JASA

1
JASA

• A high performance open-source
• auction simulator
• http//www.csc.liv.ac.uk/sphelps/jasa

Steve Phelps sphelps_at_csc.liv.ac.uk Agent Research
Technology Group University of Liverpool
2
Background auctions

• Centralised resource allocation
• Agents submit their utility functions to a
  system agent (auctioneer), which computes the
  optimal allocation and payments.
• Typically used when
• Valuations (utility functions) vary rapidly over
  time
• Agents are uncertain about their own valuation
• Speed of convergence to the optimal allocation is
  a high-priority design objective
• When we have an impromptu need to thicken the
  market gather many buyers and sellers together
  simultaneously

3
Mechanism design

• Design objectives can vary
• Maximise social welfare
• Maximise seller revenue
• Minimise time to convergence
• Minimise computational complexity
• Budget balance
• No single optimal design- auction design is a MOO
  problem
• Auction theory results fail to hold for many
  real-world auctions
• Exchanges are particular hard
• Hence simulations can sometimes shed light on the
  grey areas.

4
Requirements

• A flexible laboratory framework for Agent-based
  Computational Economics (ACE)
• In ACE we often need to run experiments very many
  times.
• Were interested in applying evolutionary
  computing to ACE
• We would like to experiment with many different
  auction mechanisms, trading strategies and
  learning algorithms
• Replication work we would like a set of
  reference-implementations for the above

5
Design

• Light-weight High-performance
• Highly extensible
• Open-source
• Readable code
• Integration with ECJ for performing experiments
  using evolutionary computing
• http//cs.gmu.edu/eclab/projects/ecj/

6
.. Demo ..
7
Open Source

• JASA is a community-led project
• Hosted at Sourceforge http//sourceforge.net/pro
  jects/jasa
• Current contributors
• Jinzhong Niu (CUNY)
• Marek Marcinkiewicz (Columbia)
• We welcome further contributions in the form of
• New functionality (eg new trading strategies,
  learning algorithms, auction types)
• Suggestions for improvement
• Bug reports
• Bug fixes
• Anything else!
• Contact sphelps_at_csc.liv.ac.uk to become involved.