dynamic matching Œpart 2 · dynamic matching processes with changing participants common to many...
TRANSCRIPT
![Page 1: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/1.jpg)
Dynamic Matching —Part 2
Leeat Yariv
Yale University
February 23, 2016
![Page 2: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/2.jpg)
Dynamic Matching Processes with Changing Participants
Common to many matching processes:
I Child Adoption:I About 1.6 million, or 2.5%, of children in the U.S. are adoptedI One adoption facilitator —11 new potential parents and 13newly relinquished children each month
I Labor Markets:I U.S. Bureau of Labor Statistics —approximately 5 million newjob openings and slightly fewer than 5 million newlyunemployed workers each month this year
I Organ Donation:I Organ Donation and Transplantation Statistics —a new patientadded to the kidney transplant list every 14 minutes and about3000 patients added each month
I In 2013, about a third of ˜17, 000 kidney transplants involvedliving donors
![Page 3: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/3.jpg)
Dynamic Matching Processes with Changing Participants
Common to many matching processes:
I Child Adoption:I About 1.6 million, or 2.5%, of children in the U.S. are adoptedI One adoption facilitator —11 new potential parents and 13newly relinquished children each month
I Labor Markets:I U.S. Bureau of Labor Statistics —approximately 5 million newjob openings and slightly fewer than 5 million newlyunemployed workers each month this year
I Organ Donation:I Organ Donation and Transplantation Statistics —a new patientadded to the kidney transplant list every 14 minutes and about3000 patients added each month
I In 2013, about a third of ˜17, 000 kidney transplants involvedliving donors
![Page 4: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/4.jpg)
Dynamic Matching Processes with Changing Participants
Common to many matching processes:
I Child Adoption:I About 1.6 million, or 2.5%, of children in the U.S. are adoptedI One adoption facilitator —11 new potential parents and 13newly relinquished children each month
I Labor Markets:I U.S. Bureau of Labor Statistics —approximately 5 million newjob openings and slightly fewer than 5 million newlyunemployed workers each month this year
I Organ Donation:I Organ Donation and Transplantation Statistics —a new patientadded to the kidney transplant list every 14 minutes and about3000 patients added each month
I In 2013, about a third of ˜17, 000 kidney transplants involvedliving donors
![Page 5: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/5.jpg)
Two Questions
I What outcomes do decentralized interactions generate?
I How to optimally match individuals in a dynamic market?
I =⇒ What are environments in which centralized interventionwould be particularly beneficial?
![Page 6: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/6.jpg)
Two Questions
I What outcomes do decentralized interactions generate?
I How to optimally match individuals in a dynamic market?
I =⇒ What are environments in which centralized interventionwould be particularly beneficial?
![Page 7: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/7.jpg)
Today
I (In brief) Kidney exchange
I (In detail) Optimal dynamic mechanisms in a particular setting
![Page 8: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/8.jpg)
Kidney Exchange
I Original References: Roth, Sonmez, and Ünver (2004, 2005,2007), Ünver (2010)
I Underlying Trade-off: Compatibility between pairs andwaiting time
![Page 9: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/9.jpg)
Compatibility between Pairs
I Blood compatibility
I Tissue type compatibility: Percentage Reactive Antibodies(PRA)
![Page 10: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/10.jpg)
Compatibility between Pairs
I Blood compatibility
I Tissue type compatibility: Percentage Reactive Antibodies(PRA)
![Page 11: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/11.jpg)
![Page 12: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/12.jpg)
Pairwise Kidney Exchange
I A patient-donor pair arrives at each period:I Each patient brings a dedicated donor (family member, friend)I Independent, and similar, arrival of patients and donors (goodsamaritans, cadavers)
I If waiting were costless, could we get everyone matched?I Discard of all compatible pairsI Consider n remaining pairs, where each two pairs arecompatible with probability p
![Page 13: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/13.jpg)
Pairwise Kidney Exchange
I A patient-donor pair arrives at each period:I Each patient brings a dedicated donor (family member, friend)I Independent, and similar, arrival of patients and donors (goodsamaritans, cadavers)
I If waiting were costless, could we get everyone matched?
I Discard of all compatible pairsI Consider n remaining pairs, where each two pairs arecompatible with probability p
![Page 14: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/14.jpg)
Pairwise Kidney Exchange
I A patient-donor pair arrives at each period:I Each patient brings a dedicated donor (family member, friend)I Independent, and similar, arrival of patients and donors (goodsamaritans, cadavers)
I If waiting were costless, could we get everyone matched?I Discard of all compatible pairsI Consider n remaining pairs, where each two pairs arecompatible with probability p
![Page 15: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/15.jpg)
Large Random Networks
I Consider a graph with n nodes and each link occurring withprobability p
I Divide the n pairs randomly into two groups of k pairs, callthem S and R
I Construct the bi-partite random graph with S and R as nodesI Each node in S is connected to (compatible with) each nodein R with probability p
![Page 16: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/16.jpg)
Large Random Networks
I Consider a graph with n nodes and each link occurring withprobability p
I Divide the n pairs randomly into two groups of k pairs, callthem S and R
I Construct the bi-partite random graph with S and R as nodes
I Each node in S is connected to (compatible with) each nodein R with probability p
![Page 17: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/17.jpg)
Large Random Networks
I Consider a graph with n nodes and each link occurring withprobability p
I Divide the n pairs randomly into two groups of k pairs, callthem S and R
I Construct the bi-partite random graph with S and R as nodesI Each node in S is connected to (compatible with) each nodein R with probability p
![Page 18: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/18.jpg)
Large Random Networks (2)
![Page 19: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/19.jpg)
Large Random Networks (3)
Erdos and Renyi (1964): As long as the graph is connectedenough, namely as long as p approaches 0 slower than log n
n , thereis a perfect matching with probability approaching 1.
I In fact, p is bounded above 0, even when we consider differentblood types
I So if we wait long enough, we can match everyone!
![Page 20: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/20.jpg)
Large Random Networks (3)
Erdos and Renyi (1964): As long as the graph is connectedenough, namely as long as p approaches 0 slower than log n
n , thereis a perfect matching with probability approaching 1.
I In fact, p is bounded above 0, even when we consider differentblood types
I So if we wait long enough, we can match everyone!
![Page 21: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/21.jpg)
Market Design —Two Problems
I In reality, waiting is costly. So, how should we create matches?I Maximize instantaneous matches; orI Keep useful donors (say, those with O blood type) for thefuture
I Overall maximization of matched pairs may not maximize aparticular hospital’s number of matched pairs (Ashlagi andRoth, 2013)
![Page 22: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/22.jpg)
Market Design —Two Problems
I In reality, waiting is costly. So, how should we create matches?I Maximize instantaneous matches; orI Keep useful donors (say, those with O blood type) for thefuture
I Overall maximization of matched pairs may not maximize aparticular hospital’s number of matched pairs (Ashlagi andRoth, 2013)
![Page 23: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/23.jpg)
To Wait or not to Wait
I Ünver (2010) —Multi-way exchanges may be useful
I Akbarpour, Li, and Oveis Gharan (2015) —Holding on tosome pairs could be useful
I No characterization of the optimal mechanism
![Page 24: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/24.jpg)
To Wait or not to Wait
I Ünver (2010) —Multi-way exchanges may be useful
I Akbarpour, Li, and Oveis Gharan (2015) —Holding on tosome pairs could be useful
I No characterization of the optimal mechanism
![Page 25: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/25.jpg)
To Wait or not to Wait
I Ünver (2010) —Multi-way exchanges may be useful
I Akbarpour, Li, and Oveis Gharan (2015) —Holding on tosome pairs could be useful
I No characterization of the optimal mechanism
![Page 26: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/26.jpg)
Optimal Dynamic Matching
I Reference: Baccara, Li, and Yariv (2015)
I Our (Broad) Goal: In a simple setting,I Analyze the optimal matching protocolI Compare it with a decentralized setting as well as with a fewsimple interventions
![Page 27: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/27.jpg)
Underlying Trade-off
I Quality of matches varies, thick markets are beneficial⇒ incentive to accumulate participants
I Adoption process —child’s gender and race, parents’demographics, etc.
I Labor markets — skills of workers, quality of firms, etc.
![Page 28: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/28.jpg)
Underlying Trade-off
I Quality of matches varies, thick markets are beneficial⇒ incentive to accumulate participants
I Adoption process —child’s gender and race, parents’demographics, etc.
I Labor markets — skills of workers, quality of firms, etc.
![Page 29: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/29.jpg)
Underlying Trade-off
I Quality of matches varies, thick markets are beneficial⇒ incentive to accumulate participants
I Waiting is costly ⇒ incentive to match quicklyI Adoption process —gestation expenses for birth mothers, timespent with social services, parents’legal fees, etc. (Baccara,Collard-Wexler, Felli, and Yariv 2014)
I Labor markets — foregone wages for the unemployed, lackingworkforce for employers, etc.
![Page 30: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/30.jpg)
Underlying Trade-off
I Quality of matches varies, thick markets are beneficial⇒ incentive to accumulate participants
I Waiting is costly ⇒ incentive to match quicklyI Adoption process —gestation expenses for birth mothers, timespent with social services, parents’legal fees, etc. (Baccara,Collard-Wexler, Felli, and Yariv 2014)
I Labor markets — foregone wages for the unemployed, lackingworkforce for employers, etc.
![Page 31: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/31.jpg)
Main Questions
I What is the optimal dynamic mechanism to balance thistrade-off?
I Hold on to a stock of certain agents up to a threshold, matchothers immediately
I What is the welfare gain relative to decentralization?I Decentralization leads to ineffi cient long queues
I What are simple, and useful, interventions?I Budget-balanced tax/subsidy schemesI Matching in fixed time intervals
![Page 32: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/32.jpg)
Main Questions
I What is the optimal dynamic mechanism to balance thistrade-off?
I Hold on to a stock of certain agents up to a threshold, matchothers immediately
I What is the welfare gain relative to decentralization?I Decentralization leads to ineffi cient long queues
I What are simple, and useful, interventions?I Budget-balanced tax/subsidy schemesI Matching in fixed time intervals
![Page 33: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/33.jpg)
Main Questions
I What is the optimal dynamic mechanism to balance thistrade-off?
I Hold on to a stock of certain agents up to a threshold, matchothers immediately
I What is the welfare gain relative to decentralization?I Decentralization leads to ineffi cient long queues
I What are simple, and useful, interventions?I Budget-balanced tax/subsidy schemesI Matching in fixed time intervals
![Page 34: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/34.jpg)
A Model of Dynamic Matching
I Market composed of squares and rounds looking to match,evolves over time
I Squares and rounds: children and adoptive parents, workersand firms,...
I At each period, one square and one round arrive at the marketI Square: type A with pr p, type B with pr 1− pI Round: type α with pr p, type β with pr 1− pI Each agent suffers a cost c > 0 for each period on the marketwithout a match
I Note: Symmetry makes things simpler (but not crucial),could assume random arrival of pairs
![Page 35: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/35.jpg)
A Model of Dynamic Matching
I Market composed of squares and rounds looking to match,evolves over time
I Squares and rounds: children and adoptive parents, workersand firms,...
I At each period, one square and one round arrive at the marketI Square: type A with pr p, type B with pr 1− pI Round: type α with pr p, type β with pr 1− pI Each agent suffers a cost c > 0 for each period on the marketwithout a match
I Note: Symmetry makes things simpler (but not crucial),could assume random arrival of pairs
![Page 36: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/36.jpg)
A Model of Dynamic Matching
I Market composed of squares and rounds looking to match,evolves over time
I Squares and rounds: children and adoptive parents, workersand firms,...
I At each period, one square and one round arrive at the marketI Square: type A with pr p, type B with pr 1− pI Round: type α with pr p, type β with pr 1− p
I Each agent suffers a cost c > 0 for each period on the marketwithout a match
I Note: Symmetry makes things simpler (but not crucial),could assume random arrival of pairs
![Page 37: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/37.jpg)
A Model of Dynamic Matching
I Market composed of squares and rounds looking to match,evolves over time
I Squares and rounds: children and adoptive parents, workersand firms,...
I At each period, one square and one round arrive at the marketI Square: type A with pr p, type B with pr 1− pI Round: type α with pr p, type β with pr 1− pI Each agent suffers a cost c > 0 for each period on the marketwithout a match
I Note: Symmetry makes things simpler (but not crucial),could assume random arrival of pairs
![Page 38: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/38.jpg)
A Model of Dynamic Matching
I Market composed of squares and rounds looking to match,evolves over time
I Squares and rounds: children and adoptive parents, workersand firms,...
I At each period, one square and one round arrive at the marketI Square: type A with pr p, type B with pr 1− pI Round: type α with pr p, type β with pr 1− pI Each agent suffers a cost c > 0 for each period on the marketwithout a match
I Note: Symmetry makes things simpler (but not crucial),could assume random arrival of pairs
![Page 39: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/39.jpg)
A Model of Dynamic Matching (2)
I Ux (y) is the utility of a type x from matching a type y
I Assortative preferences:
Ux (A) > Ux (B) for x = α, β,
Ux (α) > Ux (β) for x = A,B
I Convenient to denote:
Uxy ≡ Ux (y) + Uy (x) for x = A,B, y = α, β
I Super-modularity assumption:
U ≡ UAα + UBβ − UAβ − UBα > 0
=⇒ The effi cient matching entails the maximal number of (A, α)and (B, β) pairs
![Page 40: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/40.jpg)
A Model of Dynamic Matching (2)
I Ux (y) is the utility of a type x from matching a type yI Assortative preferences:
Ux (A) > Ux (B) for x = α, β,
Ux (α) > Ux (β) for x = A,B
I Convenient to denote:
Uxy ≡ Ux (y) + Uy (x) for x = A,B, y = α, β
I Super-modularity assumption:
U ≡ UAα + UBβ − UAβ − UBα > 0
=⇒ The effi cient matching entails the maximal number of (A, α)and (B, β) pairs
![Page 41: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/41.jpg)
A Model of Dynamic Matching (2)
I Ux (y) is the utility of a type x from matching a type yI Assortative preferences:
Ux (A) > Ux (B) for x = α, β,
Ux (α) > Ux (β) for x = A,B
I Convenient to denote:
Uxy ≡ Ux (y) + Uy (x) for x = A,B, y = α, β
I Super-modularity assumption:
U ≡ UAα + UBβ − UAβ − UBα > 0
=⇒ The effi cient matching entails the maximal number of (A, α)and (B, β) pairs
![Page 42: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/42.jpg)
A Model of Dynamic Matching (3)
I In all the matching protocols we consider, we assume agentsleave the market by matching
I Bounds on utilities from remaining unmatched assure this isindividually rational
![Page 43: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/43.jpg)
The Optimal Mechanism
I Consider general dynamic mechanisms —a social planner cancreate matches at every period
I At any time t, before a new square-round pair arrives, a queueis represented by (kA, kα, kB , kβ)
I Length of queues of squares: kA, kBI Length of queues of rounds: kα, kβ
![Page 44: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/44.jpg)
The Optimal Mechanism
I Consider general dynamic mechanisms —a social planner cancreate matches at every period
I At any time t, before a new square-round pair arrives, a queueis represented by (kA, kα, kB , kβ)
I Length of queues of squares: kA, kBI Length of queues of rounds: kα, kβ
![Page 45: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/45.jpg)
Summary Timeline
![Page 46: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/46.jpg)
The Optimal Mechanism
I We focus on stationary and deterministic mechanisms
I At time t, after a new square-round pair enters the market,there is a queue st = (stA, s
tB , s
tα, s
tβ)
I A mechanism is characterized by a mapping µ : Z4+ → Z4
+
such that for every s ∈ Z4,µ(s) = m = (mAα,mAβ,mBα,mBβ) is a feasible profile ofmatches:
I mxα +mx β ≤ sx for x ∈ {A,B}I mAy +mBy ≤ sy for y ∈ {α, β}
![Page 47: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/47.jpg)
The Optimal Mechanism
I We focus on stationary and deterministic mechanisms
I At time t, after a new square-round pair enters the market,there is a queue st = (stA, s
tB , s
tα, s
tβ)
I A mechanism is characterized by a mapping µ : Z4+ → Z4
+
such that for every s ∈ Z4,µ(s) = m = (mAα,mAβ,mBα,mBβ) is a feasible profile ofmatches:
I mxα +mx β ≤ sx for x ∈ {A,B}I mAy +mBy ≤ sy for y ∈ {α, β}
![Page 48: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/48.jpg)
The Optimal Mechanism
I We focus on stationary and deterministic mechanisms
I At time t, after a new square-round pair enters the market,there is a queue st = (stA, s
tB , s
tα, s
tβ)
I A mechanism is characterized by a mapping µ : Z4+ → Z4
+
such that for every s ∈ Z4,µ(s) = m = (mAα,mAβ,mBα,mBβ) is a feasible profile ofmatches:
I mxα +mx β ≤ sx for x ∈ {A,B}I mAy +mBy ≤ sy for y ∈ {α, β}
![Page 49: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/49.jpg)
I New queue k = (kA, kB , kα, kβ) contains remaining agents:
kx = sx − (mxα +mxβ) for x ∈ {A,B}ky = sy − (mAy +mBy ) for y ∈ {α, β}
![Page 50: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/50.jpg)
Summary Timeline
![Page 51: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/51.jpg)
The Optimal Mechanism
I The generated surplus by matches m is:
S(m) ≡ ∑(x ,y )∈{A,B}×{α,β}
mxyUxy
I Waiting costs incurred by remaining agents k is:
C (s,m) ≡ c
∑x∈{A,B ,α,β}
kx
I The welfare generated at time t is:
w(st ,mt ) ≡ S(mt )− C (st ,mt )
![Page 52: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/52.jpg)
The Optimal Mechanism
I The generated surplus by matches m is:
S(m) ≡ ∑(x ,y )∈{A,B}×{α,β}
mxyUxy
I Waiting costs incurred by remaining agents k is:
C (s,m) ≡ c
∑x∈{A,B ,α,β}
kx
I The welfare generated at time t is:
w(st ,mt ) ≡ S(mt )− C (st ,mt )
![Page 53: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/53.jpg)
The Optimal Mechanism
I The generated surplus by matches m is:
S(m) ≡ ∑(x ,y )∈{A,B}×{α,β}
mxyUxy
I Waiting costs incurred by remaining agents k is:
C (s,m) ≡ c
∑x∈{A,B ,α,β}
kx
I The welfare generated at time t is:
w(st ,mt ) ≡ S(mt )− C (st ,mt )
![Page 54: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/54.jpg)
The Optimal Mechanism
I The value of each mechanism is assessed using the averagewelfare, defined as:
W (µ) ≡ limT→∞
1T
E
[T
∑t=1w(st , µ(st ))
]I The average welfare function is well-defined: the limit existsfor every mechanism µ
I An optimal mechanism is a mechanism achieving the maximalaverage welfare
I An optimal mechanism exists since there is only a finitenumber of stationary and deterministic mechanisms leading toa bounded stock of agents in each period
![Page 55: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/55.jpg)
The Optimal Mechanism
I The value of each mechanism is assessed using the averagewelfare, defined as:
W (µ) ≡ limT→∞
1T
E
[T
∑t=1w(st , µ(st ))
]I The average welfare function is well-defined: the limit existsfor every mechanism µ
I An optimal mechanism is a mechanism achieving the maximalaverage welfare
I An optimal mechanism exists since there is only a finitenumber of stationary and deterministic mechanisms leading toa bounded stock of agents in each period
![Page 56: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/56.jpg)
The Optimal Mechanism
I The value of each mechanism is assessed using the averagewelfare, defined as:
W (µ) ≡ limT→∞
1T
E
[T
∑t=1w(st , µ(st ))
]I The average welfare function is well-defined: the limit existsfor every mechanism µ
I An optimal mechanism is a mechanism achieving the maximalaverage welfare
I An optimal mechanism exists since there is only a finitenumber of stationary and deterministic mechanisms leading toa bounded stock of agents in each period
![Page 57: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/57.jpg)
The Optimal Mechanism
Lemma 1 Any optimal mechanism requires (A, α) and (B, β)pairs to be matched as soon as they becomeavailable.
Intuition:
I If we hold on to an (A, α) pair, it is to match the agents witha future B-square and β-round
I But super-modularity ⇒ matching A with α and B with βmore effi cient
I May as well match (A, α) immediatelyI Same for a (B, β) pair
![Page 58: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/58.jpg)
The Optimal Mechanism
Lemma 1 Any optimal mechanism requires (A, α) and (B, β)pairs to be matched as soon as they becomeavailable.
Intuition:
I If we hold on to an (A, α) pair, it is to match the agents witha future B-square and β-round
I But super-modularity ⇒ matching A with α and B with βmore effi cient
I May as well match (A, α) immediatelyI Same for a (B, β) pair
![Page 59: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/59.jpg)
The Optimal Mechanism
Lemma 1 Any optimal mechanism requires (A, α) and (B, β)pairs to be matched as soon as they becomeavailable.
Intuition:
I If we hold on to an (A, α) pair, it is to match the agents witha future B-square and β-round
I But super-modularity ⇒ matching A with α and B with βmore effi cient
I May as well match (A, α) immediatelyI Same for a (B, β) pair
![Page 60: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/60.jpg)
The Optimal Mechanism
Lemma 1 Any optimal mechanism requires (A, α) and (B, β)pairs to be matched as soon as they becomeavailable.
Intuition:
I If we hold on to an (A, α) pair, it is to match the agents witha future B-square and β-round
I But super-modularity ⇒ matching A with α and B with βmore effi cient
I May as well match (A, α) immediately
I Same for a (B, β) pair
![Page 61: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/61.jpg)
The Optimal Mechanism
Lemma 1 Any optimal mechanism requires (A, α) and (B, β)pairs to be matched as soon as they becomeavailable.
Intuition:
I If we hold on to an (A, α) pair, it is to match the agents witha future B-square and β-round
I But super-modularity ⇒ matching A with α and B with βmore effi cient
I May as well match (A, α) immediatelyI Same for a (B, β) pair
![Page 62: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/62.jpg)
Proposition 1 An optimal dynamic mechanism is identified by apair of thresholds (k̄A, k̄α) ∈ Z+ such that
1. whenever more than k̄A A-squares are present, sA − k̄A pairsof type (A, β) are matched immediately, and
2. whenever more than k̄α α-rounds are present, sα − k̄α pairs oftype (B, α) are matched immediately.
Symmetry ⇒ k̄A = k̄α = k̄ .
![Page 63: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/63.jpg)
Proposition 1 An optimal dynamic mechanism is identified by apair of thresholds (k̄A, k̄α) ∈ Z+ such that
1. whenever more than k̄A A-squares are present, sA − k̄A pairsof type (A, β) are matched immediately, and
2. whenever more than k̄α α-rounds are present, sα − k̄α pairs oftype (B, α) are matched immediately.
Symmetry ⇒ k̄A = k̄α = k̄ .
![Page 64: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/64.jpg)
Structure of the Optimal Mechanism
I Denote by kAα = kA − kα
I kAα follows a Markov process, where states correspond tovalues −k̄ ≤ kAα ≤ k̄
![Page 65: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/65.jpg)
Structure of the Optimal Mechanism
![Page 66: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/66.jpg)
Structure of the Optimal Mechanism
I When −k < kAα < k,I Stock does not change if (A, α) or (B, β) arrive =⇒probability p2 + (1− p)2
I Stock changes (up or down) if (A, β) or (B, α) arrive =⇒probability for either p(1− p)
![Page 67: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/67.jpg)
Structure of the Optimal Mechanism
![Page 68: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/68.jpg)
Finding the Optimal Mechanism
I To characterize the optimal k, look for the steady state
I xt+1 = Tk̄xt (a Markov process with 2k + 1 states)
Tk=
1− p(1− p) p(1− p) 0 0p(1− p) p2+(1− p)2
0 ..
p2+(1− p)2 p(1− p)0 p(1− p) 1− p(1− p)
I This process is ergodic =⇒ There is a unique steady statedistribution, which is the uniform distribution over states =⇒each state has probability 1
2k+1
![Page 69: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/69.jpg)
Finding the Optimal Mechanism
I To characterize the optimal k, look for the steady stateI xt+1 = Tk̄xt (a Markov process with 2k + 1 states)
Tk=
1− p(1− p) p(1− p) 0 0p(1− p) p2+(1− p)2
0 ..
p2+(1− p)2 p(1− p)0 p(1− p) 1− p(1− p)
I This process is ergodic =⇒ There is a unique steady statedistribution, which is the uniform distribution over states =⇒each state has probability 1
2k+1
![Page 70: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/70.jpg)
Finding the Optimal Mechanism
I To characterize the optimal k, look for the steady stateI xt+1 = Tk̄xt (a Markov process with 2k + 1 states)
Tk=
1− p(1− p) p(1− p) 0 0p(1− p) p2+(1− p)2
0 ..
p2+(1− p)2 p(1− p)0 p(1− p) 1− p(1− p)
I This process is ergodic =⇒ There is a unique steady statedistribution, which is the uniform distribution over states =⇒each state has probability 1
2k+1
![Page 71: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/71.jpg)
Optimal Mechanism —Costs
I We use the steady state distribution to identify costs andbenefits for any k̄
I In state r , 2 |r | participants are present on the market
I Expected total waiting costs:
C (k) =1
2k + 1
(k
∑l=−k
2 |r |)c =
k(k + 1)c
2k + 1
![Page 72: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/72.jpg)
Optimal Mechanism —Costs
I We use the steady state distribution to identify costs andbenefits for any k̄
I In state r , 2 |r | participants are present on the market
I Expected total waiting costs:
C (k) =1
2k + 1
(k
∑l=−k
2 |r |)c =
k(k + 1)c
2k + 1
![Page 73: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/73.jpg)
Optimal Mechanism —Benefits
I Suppose an (A, β) pair has arrived at the market (similarly fora (B, α) pair) when the queue is r :
I If −k ≤ r < 0, the optimal mechanism creates one (A, α)match and one (B, β) match → UAα + UB β
I If 0 ≤ r < k, the mechanism creates no matches → 0I If r = k, the mechanism creates an (A, β) match → UAβ
I Expected per-period total match surplus (after some algebra):
B(k) = pUAα + (1− p)UBβ −p(1− p)U2k + 1
![Page 74: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/74.jpg)
Optimal Mechanism —Benefits
I Suppose an (A, β) pair has arrived at the market (similarly fora (B, α) pair) when the queue is r :
I If −k ≤ r < 0, the optimal mechanism creates one (A, α)match and one (B, β) match → UAα + UB β
I If 0 ≤ r < k, the mechanism creates no matches → 0I If r = k, the mechanism creates an (A, β) match → UAβ
I Expected per-period total match surplus (after some algebra):
B(k) = pUAα + (1− p)UBβ −p(1− p)U2k + 1
![Page 75: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/75.jpg)
Optimal Mechanism —Solution
I Optimal k maximizes B(k)− C (k)
Proposition 2 The threshold
k̄opt =
⌊√p(1− p)U
2c
⌋
identifies an optimal dynamic mechanism, and it isgenerically unique.
I kopt balances market thickness and cost of waiting
I koptdecreases in c and positive only when c ≤ p(1−p)U
2
I koptincreases as p(1− p) (probability of arrival of a
mismatched pair) increases, and maximized at p = 1/2
![Page 76: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/76.jpg)
Optimal Mechanism —Solution
I Optimal k maximizes B(k)− C (k)Proposition 2 The threshold
k̄opt =
⌊√p(1− p)U
2c
⌋
identifies an optimal dynamic mechanism, and it isgenerically unique.
I kopt balances market thickness and cost of waiting
I koptdecreases in c and positive only when c ≤ p(1−p)U
2
I koptincreases as p(1− p) (probability of arrival of a
mismatched pair) increases, and maximized at p = 1/2
![Page 77: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/77.jpg)
Optimal Mechanism —Solution
I Optimal k maximizes B(k)− C (k)Proposition 2 The threshold
k̄opt =
⌊√p(1− p)U
2c
⌋
identifies an optimal dynamic mechanism, and it isgenerically unique.
I kopt balances market thickness and cost of waiting
I koptdecreases in c and positive only when c ≤ p(1−p)U
2
I koptincreases as p(1− p) (probability of arrival of a
mismatched pair) increases, and maximized at p = 1/2
![Page 78: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/78.jpg)
Optimal Mechanism —Solution
I Optimal k maximizes B(k)− C (k)Proposition 2 The threshold
k̄opt =
⌊√p(1− p)U
2c
⌋
identifies an optimal dynamic mechanism, and it isgenerically unique.
I kopt balances market thickness and cost of waiting
I koptdecreases in c and positive only when c ≤ p(1−p)U
2
I koptincreases as p(1− p) (probability of arrival of a
mismatched pair) increases, and maximized at p = 1/2
![Page 79: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/79.jpg)
Optimal Mechanism —Welfare
I Welfare = Match Surplus - Waiting Costs
W opt (c) = pUAα+(1−p)UBβ−p(1− p)U2kopt+ 1− k
opt(kopt+ 1)c
2kopt+ 1
I From super-modularity, maximal conceivable welfare (absentcosts):
S∞ = pUAα + (1− p)UBβ
![Page 80: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/80.jpg)
Optimal Mechanism —Welfare
I Welfare = Match Surplus - Waiting Costs
W opt (c) = pUAα+(1−p)UBβ−p(1− p)U2kopt+ 1− k
opt(kopt+ 1)c
2kopt+ 1
I From super-modularity, maximal conceivable welfare (absentcosts):
S∞ = pUAα + (1− p)UBβ
![Page 81: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/81.jpg)
Optimal Mechanism —Welfare
I For c1 > c2, can emulate the mechanism with c1 under c2I Same matching surplus, lower costsI =⇒ Optimal welfare decreases in c
I koptdecreases in c
I As c increases, fewer individuals wait and effect of marginalcost increase is smaller
I =⇒ Welfare loss is concave in c
![Page 82: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/82.jpg)
Optimal Mechanism —Welfare
I For c1 > c2, can emulate the mechanism with c1 under c2I Same matching surplus, lower costsI =⇒ Optimal welfare decreases in c
I koptdecreases in c
I As c increases, fewer individuals wait and effect of marginalcost increase is smaller
I =⇒ Welfare loss is concave in c
![Page 83: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/83.jpg)
Optimal Mechanism —Welfare
Corollary (Optimal Welfare)The welfare under the optimal mechanism is given byW opt (c) = S∞ −Θ(c), where Θ(c) is continuous, increasing, andconcave in c, lim
c→0Θ(c) = 0, and Θ(c) = p(1− p)U for all
c ≥ p(1−p)U2 .
=⇒ For small waiting costs, the optimal mechanism achievesapproximately the maximal conceivable welfare
![Page 84: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/84.jpg)
Optimal Mechanism —Welfare
Corollary (Optimal Welfare)The welfare under the optimal mechanism is given byW opt (c) = S∞ −Θ(c), where Θ(c) is continuous, increasing, andconcave in c, lim
c→0Θ(c) = 0, and Θ(c) = p(1− p)U for all
c ≥ p(1−p)U2 .
=⇒ For small waiting costs, the optimal mechanism achievesapproximately the maximal conceivable welfare
![Page 85: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/85.jpg)
Decentralized Markets
I Many dynamic markets are decentralized (e.g., adoption inthe U.S.)
I Identify the performance of decentralized processes in order tounderstand when centralized intervention might be beneficial
![Page 86: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/86.jpg)
Decentralized Markets
I Many dynamic markets are decentralized (e.g., adoption inthe U.S.)
I Identify the performance of decentralized processes in order tounderstand when centralized intervention might be beneficial
![Page 87: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/87.jpg)
A Stark Model of Decentralized Dynamic Markets
I We assume that at each period:I A square-round pair enters the market (type distribution asbefore)
I Each square and round on the market declare their demand(e.g., for a round, whether she will match only with anA-square, or with either A- or B-squares)
I Given participants’demands, agents are matched in order ofarrival (first in first out)
I For simplicity, we assume symmetric preferences:
UA(α)− UA(β) = Uα(A)− Uα(B), and
UB (α)− UB (β) = Uβ(A)− Uβ(B)
I Focus on trembling-hand equilibrium
![Page 88: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/88.jpg)
A Stark Model of Decentralized Dynamic Markets
I We assume that at each period:I A square-round pair enters the market (type distribution asbefore)
I Each square and round on the market declare their demand(e.g., for a round, whether she will match only with anA-square, or with either A- or B-squares)
I Given participants’demands, agents are matched in order ofarrival (first in first out)
I For simplicity, we assume symmetric preferences:
UA(α)− UA(β) = Uα(A)− Uα(B), and
UB (α)− UB (β) = Uβ(A)− Uβ(B)
I Focus on trembling-hand equilibrium
![Page 89: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/89.jpg)
A Stark Model of Decentralized Dynamic Markets
I We assume that at each period:I A square-round pair enters the market (type distribution asbefore)
I Each square and round on the market declare their demand(e.g., for a round, whether she will match only with anA-square, or with either A- or B-squares)
I Given participants’demands, agents are matched in order ofarrival (first in first out)
I For simplicity, we assume symmetric preferences:
UA(α)− UA(β) = Uα(A)− Uα(B), and
UB (α)− UB (β) = Uβ(A)− Uβ(B)
I Focus on trembling-hand equilibrium
![Page 90: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/90.jpg)
Example 1
![Page 91: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/91.jpg)
Example 1
![Page 92: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/92.jpg)
Example 1
![Page 93: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/93.jpg)
Example 1
![Page 94: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/94.jpg)
Example 2
![Page 95: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/95.jpg)
Example 2
![Page 96: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/96.jpg)
Example 2 —First Option
![Page 97: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/97.jpg)
Example 2 —Second Option
![Page 98: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/98.jpg)
Example 2 —Second Option
![Page 99: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/99.jpg)
Equilibrium Characterization (1)
I There can never be an A-square and an α-round waiting in themarket (would prefer to match immediately)
I Consider the decision of an A-square entering the marketI When to wait when there is a queue of A-squares ahead?I The time till an α-round arrives is distributed geometrically:
I When first in queue → time to wait for an α-round is 1/pI When second in queue → time to wait for an α-round is 2/pI ... When k-th in queue → time to wait for an α-round is k/p
I Cost of waiting when k-th in line: kcpI Benefit of waiting: UA(α)− UA(β) (note that a β-round isalways willing to match)
![Page 100: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/100.jpg)
Equilibrium Characterization (1)
I There can never be an A-square and an α-round waiting in themarket (would prefer to match immediately)
I Consider the decision of an A-square entering the market
I When to wait when there is a queue of A-squares ahead?I The time till an α-round arrives is distributed geometrically:
I When first in queue → time to wait for an α-round is 1/pI When second in queue → time to wait for an α-round is 2/pI ... When k-th in queue → time to wait for an α-round is k/p
I Cost of waiting when k-th in line: kcpI Benefit of waiting: UA(α)− UA(β) (note that a β-round isalways willing to match)
![Page 101: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/101.jpg)
Equilibrium Characterization (1)
I There can never be an A-square and an α-round waiting in themarket (would prefer to match immediately)
I Consider the decision of an A-square entering the marketI When to wait when there is a queue of A-squares ahead?I The time till an α-round arrives is distributed geometrically:
I When first in queue → time to wait for an α-round is 1/pI When second in queue → time to wait for an α-round is 2/pI ... When k-th in queue → time to wait for an α-round is k/p
I Cost of waiting when k-th in line: kcpI Benefit of waiting: UA(α)− UA(β) (note that a β-round isalways willing to match)
![Page 102: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/102.jpg)
Equilibrium Characterization (1)
I There can never be an A-square and an α-round waiting in themarket (would prefer to match immediately)
I Consider the decision of an A-square entering the marketI When to wait when there is a queue of A-squares ahead?I The time till an α-round arrives is distributed geometrically:
I When first in queue → time to wait for an α-round is 1/pI When second in queue → time to wait for an α-round is 2/pI ... When k-th in queue → time to wait for an α-round is k/p
I Cost of waiting when k-th in line: kcpI Benefit of waiting: UA(α)− UA(β) (note that a β-round isalways willing to match)
![Page 103: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/103.jpg)
Equilibrium Characterization (1)
I There can never be an A-square and an α-round waiting in themarket (would prefer to match immediately)
I Consider the decision of an A-square entering the marketI When to wait when there is a queue of A-squares ahead?I The time till an α-round arrives is distributed geometrically:
I When first in queue → time to wait for an α-round is 1/pI When second in queue → time to wait for an α-round is 2/pI ... When k-th in queue → time to wait for an α-round is k/p
I Cost of waiting when k-th in line: kcpI Benefit of waiting: UA(α)− UA(β) (note that a β-round isalways willing to match)
![Page 104: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/104.jpg)
Equilibrium Characterization (2)
We get bounds on kAα = kA − kα:
Lemma 2 −kdec ≤ kAα ≤ kdec, where
kdec ≡ max
{k ∈ Z+ |
kcp< UA(α)− UA(β)
}= max
{k ∈ Z+ |
kcp< Uα(A)− Uα(B)
}.
