Course Outline

The goal of the course is to train students to model inter-temporal trade-offs made by consumers and firms using structural models. In the process, the course will introduce students to advanced estimation methods, identification, and problem formulation.

While the focus is on single-agent dynamic models, many of the methods and some of the papers will cover static games.

The course will require considerable commitment on the part of students in terms of reading, understanding many conceptual papers and writing programs. If you are not willing to make this commitment, this course is not for you.

Class Format

Each class, we will learn one topic. You will be assigned background readings. The class will consist of a mixture of lecture and class discussion.

Attendance

Students are expected to come to all classes.

Pre-requisites

o Good understanding of static discrete choice models.

o Basic programming skills.

o Willingness to work and have fun!

Notes

I have written detailed notes for most topics covered in the class. They are available in PDF format here: Hema_Notes.

Grading

· Assignments

Do the seven assignments and submit them on/before the specified time. Late submissions will not be graded. This carries 12.5 points per assignment, with a total of 87.5 points.

· Class participation grade

This comes from two types of class participation:

• Students will be assigned to lead the discussion of a few papers through the course. They are expected to write a brief summary of this paper, share it with others, and explain the main concepts of the paper. You will be evaluated based these discussions and the understanding of all the papers/concepts discussed in the class.
• Canvas Discussion Board – I have set up Canvas discussion boards for each assignment as well as a general discussion board. You can post questions, ask for help, and provide help to your classmates through these discussion boards. The extent to which you participate on this board (helping others) will affect your participation grade.

Class participation counts for a total of 12.5 points.

Assignments

There are seven programing assignments. Links to the assignments are listed below:

1. Assignment 1: Data generation for the Rust bus engine model.

2. Assignment 2: Estimation of Rust bus engine using Nested Fixed Point.

3. Assignment 3: Estimation of Rust bus engine using the Two-step method.

4. Assignment 4: Data generation for the Rust bus engine model with persistent unobservables.

5. Assignment 5: Estimation of the Rust bus engine with persistent unobservables using EM & Nested Fixed Point.

6. Assignment 6: Estimation of the Rust bus engine with persistent unobservables using EM & Two-step method.

7. Assignment 7: Theoretical identification proofs.

Notes about the assignments

· You can use any programming language that you are comfortable with. For each assignment, I will test the correctness of the code submitted using my dataset or by asking you to generate specific results.

· The code should be simple and clean. You are expected to put detailed comments in your code for each assignment.

· The instructor will not help in de-bugging or correcting the code.

· Do not copy code from sources available online. You are welcome to use these resources to understand how to structure your code, debug it etc., but you are expected to write your code in the end.

· You are encouraged to discuss the assignments with each other. However, each person should write his or her own code. If you are not able to explain your code to me, I will assume it is not yours. In such situations, you will be automatically given zero points for the assignment.

· Your grade for each assignment will solely depend on the correctness of the code. If the code produces correct results, you get full points. If it produces incorrect results, you get zero points. However, if it happens that the code gives correct answers using incorrect logic or some other flaw in the system, you will get zero points.

· There is no partial grading. While this may seem harsh, remember that this is how the publication process works. Incorrect codes and research papers can significantly harm your career. The goal is to get you to focus on writing thoughtful, correct code, and get you into the habit of putting in checks in your code to ensure that it is correct.

FAQ

1. I am not in the marketing department. Can I still take the course?

Yes, you can. However, please check with the instructor before doing so.

2. Can I audit the course?

No, auditors are not allowed.

3. What kind of programming skill do I need for the class?

You are expected to have basic to good programming skills. The class itself will not cover any programming languages or basics. You need to be conversant with or willing to learn one good programming language (with an in-built optimizer) to be able to complete the assignments. Depending your prior knowledge, you could use Stata (or Mata), Matlab, Python/C in conjunction with Stata etc.

4. Is this a programming class or will we cover theory too?

This is not a programming class. We will cover many theoretical aspects of dynamic structural models such as how to write formal identification proofs and the convergence properties of estimators. In the end, a good structural model is a good theory model taken to data and the main goal of these models is to answer interesting substantive questions. So understanding the theory and the substantive contribution of the papers discussed is a primary objective of this class.

5. I have heard a rumor that Prof. Yoganarasimhan has a policy of only working with students who do well in this course. Is this true?

It is not a rumor; it is a fact! I have a very simple policy. If you want to work with me on a project, or write you a recommendation letter, or serve on your committee, or be your dissertation advisor, then you have to: 1) Take this course with grading system 2 and do well in class, and 2) answer my qualification exam question based on this class and get a high pass. If you are outside the Marketing department (IS/Economics), these rules still apply.

For more details, please read the “Doctoral Mentoring Policy” document: Hema_doctoral_mentoring_policies.

Class Schedule

CLASS	DATE	TOPIC	ASSIGNMENT DUE
1	September 28	Introduction to Discrete Choice Models
2	October 5 (TB rescheduled)	Introduction to Single Agent Dynamic Models & Nested Fixed Point
3	October 12	Two-step Methods – I	1
4	October 19	Two-step Methods – II	2
	October 26	No class	3
5	November 2	Persistent Unobservables & Nested Fixed Point with EM	4
6	November 9	Guest Speaker
7	November 16	Two-step methods with EM	5
8	November 30	Identification	6
9	December 7 (TB rescheduled)	MPEC Methods	7

Class Readings

Class 1: Introduction to Discrete Choice Models

· Chapters 1-5 of Kenneth Train’s textbook Discrete Choice Methods with Simulation available online.

