27
Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

  • View
    215

  • Download
    1

Embed Size (px)

Citation preview

Page 1: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Multi-Attribute Exchange Market:Search for Optimal Matches

Eugene FinkJianli Gong

John Hershberger

Page 2: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Motivation• Build an automated exchange for trading goods and services

• Combine the speed of the stock exchange with the flexibility of eBay

+

Page 3: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Previous work

• Combinatorial auctions

• Standardized exchanges

• Standardized exchanges

- Complex goods - Asymmetry between buyers and sellers - Illiquid

- Simple goods - Symmetry between buyers and sellers - Liquid

Page 4: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Research goals

Build an automated exchangefor non-standardized goods.

• Support fast-paced trading for markets with millions of orders

• Include optimization techniques to maximize traders’ satisfaction

Page 5: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Outline

• Multi-attribute orders

• Best-price matches

• Quality functions

• Experimental results

Page 6: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Market

A market is a set of items that can betraded, defined by a list of attributes.

ExampleA used-car market is a set of all conceivable vehicles, defined bymodel, year, and mileage.

Page 7: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

OrdersA trader specifies a buy or sell orderby attribute values and a price limit.

Buy orderModel: Mustang or CorvetteYear: 2002..2004Mileage: 0..10KPrice: $32,000

ExampleSell orderModel: MustangYear: 2004Mileage: 1KPrice: $30,000

A value specification may includelists of values and numeric ranges.

Page 8: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Matching

0102

0304

Year

Model

Price

Camaro Mustang Corvette

$32,000

$30,000

Sell orderMustang,

made in 2004,$30,000

Buy orderMustang,

made after 2001,$32,000

TradeMustang,

made in 2004,$31,000

Page 9: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Outline

• Multi-attribute orders

• Best-price matches

• Quality functions

• Experimental results

Page 10: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Main structures

• Tree of fully specified orders

• Unordered list of the other orders

Page 11: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Depth-first search

Model

Mileage

Year Year

CorvetteMustang

1996

50K

1996 Mustang,50K miles,

$7,000

Year

Camry

1999

Mileage

1999 Camry,10K miles,

$18,000

10K

Mileage

1988

1988 Corvette,20K miles,

$4,000

20K

Mileage

2000

2000 Camry,180K miles,

$5,000

180K

Buy Order:Any car made

after 1990

Best price

Page 12: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Depth-first search

Drawback:

If there are many matching leaves,the search takes a long time.

Solution:

Apply best-first search :• Store the best price for each subtree• Use these prices to guide the search

Page 13: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Best prices for subtrees

Model

Mileage

Year Year

CorvetteMustang

1996

50K

1996 Mustang,50K miles,

$7,000

Year

Camry

1999

Mileage

1999 Camry,10K miles,

$18,000

10K

Mileage

1988

1988 Corvette,20K miles,

$4,000

20K

Mileage

2000

2000 Camry,180K miles,

$5,000

180K

$7,000 $18,000 $5,000 $4,000

$7,000 $5,000 $4,000

$4,000

Page 14: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Search for the best price

Model

Mileage

Year Year

CorvetteMustang

1996

50K

1996 Mustang,50K miles,

$7,000

Year

Camry

1999

Mileage

1999 Camry,10K miles,

$18,000

10K

Mileage

1988

1988 Corvette,20K miles,

$4,000

20K

Mileage

2000

2000 Camry,180K miles,

$5,000

180K

$7,000 $18,000 $5,000 $4,000

$7,000 $5,000 $4,000

$4,000

Buy Order:Any car made

after 1990

2000 Camry,180K miles,

$5,000

Best price

Page 15: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Outline

• Multi-attribute orders

• Best-price matches

• Quality functions

• Experimental results

Page 16: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Quality functions

A trader can specify a quality function that ranks the acceptable transactions.

The transaction quality may depend on an item and its price.

Example:

Quality of a car purchase+4 for each year after 1970–5 for each 10K miles–6 for each $1,000

Page 17: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

The transaction quality may depend on an item and its price.

The system searches for the matches with the highest quality.

Quality functions

A trader can specify a quality function that ranks the acceptable transactions.

Page 18: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Depth-first search

Model

Mileage

Year Year

CorvetteMustang

1996

50K

1996 Mustang,50K miles,

$7,000

Year

Camry

1999

Mileage

1999 Camry,10K miles,

$18,000

10K

Mileage

1988

1988 Corvette,20K miles,

$4,000

20K

Mileage

2000

2000 Camry,180K miles,

$5,000

180K

Buy Order:Any car made

after 1990

Quality: 37 Quality: 3 Quality: 0 Highest quality: 37

Page 19: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Monotonic attributes

• The quality monotonically changes with the price

• Usually, it is also monotonic on several other attributes

Example:

Car quality• Increases with the year• Decreases with the mileage

Page 20: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Best-first search

• For every subtree, store the best value of each monotonic attribute

• Use these values to estimate the quality of the best match in every subtree

Page 21: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Best values for subtrees

Model

Mileage

Year Year

CorvetteMustang

1996

50K

1996 Mustang,50K miles,

$7,000

Year

Camry

1999

Mileage

1999 Camry,10K miles,

$18,000

10K

Mileage

1988

1988 Corvette,20K miles,

$4,000

20K

Mileage

2000

2000 Camry,180K miles,

$5,000

180K

2000,10K,

$4,000

1988,20K,

$4,000

2000,10K,

$5,000

1996,50K,

$7,000

20K,$4,000

180K,$5,00010K,

$18,00050K,

$7,000

Page 22: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Search for the best match

Model

Mileage

Year Year

CorvetteMustang

1996

50K

1996 Mustang,50K miles,

$7,000

Year

Camry

1999

Mileage

1999 Camry,10K miles,

$18,000

10K

Mileage

1988

1988 Corvette,20K miles,

$4,000

20K

Mileage

2000

2000 Camry,180K miles,

$5,000

180K

2000,10K,

$4,000

1988,20K,

$4,000

2000,10K,

$5,000

1996,50K,

$7,000

20K,$4,000

180K,$5,00010K,

$18,00050K,

$7,000

Buy Order:Any car made

after 1990

91

85 3837

37

Highest quality: 37

Page 23: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Outline

• Multi-attribute orders

• Search for matches

• Quality functions

• Experimental results

Page 24: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Performance

:

• 2 GHz CPU

• 1 Gbyte memory

• 166 MHz bus

Experiments using aPentium computer

Page 25: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

• Car market with eight attributes

number of orders

orde

rs p

er s

econ

d

102

103

104

105

101 103 104 105102

: 200 to 20,000 orders per second

Cars and bonds

: 500 to 50,000 orders per second

bonds

• Bond market with two attributes

cars

Page 26: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Artificial markets Synthetic market data:• 1 to 100 attributes• 300,000 orders

Best-First

Depth-First

number of attributes3 10 30 100

orde

rs p

er s

econ

d

1

102

103

104

105

101

Page 27: Multi-Attribute Exchange Market: Search for Optimal Matches Eugene Fink Jianli Gong John Hershberger

Summary

• General model for trading

of multi-attribute goods• Fast identification of matches between buy and sell orders