I Regular environment: Agents are never indifferent betweenwaiting and leaving
I Note that if an A-square decides to wait for a superior match,she will wait in subsequent periods as well, as her position canonly improve; Similarly for an α-round
![Page 105: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/105.jpg)
Equilibrium Characterization (2)
We get bounds on kAα = kA − kα:
Lemma 2 −kdec ≤ kAα ≤ kdec, where
kdec ≡ max
{k ∈ Z+ |
kcp< UA(α)− UA(β)
}= max
{k ∈ Z+ |
kcp< Uα(A)− Uα(B)
}.
I Regular environment: Agents are never indifferent betweenwaiting and leaving
I Note that if an A-square decides to wait for a superior match,she will wait in subsequent periods as well, as her position canonly improve; Similarly for an α-round
![Page 106: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/106.jpg)
Equilibrium Characterization (2)
We get bounds on kAα = kA − kα:
Lemma 2 −kdec ≤ kAα ≤ kdec, where
kdec ≡ max
{k ∈ Z+ |
kcp< UA(α)− UA(β)
}= max
{k ∈ Z+ |
kcp< Uα(A)− Uα(B)
}.
I Regular environment: Agents are never indifferent betweenwaiting and leaving
I Note that if an A-square decides to wait for a superior match,she will wait in subsequent periods as well, as her position canonly improve; Similarly for an α-round
![Page 107: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/107.jpg)
Equilibrium Characterization (3)
I Consider the decision of β-rounds to wait in lineI A β-round may decide to wait for an A-square, who becomesavailable for β-rounds every time the A− α queue exceeds
kdec
I Two conflicting forces:I the longer the line of β-rounds, the weaker the incentive towait
I the closer kAα is to the threshold, the stronger the incentive towait
![Page 108: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/108.jpg)
Equilibrium Characterization (3)
I Consider the decision of β-rounds to wait in lineI A β-round may decide to wait for an A-square, who becomesavailable for β-rounds every time the A− α queue exceeds
kdec
I Two conflicting forces:I the longer the line of β-rounds, the weaker the incentive towait
I the closer kAα is to the threshold, the stronger the incentive towait
![Page 109: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/109.jpg)
Equilibrium Characterization (4)
There can never be a B-square and a β-round waiting in the market
Intuition:
I Consider the first β-round arriving at the marketI No previous A-squares waiting since such squares would havearrived with β-rounds
I Suppose the β-round arrives with a B-squareI To match with an A-square, she needs to wait for at leastkdec+ 1 of A-squares to arrive (first k
decwait in queue)
![Page 110: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/110.jpg)
Equilibrium Characterization (4)
There can never be a B-square and a β-round waiting in the market
Intuition:
I Consider the first β-round arriving at the market
I No previous A-squares waiting since such squares would havearrived with β-rounds
I Suppose the β-round arrives with a B-squareI To match with an A-square, she needs to wait for at leastkdec+ 1 of A-squares to arrive (first k
decwait in queue)
![Page 111: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/111.jpg)
Equilibrium Characterization (4)
There can never be a B-square and a β-round waiting in the market
Intuition:
I Consider the first β-round arriving at the marketI No previous A-squares waiting since such squares would havearrived with β-rounds
I Suppose the β-round arrives with a B-squareI To match with an A-square, she needs to wait for at leastkdec+ 1 of A-squares to arrive (first k
decwait in queue)
![Page 112: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/112.jpg)
Equilibrium Characterization (4)
There can never be a B-square and a β-round waiting in the market
Intuition:
I Consider the first β-round arriving at the marketI No previous A-squares waiting since such squares would havearrived with β-rounds
I Suppose the β-round arrives with a B-square
I To match with an A-square, she needs to wait for at leastkdec+ 1 of A-squares to arrive (first k
decwait in queue)
![Page 113: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/113.jpg)
Equilibrium Characterization (4)
There can never be a B-square and a β-round waiting in the market
Intuition:
I Consider the first β-round arriving at the marketI No previous A-squares waiting since such squares would havearrived with β-rounds
I Suppose the β-round arrives with a B-squareI To match with an A-square, she needs to wait for at leastkdec+ 1 of A-squares to arrive (first k
decwait in queue)
![Page 114: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/114.jpg)
Equilibrium Characterization (5)
Intuition (continued):
I Cost of waiting: at least (kdec + 1)c/p, greater than themaximal waiting cost k
decc/p that A-squares experience
I Benefit of waiting: (from super-modularity)
Uβ(A)− Uβ(B) < Uα(A)− Uα(B) = UA(α)− UA(β)
I Since A-squares are not willing to wait when (kdec+ 1)-th in
line, nor would the β-roundI β-rounds who arrive later in the process face similar calculus(Lemma 3)
![Page 115: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/115.jpg)
Equilibrium Characterization (5)
Intuition (continued):
I Cost of waiting: at least (kdec + 1)c/p, greater than themaximal waiting cost k
decc/p that A-squares experience
I Benefit of waiting: (from super-modularity)
Uβ(A)− Uβ(B) < Uα(A)− Uα(B) = UA(α)− UA(β)
I Since A-squares are not willing to wait when (kdec+ 1)-th in
line, nor would the β-roundI β-rounds who arrive later in the process face similar calculus(Lemma 3)
![Page 116: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/116.jpg)
Equilibrium Characterization (5)
Intuition (continued):
I Cost of waiting: at least (kdec + 1)c/p, greater than themaximal waiting cost k
decc/p that A-squares experience
I Benefit of waiting: (from super-modularity)
Uβ(A)− Uβ(B) < Uα(A)− Uα(B) = UA(α)− UA(β)
I Since A-squares are not willing to wait when (kdec+ 1)-th in
line, nor would the β-round
I β-rounds who arrive later in the process face similar calculus(Lemma 3)
![Page 117: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/117.jpg)
Equilibrium Characterization (5)
Intuition (continued):
I Cost of waiting: at least (kdec + 1)c/p, greater than themaximal waiting cost k
decc/p that A-squares experience
I Benefit of waiting: (from super-modularity)
Uβ(A)− Uβ(B) < Uα(A)− Uα(B) = UA(α)− UA(β)
I Since A-squares are not willing to wait when (kdec+ 1)-th in
line, nor would the β-roundI β-rounds who arrive later in the process face similar calculus(Lemma 3)
![Page 118: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/118.jpg)
Equilibrium Characterization —Summary (Proposition 3)
I (A, α) or (B, β) pairs match immediatelyI There is a stock of (A, β) or (B, α) pairs up to a threshold
k̄dec =⌊p (UA(α)− UA(β))
c
⌋=
⌊p (Uα(A)− Uα(B))
c
⌋
I =⇒ Same protocol as that of the optimal mechanism, butwith a threshold k̄dec
I Unique steady state distribution, which is uniform over valuesof kAα = −k̄dec , ..., k̄dec
![Page 119: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/119.jpg)
Equilibrium Characterization —Summary (Proposition 3)
I (A, α) or (B, β) pairs match immediatelyI There is a stock of (A, β) or (B, α) pairs up to a threshold
k̄dec =⌊p (UA(α)− UA(β))
c
⌋=
⌊p (Uα(A)− Uα(B))
c
⌋
I =⇒ Same protocol as that of the optimal mechanism, butwith a threshold k̄dec
I Unique steady state distribution, which is uniform over valuesof kAα = −k̄dec , ..., k̄dec
![Page 120: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/120.jpg)
Equilibrium Characterization —Summary (Proposition 3)
I (A, α) or (B, β) pairs match immediatelyI There is a stock of (A, β) or (B, α) pairs up to a threshold
k̄dec =⌊p (UA(α)− UA(β))
c
⌋=
⌊p (Uα(A)− Uα(B))
c
⌋
I =⇒ Same protocol as that of the optimal mechanism, butwith a threshold k̄dec
I Unique steady state distribution, which is uniform over valuesof kAα = −k̄dec , ..., k̄dec
![Page 121: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/121.jpg)
Comparing Optimal and Decentralized Processes
I There is a negative externality of waiting — imposes longerwaits (and possibly inferior matches) on those that follow,which dominates the positive externality of waiting
I =⇒ Excessive waiting in decentralized markets
I Formally: k̄dec ≥ k̄opt (Corollary 2)
![Page 122: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/122.jpg)
Comparing Optimal and Decentralized Processes
I There is a negative externality of waiting — imposes longerwaits (and possibly inferior matches) on those that follow,which dominates the positive externality of waiting
I =⇒ Excessive waiting in decentralized markets
I Formally: k̄dec ≥ k̄opt (Corollary 2)
![Page 123: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/123.jpg)
Welfare in a Decentralized Market
I Analogous calculation to that for the optimal mechanism(though steady state distribution is uniform on a widersupport)
I Recall:S∞ = pUAα + (1− p)UBβ
I W opt (c) = S∞ −Θ(c) with limc→0 Θ(c) = 0 (Corollary 1)
Corollary (Decentralized Welfare)The welfare under the decentralized process is given byW dec (c) = S∞ −Ψ(c), where lim
c→0Ψ(c) = p (UA(α)− UA(β)) ,
and Ψ(c) = p(1− p)U for all c ≥ p (UA(α)− UA(β)) .
![Page 124: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/124.jpg)
Welfare in a Decentralized Market
I Analogous calculation to that for the optimal mechanism(though steady state distribution is uniform on a widersupport)
I Recall:S∞ = pUAα + (1− p)UBβ
I W opt (c) = S∞ −Θ(c) with limc→0 Θ(c) = 0 (Corollary 1)
Corollary (Decentralized Welfare)The welfare under the decentralized process is given byW dec (c) = S∞ −Ψ(c), where lim
c→0Ψ(c) = p (UA(α)− UA(β)) ,
and Ψ(c) = p(1− p)U for all c ≥ p (UA(α)− UA(β)) .
![Page 125: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/125.jpg)
Welfare in a Decentralized Market
I Analogous calculation to that for the optimal mechanism(though steady state distribution is uniform on a widersupport)
I Recall:S∞ = pUAα + (1− p)UBβ
I W opt (c) = S∞ −Θ(c) with limc→0 Θ(c) = 0 (Corollary 1)
Corollary (Decentralized Welfare)The welfare under the decentralized process is given byW dec (c) = S∞ −Ψ(c), where lim
c→0Ψ(c) = p (UA(α)− UA(β)) ,
and Ψ(c) = p(1− p)U for all c ≥ p (UA(α)− UA(β)) .
![Page 126: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/126.jpg)
Welfare Comparison
I By definition, W opt (c) ≥ W dec (c)
I When might centralization be particularly beneficial?I What are the comparative statics of the welfare wedgeW opt (c)−W dec (c) with respect to c , p, and U?
![Page 127: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/127.jpg)
Welfare Comparison —Waiting Costs
I An increase in costs has two effects on the welfare gap:I Direct Effect: Since k̄dec ≥ k̄opt , greater costs → greaterwelfare gap
I Indirect Effect: Greater costs lead to smaller thresholds, andthe mechanisms are more similar → smaller welfare gap
I Points of discontinuity: costs at which the decentralizedthreshold decreases
I As costs become prohibitively high, both processes lead toinstantaneous matches and identical welfare levels
![Page 128: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/128.jpg)
Welfare Comparison —Waiting Costs
I An increase in costs has two effects on the welfare gap:I Direct Effect: Since k̄dec ≥ k̄opt , greater costs → greaterwelfare gap
I Indirect Effect: Greater costs lead to smaller thresholds, andthe mechanisms are more similar → smaller welfare gap
I Points of discontinuity: costs at which the decentralizedthreshold decreases
I As costs become prohibitively high, both processes lead toinstantaneous matches and identical welfare levels
![Page 129: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/129.jpg)
Welfare Comparison —Waiting Costs
I An increase in costs has two effects on the welfare gap:I Direct Effect: Since k̄dec ≥ k̄opt , greater costs → greaterwelfare gap
I Indirect Effect: Greater costs lead to smaller thresholds, andthe mechanisms are more similar → smaller welfare gap
I Points of discontinuity: costs at which the decentralizedthreshold decreases
I As costs become prohibitively high, both processes lead toinstantaneous matches and identical welfare levels
![Page 130: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/130.jpg)
Welfare Comparison —Waiting Costs
![Page 131: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/131.jpg)
Welfare Comparison —Type Distribution and Preferences
Proposition 4
1. For any interval [c, c), where c > 0, there is a partition{[ci , ci+1)}M−1i=1 , where c = c1 < c2 < ... < cM = c , suchthat W opt (c)−W dec (c) is continuous and increasing over(ci , ci+1) and
W opt (ci )−W dec (ci ) > W opt (ci+1)−W dec (ci+1)
for all i = 1, ...,M − 1.
2. As c becomes vanishingly small, the welfare gapW opt (c)−W dec (c) converges to a value that is increasing inp ∈ [0, 1) and in Uα(A)− Uα(B).
![Page 132: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/132.jpg)
Welfare Comparison —Type Distribution and Preferences
Proposition 4
1. For any interval [c, c), where c > 0, there is a partition{[ci , ci+1)}M−1i=1 , where c = c1 < c2 < ... < cM = c , suchthat W opt (c)−W dec (c) is continuous and increasing over(ci , ci+1) and
W opt (ci )−W dec (ci ) > W opt (ci+1)−W dec (ci+1)
for all i = 1, ...,M − 1.2. As c becomes vanishingly small, the welfare gapW opt (c)−W dec (c) converges to a value that is increasing inp ∈ [0, 1) and in Uα(A)− Uα(B).
![Page 133: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/133.jpg)
Welfare Comparison —Type Distribution
Intuition:
I Consider p1 and p2 such that p1 < p2 = mp1, m > 1I Suppose costs are very low → match surplus ≈ S∞ under p1
I Individual incentives to wait are higher under p2 than under p1I In the decentralized setting, under p2, roughly 1− 1/m of thesteady-state probability mass allocated to queue lengths largerthan those seen under p1
I Large steady state queue lengths → increased waiting costsI Marginal effect on match surplus minuscule
I In contrast, the optimal mechanism internalizes externalities→ impacts of type distribution on induced welfare levels areweak for low c
![Page 134: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/134.jpg)
Simple Interventions
The optimal mechanism may be diffi cult to implement in certainenvironments:
I Imposes matches on participants
I Requires continuous monitoring of the market
![Page 135: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/135.jpg)
Simple Interventions
The optimal mechanism may be diffi cult to implement in certainenvironments:
I Imposes matches on participantsI Linear balanced-budget tax schemes that lead to optimalwelfare
I Requires continuous monitoring of the marketI Fixed-window mechanisms potentially provide a substantialimprovement over decentralized ones
![Page 136: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/136.jpg)
Extensions
I More than two types
I Independent arrivals
I Incomplete information
I Transfers
I Non-linear waiting costs
![Page 137: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/137.jpg)
THE END
![Page 138: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/138.jpg)
Individual Rationality
I Assume all agents are acceptable
I Since k̄dec ≥ k̄opt , no suffi ces to show individual rationality fordecentralized processes
I A-squares (similarly, α-rounds) can always match withβ-rounds (B-squares) immediately → IR satisfied for them
I Consider a B-square who enters as k-th in line, k ≤ k̄decI Suppose the B-square is willing to match with any round inall periods
I =⇒ Expected time to match with a β-round is at mostk1−p ≤
k̄dec1−p
I =⇒ Expected utility is at least
UB (β)−k̄decc1− p ≥ UB (β)−
p1− p [UA(α)− UA(β)] ≡ U
min
I As long as utility from remaining unmatched < Umin, IRsatisfied for B-squares
![Page 139: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/139.jpg)
Individual Rationality
I Assume all agents are acceptableI Since k̄dec ≥ k̄opt , no suffi ces to show individual rationality fordecentralized processes
I A-squares (similarly, α-rounds) can always match withβ-rounds (B-squares) immediately → IR satisfied for them
I Consider a B-square who enters as k-th in line, k ≤ k̄decI Suppose the B-square is willing to match with any round inall periods
I =⇒ Expected time to match with a β-round is at mostk1−p ≤
k̄dec1−p
I =⇒ Expected utility is at least
UB (β)−k̄decc1− p ≥ UB (β)−
p1− p [UA(α)− UA(β)] ≡ U
min
I As long as utility from remaining unmatched < Umin, IRsatisfied for B-squares
![Page 140: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/140.jpg)
Individual Rationality
I Assume all agents are acceptableI Since k̄dec ≥ k̄opt , no suffi ces to show individual rationality fordecentralized processes
I A-squares (similarly, α-rounds) can always match withβ-rounds (B-squares) immediately → IR satisfied for them
I Consider a B-square who enters as k-th in line, k ≤ k̄decI Suppose the B-square is willing to match with any round inall periods
I =⇒ Expected time to match with a β-round is at mostk1−p ≤
k̄dec1−p
I =⇒ Expected utility is at least
UB (β)−k̄decc1− p ≥ UB (β)−
p1− p [UA(α)− UA(β)] ≡ U
min
I As long as utility from remaining unmatched < Umin, IRsatisfied for B-squares
![Page 141: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/141.jpg)
Individual Rationality
I Assume all agents are acceptableI Since k̄dec ≥ k̄opt , no suffi ces to show individual rationality fordecentralized processes
I A-squares (similarly, α-rounds) can always match withβ-rounds (B-squares) immediately → IR satisfied for them
I Consider a B-square who enters as k-th in line, k ≤ k̄dec
I Suppose the B-square is willing to match with any round inall periods
I =⇒ Expected time to match with a β-round is at mostk1−p ≤
k̄dec1−p
I =⇒ Expected utility is at least
UB (β)−k̄decc1− p ≥ UB (β)−
p1− p [UA(α)− UA(β)] ≡ U
min
I As long as utility from remaining unmatched < Umin, IRsatisfied for B-squares
![Page 142: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/142.jpg)
Individual Rationality
I Assume all agents are acceptableI Since k̄dec ≥ k̄opt , no suffi ces to show individual rationality fordecentralized processes
I A-squares (similarly, α-rounds) can always match withβ-rounds (B-squares) immediately → IR satisfied for them
I Consider a B-square who enters as k-th in line, k ≤ k̄decI Suppose the B-square is willing to match with any round inall periods
I =⇒ Expected time to match with a β-round is at mostk1−p ≤
k̄dec1−p
I =⇒ Expected utility is at least
UB (β)−k̄decc1− p ≥ UB (β)−
p1− p [UA(α)− UA(β)] ≡ U
min
I As long as utility from remaining unmatched < Umin, IRsatisfied for B-squares
![Page 143: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/143.jpg)
Individual Rationality
