CS 286r. Iterative Combinatorial Exchanges.

Prof. David C. Parkes

schedule | announcements | assignments | projects | discussion | final papers


General Information
Note: New time, TR 11.30am-1pm.
Location: MD 319
Section: Fridays, 2pm. Room MD 221

This graduate-level seminar will focus on the design, prototyping, and testing of an iterative combinatorial exchange. A combinatorial exchange is an electronic market in which multiple buyers and sellers can express complex preferences, for example "I will sell A and B if I can also buy C", or "I will sell one of A, B or C." An iterative combinatorial exchange is one in which participants (call them "agents") can refine their bids over time, rather than submitting a single sealed bid. Iterative combinatoral exchanges (ICE's) are of considerable interest to a number of government agencies, including the Federal Communications Commission (FCC), see the recent combinatorial auction and bidding conference, the Federal Aviation Administration (FAA), see the recent combinatorial auction workshop, the Navy (in the guise of an exchange to match sailors with particular duties), and the Federal Energy Regulatory Commission (FERC).

The FCC is interested in using combinatorial exchanges as a "big bang" device, to enable the efficient reallocation of wireless spectrum that has been inefficiently allocated over the past 50 years through administrative processes and lotteries. Current spectrum is highly fragmented, in terms of control, geography, and frequence. An exchange could allow incumbents, such as TV stations and cellular operators, to sell license to new entrants and amongst themselves. The FAA is interested in using combinatorial exchanges as a device to reallocate take-off and landing slots at airports, with airlines and airports acting as buyers and sellers. FERC is exploring the idea of using combinatorial exchanges to design markets that allow the simultaneous clearing of generation capacity and transmission capacity. Current markets tend to clear these two resources separately, leading to considerable strategic vulnerabilities and inefficiencies. Unfortunately, there is currently no good, tested, design for an iterative combinatorial exchange. Recently, Prof. Parkes presented a proposed design to the FCC combinatorial auction workshop held this past Fall. The goal of this seminar is to:

The class will form four groups, with each group focused on one particular component of this problem. The groups will be designed around the following focus areas: Along the way we will read papers on winner determination algorithms, bidding languages, auction theory, double auctions, and matching markets. Outside speakers will include FCC Senior Economist, Evan Kwerel, and George Donohue, Proferssor of Systems Engineering and Operations Research, George Mason University, and former Associate Administrator of Research and Acquistions, FAA. We will work with the FCC, and with George Donohue, to get some real data on which to test out final prototype.

Prerequisites: This course draws on a broad set of ideas, from CS theory, to AI, to economics. Formal requirements include a basic course in linear algebra (such as M 21b, AM 21b, or equivalent), CS 121 (complexity theory) and CS 124 (algorithms), and a course in AI, such as CS 181 or CS 182. Students can petition to substitute an economic theory course for an AI course. A course in advanced algorithms, such as CS 223, Probabilistic algorithms, or CS 226r, Efficient algorithms is preferred but not required. Knowledge of game theory, microeconomics, and auction theory, is very helpful but not required. EC 1011a and EC 1011b, or equivalents, are recommended. Mathematics will be fundamental to the course (especially linear-programming and duality), and students should expect to learn additional mathematics on their own as necessary. I recommend that students unsure about their background for the course read a couple of papers from the reading list, and attend my office hours during the first week.

Time/Location: The course meets on Tuesdays and Thursdays, 11.30am-1pm, room MD 319

Limited Enrollment: Enrollment will be restricted to 20-25 students to facilitate seminar-style discussion of papers. A form will be distributed in the first class to help with the selection of students, with preference given to graduate students, and students with the strongest background.

Course Staff:
Name E-mail Office Hours
David Parkes parkes@eecs.harvard.edu Mon, Wed, 2-3pm, MD 229 (not 2/11)
Sebastien Lahaie slahaie@eecs.harvard.edu Mon, Wed 4-5pm, MD 219

Additional early-term office hours: Mon 2/9, 10am-12pm; Tue 2/10, 1.30-2.30pm. (Maxwell-Dworkin, 229).

Missed Course Materials: Sebastien Lahaie, Maxwell-Dworkin 219.

Course Structure: This will be a non-standard course. Our time will be split between lectures (40%), student presentations and discussion (50%) and outside speakers (10%).

The lectures are designed to provide a high level overview of the technical material that you will need to carefully read and understand for the class to make progress in designing, prototyping, and testing an iterative combinatorial exchage. They will necessarily move along quickly. There will be a few small homework exercises on some basic economic concepts. These problem sets are designed to help with understanding of the background material that is covered in the introductory lectures.

Students will form groups, with each group focusing on one aspect of the iterative combinatorial exchange design problem. Student presentations will mainly focus on academic papers related to their slice of their problem, along with progress reports. Sometimes we'll also have the whole class read a particularly important paper.

At the end of the class, each group will give a final presentation on what they learned, and what they achieved in the class. Each group will also be expected to submit one or two reports, describing either the theory that related to their component of the design problem or their practical experience and results.

Final grades will be determined according to the following approximate guidelines.
Problem Sets 20% 3 Short Problem Sets on Introductory Material
Participation 20% Particpation in class discussion.
Class presentations. 35% Presentations in class, of research papers and ideas related to a particular component of the exchange design problem.
Final Results, and Report. 35% Amount of progress made, Quality of research, Quality of final report and presentation.

Readings: There is no set text for this course. Readings will be made available electronically, and also distributed in class. Student presentations: In later weeks, two groups will make presentations to the class on one, or both, days of the week. Each group will provide a progress report and present a paper to the class from their set of papers that relates to their problem, clearly explaining how it helps them to understand then task at hand.

Groups can choose to come to office hours ahead of class to talk about the paper that they are presenting.


Course Announcements:

Class Project: The goal of this class is to deliver a working prototype of an iterative combinatorial exchange. Final reports will be due (up to two per group) on Tuesday May 17th, with presentations on Wednesday May 12th.

Related Courses at Harvard:

Related Courses at Other Universities:

Final Papers: