Difference between revisions of "Papers on Economic Agent-Based simulation"

From CSWiki
Jump to: navigation, search
(The ACE Trading World: Technical Details)
Line 48: Line 48:
  
 
You can access this paper [http://www.econ.iastate.edu/tesfatsi/AEAPP2008.LeBaronTesfatsion.ACEMacroModeling.Final.pdf here].
 
You can access this paper [http://www.econ.iastate.edu/tesfatsi/AEAPP2008.LeBaronTesfatsion.ACEMacroModeling.Final.pdf here].
 +
 +
==Constructive Modeling of Decentralized Market Economies: An Agent-Based Computational Economics Approach==
 +
Presentation given by Leigh Tesfatsion
 +
 +
A presentation on the general working of agent-based economic simulation.
 +
 +
*Difficulty of modeling real world dynamic economic systems
 +
**Distributed local interactions
 +
**Strategic behavior and uncertainty
 +
**Institutional constraints
 +
*Previously, assumed everyone rational, so no need to look deeper into cognition and social interactions
 +
*Agent-Based Computational Economics (ACE)
 +
**Economic world with various agent types
 +
***Agents adapt behavior, communicate, learn to complete goals
 +
**Set initial conditions
 +
**World develops over time
 +
**Events driven by agent interaction
 +
*ACE research
 +
**Empirical Understanding
 +
***Explanation for persistently observed empirical regularities
 +
***Are these trends always observed in the models?
 +
**Market Design
 +
***Efficient, fair, and orderly social outcomes?
 +
***Even if agents try to ‘game’ the system?
 +
**Qualitative Analysis
 +
***What are the performance capabilities?
 +
***Does coordination occur?
 +
*Starting point
 +
**Two-sector Walrasian equilibrium economy
 +
***No imposing equilibrium conditions
 +
***Agent driven production, pricing, and trading
 +
***Does equilibrium emerge?
 +
**Look for market organization
 +
***How does trading occur?
 +
***Bilateral or mediated trade?
 +
****Mediators: brokers (no inventory) and dealers (inventory)
 +
**Mediated markets: auction (central, clearing houses), OTC (not central, managed by dealers), exchanges (mix)
 +
**Look for learning behavior
 +
***Price/quantity discovery process
 +
***Buyer-seller interaction networks?
 +
****Long term relationships: trust?
 +
****Trading
 +
*****Terms of trade
 +
*****Seller-buyer matching
 +
*****Trade
 +
*****Settlement
 +
*****Manage longer term relationships
 +
*Illustration by Hash-and-Beans economy
 +
**Learning method can matter
 +
*Issues
 +
**Setting initial conditions?
 +
***Carrying capacity
 +
***Market clearing
 +
***Market efficiency
 +
*Disadvantages
 +
**Intensive experiments often needed
 +
**Multi-peak rather than central-tendency outcome distributions can arise
 +
**Difficult to learn and build platform
 +
*Advantages
 +
**Permits systematic study of behavior and markets
 +
**Can look at interesting problems and get quick feedback
 +
 +
The presentation can be found [http://www.econ.iastate.edu/tesfatsi/ACEIntro.Econ502.pdf here].

Revision as of 03:01, 1 October 2008

Agent-Based Computational Economics: A Constructive Approach to Economic Theory

Leigh Tesfatsion

  • Uses Walrasian Auctioneer pricing mechanism: simultaneous auction where each agent calculates its demand for the good at every possible price and submits this to auctioneer.
  • Difference between ACE and other models: Events are driven by agent interactions after initial conditions; goes TOWARDS equilibrium rather than starting with one.
  • Disadvantages: validating ACE model outcomes against empirical data. An empirically observed outcome might be a low-probability event lying in a relatively small peak of the outcome of distribution for this true data-generating process, or in a thin tail of this distribution.
  • States that ACE models are still just a complement, not a substitute, for analytical and statistical modeling approaches.