I Assume all agents are acceptableI Since k̄dec ≥ k̄opt , no suffi ces to show individual rationality fordecentralized processes
I A-squares (similarly, α-rounds) can always match withβ-rounds (B-squares) immediately → IR satisfied for them
I Consider a B-square who enters as k-th in line, k ≤ k̄decI Suppose the B-square is willing to match with any round inall periods
I =⇒ Expected time to match with a β-round is at mostk1−p ≤
k̄dec1−p
I =⇒ Expected utility is at least
UB (β)−k̄decc1− p ≥ UB (β)−
p1− p [UA(α)− UA(β)] ≡ U
min
I As long as utility from remaining unmatched < Umin, IRsatisfied for B-squares
![Page 144: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/144.jpg)
Individual Rationality
I Assume all agents are acceptableI Since k̄dec ≥ k̄opt , no suffi ces to show individual rationality fordecentralized processes
I A-squares (similarly, α-rounds) can always match withβ-rounds (B-squares) immediately → IR satisfied for them
I Consider a B-square who enters as k-th in line, k ≤ k̄decI Suppose the B-square is willing to match with any round inall periods
I =⇒ Expected time to match with a β-round is at mostk1−p ≤
k̄dec1−p
I =⇒ Expected utility is at least
UB (β)−k̄decc1− p ≥ UB (β)−
p1− p [UA(α)− UA(β)] ≡ U
min
I As long as utility from remaining unmatched < Umin, IRsatisfied for B-squares
![Page 145: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/145.jpg)
Individual Rationality
I Assume all agents are acceptableI Since k̄dec ≥ k̄opt , no suffi ces to show individual rationality fordecentralized processes
I A-squares (similarly, α-rounds) can always match withβ-rounds (B-squares) immediately → IR satisfied for them
I Consider a B-square who enters as k-th in line, k ≤ k̄decI Suppose the B-square is willing to match with any round inall periods
I =⇒ Expected time to match with a β-round is at mostk1−p ≤
k̄dec1−p
I =⇒ Expected utility is at least
UB (β)−k̄decc1− p ≥ UB (β)−
p1− p [UA(α)− UA(β)] ≡ U
min
I As long as utility from remaining unmatched < Umin, IRsatisfied for B-squares
![Page 146: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/146.jpg)
Sub-modular Preferences
I Suppose sub-modular preferences
U ≡ UAα + UBβ − UAβ − UBα < 0
=⇒ The effi cient matching entails the maximal number of (A, β)and (B, α) pairs
I Characterization of optimal mechanism and fixed-windowmechanisms similar
I Effi ciency losses in decentralized processes due to two effects:I Negative externalities of waitingI Private match incentives misaligned with market-wide ones
=⇒ Greater welfare wedge between decentralized and optimal orfixed-window protocols under sub-modularity
![Page 147: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/147.jpg)
Sub-modular Preferences
I Suppose sub-modular preferences
U ≡ UAα + UBβ − UAβ − UBα < 0
=⇒ The effi cient matching entails the maximal number of (A, β)and (B, α) pairs
I Characterization of optimal mechanism and fixed-windowmechanisms similar
I Effi ciency losses in decentralized processes due to two effects:I Negative externalities of waitingI Private match incentives misaligned with market-wide ones
=⇒ Greater welfare wedge between decentralized and optimal orfixed-window protocols under sub-modularity
![Page 148: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/148.jpg)
Sub-modular Preferences
I Suppose sub-modular preferences
U ≡ UAα + UBβ − UAβ − UBα < 0
=⇒ The effi cient matching entails the maximal number of (A, β)and (B, α) pairs
I Characterization of optimal mechanism and fixed-windowmechanisms similar
I Effi ciency losses in decentralized processes due to two effects:I Negative externalities of waitingI Private match incentives misaligned with market-wide ones
=⇒ Greater welfare wedge between decentralized and optimal orfixed-window protocols under sub-modularity
![Page 149: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/149.jpg)
Sub-modular Preferences
I Suppose sub-modular preferences
U ≡ UAα + UBβ − UAβ − UBα < 0
=⇒ The effi cient matching entails the maximal number of (A, β)and (B, α) pairs
I Characterization of optimal mechanism and fixed-windowmechanisms similar
I Effi ciency losses in decentralized processes due to two effects:I Negative externalities of waitingI Private match incentives misaligned with market-wide ones
=⇒ Greater welfare wedge between decentralized and optimal orfixed-window protocols under sub-modularity
![Page 150: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/150.jpg)
Sub-modular Preferences
I Suppose sub-modular preferences
U ≡ UAα + UBβ − UAβ − UBα < 0
=⇒ The effi cient matching entails the maximal number of (A, β)and (B, α) pairs
I Characterization of optimal mechanism and fixed-windowmechanisms similar
I Effi ciency losses in decentralized processes due to two effects:I Negative externalities of waitingI Private match incentives misaligned with market-wide ones
=⇒ Greater welfare wedge between decentralized and optimal orfixed-window protocols under sub-modularity
![Page 151: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/151.jpg)
Asymmetric Costs
I Suppose squares experience waiting costs cS and roundsexperience waiting costs cR
I Assume (cS + cR )/2 = c (mean-preserving spread)I Per-pair costs of waiting remain the same → optimal andfixed-window mechanisms identical
I In decentralized settings, A-squares willing to wait k̄decS andα-rounds willing to wait k̄decR ,
k̄decS =
⌊p (UA(α)− UA(β))
cS
⌋k̄decR =
⌊p (Uα(A)− Uα(B))
cR
⌋
I From convexity, k̄dec ≤ (k̄decS + k̄decR )/2 → waiting indecentralized processes is longer
I =⇒ Greater welfare gap between decentralized and optimal orfixed-window protocols under asymmetric costs
![Page 152: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/152.jpg)
Asymmetric Costs
I Suppose squares experience waiting costs cS and roundsexperience waiting costs cR
I Assume (cS + cR )/2 = c (mean-preserving spread)
I Per-pair costs of waiting remain the same → optimal andfixed-window mechanisms identical
I In decentralized settings, A-squares willing to wait k̄decS andα-rounds willing to wait k̄decR ,
k̄decS =
⌊p (UA(α)− UA(β))
cS
⌋k̄decR =
⌊p (Uα(A)− Uα(B))
cR
⌋
I From convexity, k̄dec ≤ (k̄decS + k̄decR )/2 → waiting indecentralized processes is longer
I =⇒ Greater welfare gap between decentralized and optimal orfixed-window protocols under asymmetric costs
![Page 153: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/153.jpg)
Asymmetric Costs
I Suppose squares experience waiting costs cS and roundsexperience waiting costs cR
I Assume (cS + cR )/2 = c (mean-preserving spread)I Per-pair costs of waiting remain the same → optimal andfixed-window mechanisms identical
I In decentralized settings, A-squares willing to wait k̄decS andα-rounds willing to wait k̄decR ,
k̄decS =
⌊p (UA(α)− UA(β))
cS
⌋k̄decR =
⌊p (Uα(A)− Uα(B))
cR
⌋
I From convexity, k̄dec ≤ (k̄decS + k̄decR )/2 → waiting indecentralized processes is longer
I =⇒ Greater welfare gap between decentralized and optimal orfixed-window protocols under asymmetric costs
![Page 154: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/154.jpg)
Asymmetric Costs
I Suppose squares experience waiting costs cS and roundsexperience waiting costs cR
I Assume (cS + cR )/2 = c (mean-preserving spread)I Per-pair costs of waiting remain the same → optimal andfixed-window mechanisms identical
I In decentralized settings, A-squares willing to wait k̄decS andα-rounds willing to wait k̄decR ,
k̄decS =
⌊p (UA(α)− UA(β))
cS
⌋k̄decR =
⌊p (Uα(A)− Uα(B))
cR
⌋
I From convexity, k̄dec ≤ (k̄decS + k̄decR )/2 → waiting indecentralized processes is longer
I =⇒ Greater welfare gap between decentralized and optimal orfixed-window protocols under asymmetric costs
![Page 155: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/155.jpg)
Asymmetric Costs
I Suppose squares experience waiting costs cS and roundsexperience waiting costs cR
I Assume (cS + cR )/2 = c (mean-preserving spread)I Per-pair costs of waiting remain the same → optimal andfixed-window mechanisms identical
I In decentralized settings, A-squares willing to wait k̄decS andα-rounds willing to wait k̄decR ,
k̄decS =
⌊p (UA(α)− UA(β))
cS
⌋k̄decR =
⌊p (Uα(A)− Uα(B))
cR
⌋
I From convexity, k̄dec ≤ (k̄decS + k̄decR )/2 → waiting indecentralized processes is longer
I =⇒ Greater welfare gap between decentralized and optimal orfixed-window protocols under asymmetric costs
![Page 156: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/156.jpg)
Asymmetric Costs
I Suppose squares experience waiting costs cS and roundsexperience waiting costs cR
I Assume (cS + cR )/2 = c (mean-preserving spread)I Per-pair costs of waiting remain the same → optimal andfixed-window mechanisms identical
I In decentralized settings, A-squares willing to wait k̄decS andα-rounds willing to wait k̄decR ,
k̄decS =
⌊p (UA(α)− UA(β))
cS
⌋k̄decR =
⌊p (Uα(A)− Uα(B))
cR
⌋
I From convexity, k̄dec ≤ (k̄decS + k̄decR )/2 → waiting indecentralized processes is longer
I =⇒ Greater welfare gap between decentralized and optimal orfixed-window protocols under asymmetric costs
![Page 157: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/157.jpg)
