Adam Arsenault

1
A Multi-Agent Systems Approach to Farmland
Auction Markets

• Adam Arsenault
• Department of Agricultural Economics
• University of Saskatchewan

UNIVERSITY OF SASKATCHEWAN Saskatoon,
Saskatchewan, Canada. www.usask.ca
2
The Issue

• Multiple farmland auction markets and
  interactions are poorly understood
• Current methodologies cannot account for
• Interaction between farmers (agents) and their
  environment
• Heterogeneity of land and farmer characteristics
• Spatial and temporal aspects of land purchases
  and sales
• Little is known about the impacts of bidder
  learning on farmland prices

3
Project Objectives

• Develop modern auction theory applicable to
  Canadian prairies farmland market
• Specifically, this entails
• Incorporating heterogeneity into farmer
  characteristics and geographical landscape
• Incorporating learning mechanisms and feedback
  into bidding strategies for agricultural land
• Better understanding the effects of farmer
  interactions and adaptive learning in a repeated
  game of bidding in auction markets

4
Difficulties with the Classical Model - Agents

• Classical models assume rational expectations and
  optimization in bidding strategies
• Optimization strategies tell us nothing about the
  negotiation and bidding process (Zeng et al 1998)
  
• Results in only strategic moves No interaction
  or learning (Selten 2001, Selten and Neugebauer
  2006)
• Classical models cannot find solutions to
  analytically complex problems (Tesfatsion 2002)
• Simple models of interactions and feedback
  (learning) become complex very quickly ?
  Nonlinearity

5
Difficulties with the Classical Model - Land

• Space and the heterogeneity of land
• Law of one price may not always apply to land
• Land Price F ( Location, Quality, Time, ß )
• Time is crucial in land acquisitions
• Land is a lumpy investment
• Desired land not always available
• Outcome of auction (win/lose) is critical in
  success of individual farms ? Bidding strategies
  and learning are paramount

6
The Model

• Multi-Agent Systems (MAS)/ Agent-Based
  Computational Economics (ACE)
• The computational study of economies modeled as
  evolving systems of autonomous interacting
  agents. (Tesfatsion 2002)
• Agents have Goals, Actions, and Domain Knowledge
  (Stone and Veloso 2000)
• Dynamic model of heterogeneous agent interactions
  
• Learning in repeated games of bidding using
  behavioral adaptations
• Dynamic model of farmland markets with spatial
  and temporal factors
• Land is lumpy, heterogeneous, and
  non-transportable

7
Why Agents Matter

• Heterogeneity, interaction, and feedback (Stone
  and Veloso 2000)
• Agents interact with one-another and surrounding
  environment not simply market clearing prices

8
System Flow
9
How Efficient Are Current Farmland Markets?

• How well does game theoretic predicted price map
  actual price?
• Is farmland available to those who need it when
  they need it?
• Do better informed farmers actually do better? ?
  Better bidders
• What salient features of farmer behavior/learning
  are driving trends in farmland prices?

10
QUESTIONS?
Thank You!