Upload
rebekah-mcclain
View
28
Download
2
Tags:
Embed Size (px)
DESCRIPTION
September 25, 2007. Reorganization in Matching Markets The Dynamics of Reorganization in Matching Markets: A Laboratory Experiment Motivated by A Natural Experiment - John H Kagel and Alvin E Roth. Presentation for: MGT 703: Experimental Economics. What’s so special about March 15, 2007?. - PowerPoint PPT Presentation
Citation preview
Reorganization in Matching Markets
The Dynamics of Reorganization in Matching Markets: A Laboratory Experiment Motivated by A Natural Experiment- John H Kagel and Alvin E Roth
September 25, 2007
Presentation for: MGT 703: Experimental Economics
Yale School of Management 20070925_Matching_Games.ppt 2
It’s Match Day!
What’s so special about March 15, 2007?
The National Residency Matching Program matches residents and independent applications to residencies all over the country
In 2006, the program matched– 16,000 medical school students– 18,000 independent applicants– 24,085 positions
Efficient matching is what keeps thousands of medical residents across the country happy!
This is what matching markets are about
Yale School of Management 20070925_Matching_Games.ppt 3
Matching markets
Experimental setup
Results from experiment
Discussion
Agenda
Yale School of Management 20070925_Matching_Games.ppt 4
Why is matching important?
Unstable matches can occur because– Types are not known (signaling, pre-emption, etc.)– Known congestion in the system
But matching efficiently is not easy– Decentralized matching may be acceptable in some cases– Other situations call for centralized mechanisms
Matching markets are centralized clearinghouses– They attempt to match up as ‘fairly’ as
possible
Efficient matching markets allow parties to match with their best possible counterpart– The “fit” factor in job interviews– A stable match is beneficial to both parties
Yale School of Management 20070925_Matching_Games.ppt 5
What happens when matching markets fail?
An example: In mid-1960s regional market for new physicians in UK unraveled– Medical students were being made offers a year and a half prior o
graduation
Matching markets can unravel– Individual agents attempt to pre-empt the matching market– This can have significant costs
- Mismatches based on insufficient data- Lost opportunities
Another example: Market for Federal court clerks (US) – Becker, Beyer and Calabresi (1994) call for reorganization of
matching market due to earlier unraveling– This attempt was also unsuccessful
Yale School of Management 20070925_Matching_Games.ppt 6
How do we match?
In the example from the UK, two algorithms were used– “Newcastle” algorithm – introduced in Newcastle and Birmingham in 1966-67
- Resulted in unstable match-ups- Unraveled: by 1981, 80% of preferences submitted by both students and positions indicated only a mutual first choice - This is “match-fixing” in its truest form!
– Delayed Acceptance (DA) algorithm – introduced in Edinburgh in 1969 and Cardiff in 1971- Based on Gale-Shapley (1962) result- Resulted in stable match-ups
Centralized matching is usually achieved through algorithms– Take into account preferences of both parties– Result in ‘market-clearing’ (to the extent possible…)
Yale School of Management 20070925_Matching_Games.ppt 7
The Newcastle Algorithm
Step 1: Firms and workers submit their preferences
Step 2: Each firm-worker combination is ranked (ascending order)– Based on product of firm ranking of worker multiplied by worker’s ranking of firm
Step 3: If necessary, ties are broken based on local rules– i.e. more weight can be given to the worker’s preference or the firm’s preference
Step 4: Matches are identified from the ranked list
Yale School of Management 20070925_Matching_Games.ppt 8
Delayed Acceptance Algorithm
Step 1:
Each employer makes an offer to its highest ranked acceptable student
Step k:
(i) Each worker rejects all but the highest-ranked of the acceptable offers she received in steps 1 through k – 1
(ii) Each employer who has had an offer rejected in part (i) of step k, makes an offer to its highest ranked acceptable worker who has not yet rejected its offer
(iii) The algorithm stops at any step k = T at which no rejections are issued, and the resulting matching places each worker with the employer (if any) whose offer she has not rejected
Yale School of Management 20070925_Matching_Games.ppt 9
Motivation of the Experiment
Authors have seen a “natural experiment” in introduction of centralized market clearinghouses (UK, late 1960s)
In this experiment, regions introduced centralized matching algorithms to halt unraveling– Inconsistent success; some regions succeeded in slowing unraveling; others did not– Initial “pull-back” in pre-emptive decentralized matching– Longer term effects depend on algorithm efficacy
Can we replicate such an experiment in the lab?
What will it tell us about transitions to centralized matching markets?
Yale School of Management 20070925_Matching_Games.ppt 10
Experimental Setup
Steps in Experiment
Six firms are asked to match with six workers
– Three “high” type and three “low” type of each
–Known payoff tables
Three time periods: -2, -1 and 0
One-offer per period rule except when centralized matching occurs
Offering/accepting early has costs
First 10 rounds are all decentralized matching
Subsequent 15 rounds have centralized matching in period 0
Sample Payoff Table
Payoffs shown to Firm 1 (a high-productivity firm)
Yale School of Management 20070925_Matching_Games.ppt 11
Sample Payoff Matrix
Comments
Setup is designed to cause congestion
Making any match is better than making no match
Big upside for correct matches for “high” types
Yale School of Management 20070925_Matching_Games.ppt 12
Results (1): Efficacy of centralized matching
Yale School of Management 20070925_Matching_Games.ppt 13
: Centralized matching initially reduces pre-emptive matchesResults (2)
Yale School of Management 20070925_Matching_Games.ppt 14
Results (3): Mismatch costs modulate levels of unraveling
Yale School of Management 20070925_Matching_Games.ppt 15
Discussion
Simple experiment that maps closely to what was observed in reality
Creation of an in-lab model of regime changes– Provides a lot of insight over and above a static model– Authors also explore modulating factors
In our preliminary in-class test, I hope to see some unraveling occurring
Questions/Comments?