Asymmetric Distributions
I xt+1 = Tk̄A ,k̄αxt
![Page 158: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/158.jpg)
Asymmetric Distributions
I If pα < pA, for small waiting costs, can approximate theoptimal mechanism with a one-threshold mechanism
I α-rounds wait without boundI A-squares wait until their stock reaches k̄A
I When c → 0, the optimal mechanism generates welfareapproaching S∞
![Page 159: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/159.jpg)
Simple Interventions —Tax Schemes
I Decentralized processes are ineffi cient due to excessive waiting
I =⇒ Can tax waiting in order to induce optimal waiting(without forcing matches)
I Recall:
k̄dec =⌊p (UA(α)− UA(β))
c
⌋≥ k̄opt
I Tax of τ on waiting each period =⇒ effective cost of waitingis c + τ
I Choose τ∗ such that
k̃dec =⌊p (UA(α)− UA(β))
c + τ∗
⌋= k̄opt
![Page 160: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/160.jpg)
Simple Interventions —Tax Schemes
I Decentralized processes are ineffi cient due to excessive waitingI =⇒ Can tax waiting in order to induce optimal waiting(without forcing matches)
I Recall:
k̄dec =⌊p (UA(α)− UA(β))
c
⌋≥ k̄opt
I Tax of τ on waiting each period =⇒ effective cost of waitingis c + τ
I Choose τ∗ such that
k̃dec =⌊p (UA(α)− UA(β))
c + τ∗
⌋= k̄opt
![Page 161: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/161.jpg)
Simple Interventions —Tax Schemes
I Decentralized processes are ineffi cient due to excessive waitingI =⇒ Can tax waiting in order to induce optimal waiting(without forcing matches)
I Recall:
k̄dec =⌊p (UA(α)− UA(β))
c
⌋≥ k̄opt
I Tax of τ on waiting each period =⇒ effective cost of waitingis c + τ
I Choose τ∗ such that
k̃dec =⌊p (UA(α)− UA(β))
c + τ∗
⌋= k̄opt
![Page 162: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/162.jpg)
Simple Interventions —Tax Schemes
I Decentralized processes are ineffi cient due to excessive waitingI =⇒ Can tax waiting in order to induce optimal waiting(without forcing matches)
I Recall:
k̄dec =⌊p (UA(α)− UA(β))
c
⌋≥ k̄opt
I Tax of τ on waiting each period =⇒ effective cost of waitingis c + τ
I Choose τ∗ such that
k̃dec =⌊p (UA(α)− UA(β))
c + τ∗
⌋= k̄opt
![Page 163: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/163.jpg)
Simple Interventions —Tax Schemes
I Decentralized processes are ineffi cient due to excessive waitingI =⇒ Can tax waiting in order to induce optimal waiting(without forcing matches)
I Recall:
k̄dec =⌊p (UA(α)− UA(β))
c
⌋≥ k̄opt
I Tax of τ on waiting each period =⇒ effective cost of waitingis c + τ
I Choose τ∗ such that
k̃dec =⌊p (UA(α)− UA(β))
c + τ∗
⌋= k̄opt
![Page 164: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/164.jpg)
Simple Interventions —Tax Schemes
I What do we do with collected taxes?I Could distribute to new participants entering the market toassure budget balance
I But that distorts incentives to enter the market
I Can find linear balanced-budget tax schemes that lead tooptimal welfare and have expected returns of 0 to a newentrant
![Page 165: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/165.jpg)
Simple Interventions —Tax Schemes
I What do we do with collected taxes?I Could distribute to new participants entering the market toassure budget balance
I But that distorts incentives to enter the market
I Can find linear balanced-budget tax schemes that lead tooptimal welfare and have expected returns of 0 to a newentrant
![Page 166: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/166.jpg)
Fixed-Window Mechanisms
I Fix time intervals at the end of which all squares and roundsthat arrived are matched
I “Simpler” than monitoring the market every period
I Consider a window of size n at the end of which kx agents oftype x are present
I The only effi cient (and stable) matching is:I min {kA, kα} A-squares matched with the same number of
α-roundsI max {kA, kα} −min {kA, kα} mismatched pairsI min
{kB , kβ
}B-squares matched with β-rounds
![Page 167: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/167.jpg)
Fixed-Window Mechanisms
I Fix time intervals at the end of which all squares and roundsthat arrived are matched
I “Simpler” than monitoring the market every period
I Consider a window of size n at the end of which kx agents oftype x are present
I The only effi cient (and stable) matching is:I min {kA, kα} A-squares matched with the same number of
α-roundsI max {kA, kα} −min {kA, kα} mismatched pairsI min
{kB , kβ
}B-squares matched with β-rounds
![Page 168: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/168.jpg)
Fixed-Window Mechanisms
I Fix time intervals at the end of which all squares and roundsthat arrived are matched
I “Simpler” than monitoring the market every period
I Consider a window of size n at the end of which kx agents oftype x are present
I The only effi cient (and stable) matching is:I min {kA, kα} A-squares matched with the same number of
α-roundsI max {kA, kα} −min {kA, kα} mismatched pairsI min
{kB , kβ
}B-squares matched with β-rounds
![Page 169: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/169.jpg)
Fixed-Window Mechanisms
I Can find the optimal window size no that balances marketthickness and waiting time
I We show no ≤ k̄dec
![Page 170: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/170.jpg)
Welfare Comparisons
Corollary (Performance of Fixed-Window Mechanisms)
limc→0
W opt (c)−W fix (c)W opt (c)−W dec (c)
=(1− p)U
2 (UA(α)− UA(β))< 1.
=⇒ Fixed-window mechanisms could potentially provide asubstantial improvement over decentralized ones
![Page 171: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/171.jpg)
Welfare Comparisons
Corollary (Performance of Fixed-Window Mechanisms)
limc→0
W opt (c)−W fix (c)W opt (c)−W dec (c)
=(1− p)U
2 (UA(α)− UA(β))< 1.
=⇒ Fixed-window mechanisms could potentially provide asubstantial improvement over decentralized ones
![Page 172: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/172.jpg)
Linear Taxation
I τ can be set so that
p (UA(α)− UA(β))c + τ
=
√p(1− p)U
2c
I A linear tax scheme —an agent who is k-th in line pays τ∗k(whether square or round)
![Page 173: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/173.jpg)
Linear Taxation
I When the length of queue is k̂, charge net tax for an agentwho is k-th in line:
τ∗k − 2 ·∑k̂k=1 τ∗k
2k̂= τ∗
(k − k̂ + 1
2
)
I The net added tax on waiting for the last person in the queue
is τ(k̂−1)2 , increasing in the queue’s length k̂
I Choose
τ∗(k̄opt − 1)2
= τ ⇐⇒ τ∗ =2τ
k̄opt − 1
![Page 174: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/174.jpg)
Linear Taxation
I When the length of queue is k̂, charge net tax for an agentwho is k-th in line:
τ∗k − 2 ·∑k̂k=1 τ∗k
2k̂= τ∗
(k − k̂ + 1
2
)
I The net added tax on waiting for the last person in the queue
is τ(k̂−1)2 , increasing in the queue’s length k̂
I Choose
τ∗(k̄opt − 1)2
= τ ⇐⇒ τ∗ =2τ
k̄opt − 1
![Page 175: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/175.jpg)
Linear Taxation
I When the length of queue is k̂, charge net tax for an agentwho is k-th in line:
τ∗k − 2 ·∑k̂k=1 τ∗k
2k̂= τ∗
(k − k̂ + 1
2
)
I The net added tax on waiting for the last person in the queue
is τ(k̂−1)2 , increasing in the queue’s length k̂
I Choose
τ∗(k̄opt − 1)2
= τ ⇐⇒ τ∗ =2τ
k̄opt − 1
![Page 176: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/176.jpg)
Fixed-Window Mechanisms
I The total matching surplus:
S(kA, kα) ≡{kαUAα + (kA − kα)UAβ + (n− kA)UBβ if kA ≥ kα,kAUAα + (kα − kA)UBα + (n− kα)UBβ otherwise
I The first square and round to arrive wait for n− 1 periods,the second square and round to arrive wait for n− 2 periods,etc. Thus, the total waiting cost is
2 · c · ((n− 1) + (n− 2) + · · ·+ 0) = cn(n− 1)
![Page 177: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/177.jpg)
Fixed-Window Mechanisms
I The total matching surplus:
S(kA, kα) ≡{kαUAα + (kA − kα)UAβ + (n− kA)UBβ if kA ≥ kα,kAUAα + (kα − kA)UBα + (n− kα)UBβ otherwise
I The first square and round to arrive wait for n− 1 periods,the second square and round to arrive wait for n− 2 periods,etc. Thus, the total waiting cost is
2 · c · ((n− 1) + (n− 2) + · · ·+ 0) = cn(n− 1)
![Page 178: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/178.jpg)
Fixed-Window Mechanisms
I Therefore, the expected per-period welfare for eachsquare-round pair generated by a window size n is
1n ∑0≤kα,kA≤n
(nkA
)(nkα
)pkAA (1− pA)
n−kApkαα (1− pα)
n−kαS(kA, kα)
−c(n− 1).
I Easy to show there is a finite solution (first term boundedabove)
I Hard to find closed-form solution
![Page 179: Dynamic Matching ŒPart 2 · Dynamic Matching Processes with Changing Participants Common to many matching processes: I Child Adoption: I About 1.6 million, or 2.5%, of children in](https://reader034.vdocuments.site/reader034/viewer/2022042612/5f3bc4aa99f92f5e5e3c6fec/html5/thumbnails/179.jpg)
Fixed-Window Mechanisms
Proposition 5
1. An upper bound for the optimal window size is given by
no ≤ p(1− p)U4c
.
2. For any ε > 0, there exists c∗ such that if c < c∗, a lowerbound for the optimal window size is given by
no ≥ p(1− p)U(4+ ε)c
.