· Stata notes posted on Canvas

· Application Papers

o Guadagni & Little, “A Logit Model of Brand Choice Calibrated on Scanner Data,” Marketing Science, (1983)

o Kamakura & Russell, "A Probabilistic Choice Model for Market Segmentation and Elasticity Structure," Journal of Marketing Research, (1989)

Class 2: Introduction to Single Agent Dynamic Models & Nested Fixed Point

· Chapters 1 and 2 of my notes: Hema_Notes

· Rust, "Optimal Replacement of GMC Bus Engines: An Empirical Model of Harold Zurcher," Econometrica, (1987)

· Application Papers

o Hendel & Nevo, “Measuring the Implications of Sales and Consumer Inventory Behavior,” Econometric (2006)

o Song & Chintagunta, “A Micromodel of New Product Adoption with Heterogeneous and Forward-Looking Consumers: Application to the Digital Camera Category,” QME (2003)

o Erdem, Imai, & Keane, “Brand and Quantity Choice Dynamics Under Price Uncertainty” QME (2003)

Classes 3 and 4: Two-step methods

· Hotz & Miller, “Conditional Choice Probabilities and the Estimation of Dynamic Models,” The Review of Economic Studies (1993)

· Bajari, Benkard, & Levin, “Estimating Dynamic Models of Imperfect Competition,” Econometrica (2007)

· Arcidiacono & Ellickson, “Practical Methods for Estimation of Dynamic Discrete Choice Models,” Annual Review of Economics, (2011). (Read only the parts without persistent unobservables for this class)

· Notes posted on Canvas

· Application Papers

o Ellickson & Misra, “Supermarket Pricing Strategies,” Marketing Science (2008)

o Ryan & Tucker, “Heterogeneity and the dynamics of technology adoption,” QME (2012)

Class 5: Persistent Unobservables & Nested Fixed Point with EM

· Chapter 5 of my notes: Hema_Notes

· Arcidiacono & Jones, “Finite mixture distributions, sequential likelihood and the EM algorithm,” Econometrica (2003)

· Get background reading on EM algorithm. In this class, you should know what an EM algorithm. Start with Dempster’s classic paper from 1977.

· Application Papers

o Arcidiacono, “Affirmative action in higher education: How do admission and financial aid rules affect future earnings?” Econometrica (2005)

o Arcidiacono, Sieg, Sloan, “Living Rationally Under the Volcano? An Empirical Analysis of Heavy Drinking and Smoking” International Economic Review (2007)

Class 7: Two-step methods with EM

· Chapter 6 of my notes: Hema_Notes

· Arcidiacono & Miller, “Conditional choice probability estimation of dynamic discrete choice models with unobserved heterogeneity,” Econometrica (2011)

· Arcidiacono & Ellickson, “Practical Methods for Estimation of Dynamic Discrete Choice Models,” Annual Review of Economics, 2011. (Finish the parts with persistent unobservables for this class)

· Application Papers

o Yoganarasimhan, “The Value of Reputation in an Online Freelance Marketplace,” Marketing Science (2013)

o Yoganarasimhan, “Estimation of Beauty Contest Auctions,” Marketing Science (2016)

Class 8: Identification

· Chapter 7 of my notes: Hema_Notes

· Magnac & Thesmar, “Identifying Dynamic Discrete Decision Processes,” Econometrica (2002)

· Kasahara & Shimotsu, “Nonparametric Identification of Finite Mixture Models of Dynamic Discrete Choices,” Econometrica (2009)

· Application Papers

o Read the identification proof in the Web Appendix of Yoganarasimhan, “The Value of Reputation in an Online Freelance Marketplace,” Marketing Science (2013)

o Read the identification proof in the Web Appendix of Yoganarasimhan, “Estimation of Beauty Contest Auctions,” Marketing Science (2016)

Class 9: MPEC Methods

· Chapter 8 of my notes: Hema_Notes

· Su & Judd, “Constrained Optimization Approaches to Estimation of Structural Models,” Econometrica (2012)

· Application Papers

o Berry, Levinsohn, Pakes, “Automobile Prices in Market Equilibrium” Econometrica (1995)

o Dube, Fox, & Su, “Improving the Numerical Performance of Static and Dynamic Aggregate Discrete Choice Random Coefficients Demand Estimation,” Econometrica (2012)

· Background Reading:

o Luo, Pang, Ralph, “Mathematical Programs with Equilibrium Constraints,” Cambridge University Press (1996)

Some survey papers and/or notes that may help you.

· Aguirregabiria & Mira, “Dynamic discrete choice structural models: A survey,” 2010

· The class notes of Matthew Shum available at: http://people.hss.caltech.edu/~mshum/gradio/ioclass.html

· Aguirregabiria’s class notes available at: http://www.iub.edu/~caepr/visitors/2007/files/2007-02-01-03.pdf