The ACE Trading World: Technical Details

Initial Conditions

  • 0 to T periods, finite number of profit seeking hash and bean firms, finite number of consumers
  • Each firm has positive amount of money and production capacity
  • Each total cost function has amortized fixed costs proportional to capacity.
  • Known to the companies: # of hash firms, bean firms, and # of consumers, and that hash and beans are both perishable.
  • Unknown: income levels and utility functions of consumers or cost functions/capacities of other firms

Other Information

  • Collusion is prohibited
  • Each consumer has a lifetime money endowment profile and utility function measuring preferences and subsistence
  • Income is determined partly by money endowment, savings from previous periods and dividends

How Things Work

Firms use supply offer and learning method. Consumers can acquire complete info about the supply offers when they are posted at the beginning of the period. Price discovery process: Person who can’t meet subsistence needs at lowest prices will exit the process. Remaining people determine demands conditional on remaining income, unmet needs, and lowest prices. They submit demands to firms that posted low prices, then the firms try to satisfy demands with rationing method. Consumers rationed with less than needs adjust their demand downward for remaining goods to preserve income. Trades happen, then firms with surplus/consumers with unspent income continue into the next round. End of rounds: firms stocked out or when there is no more income. Consumers who exit or finish with needs still-> die at end of period and their unspent money holdings are lost to the economy but stock shares are redistributed.

SUMMARY OF TECHNICAL PORTION: ACE trading world is a deterministhic Markov process where X(T+1) is determined as a function of X(T), where profiles and metrics are initialized. It is supposedly being implemented, as detailed in COOK AND TESFATSION (2006) as a computational lab with graphical interface.

There is an appendix detailing the mathematical models and classes.

Link to Paper

Other Relevant Materials

  • ACE course relying heavily on computational lab exercises
  • Other ACE-related course materials

Modeling Macroeconomies as Open-Ended Dynamic Systems of Interacting Agents

Blake LeBaron and Leigh Tesfatsion

This article just gives an overview about ACE: what ACE is about and used for. It is not as useful in giving any technical specifics. It states problems of ACE, such as degrees of freedom, the fact that modeling tools may be too rich to fit to data, and that the properties of ACE are not understood well enough to model human behavior. It also lists some things that ACE can model (National economy, financial sector, etc.). Policy makers and how they can use ACE is stated on page 9.

The article's three criteria for ACE are: 1. Model must include appropriate empirically-based taxonomy of agents 2. Scale of model must be suitable for particular purpose 3. Model specifications must be subject to empirical validation

You can access this paper here.

Constructive Modeling of Decentralized Market Economies: An Agent-Based Computational Economics Approach

Presentation given by Leigh Tesfatsion

A presentation on the general working of agent-based economic simulation.

  • Difficulty of modeling real world dynamic economic systems
    • Distributed local interactions
    • Strategic behavior and uncertainty
    • Institutional constraints
  • Previously, assumed everyone rational, so no need to look deeper into cognition and social interactions
  • Agent-Based Computational Economics (ACE)
    • Economic world with various agent types
      • Agents adapt behavior, communicate, learn to complete goals
    • Set initial conditions
    • World develops over time
    • Events driven by agent interaction
  • ACE research
    • Empirical Understanding
      • Explanation for persistently observed empirical regularities
      • Are these trends always observed in the models?
    • Market Design
      • Efficient, fair, and orderly social outcomes?
      • Even if agents try to ‘game’ the system?
    • Qualitative Analysis
      • What are the performance capabilities?
      • Does coordination occur?
  • Starting point
    • Two-sector Walrasian equilibrium economy
      • No imposing equilibrium conditions
      • Agent driven production, pricing, and trading
      • Does equilibrium emerge?
    • Look for market organization
      • How does trading occur?
      • Bilateral or mediated trade?
        • Mediators: brokers (no inventory) and dealers (inventory)
    • Mediated markets: auction (central, clearing houses), OTC (not central, managed by dealers), exchanges (mix)
    • Look for learning behavior
      • Price/quantity discovery process
      • Buyer-seller interaction networks?
        • Long term relationships: trust?
        • Trading
          • Terms of trade
          • Seller-buyer matching
          • Trade
          • Settlement
          • Manage longer term relationships
  • Illustration by Hash-and-Beans economy
    • Learning method can matter
  • Issues
    • Setting initial conditions?
      • Carrying capacity
      • Market clearing
      • Market efficiency
  • Disadvantages
    • Intensive experiments often needed
    • Multi-peak rather than central-tendency outcome distributions can arise
    • Difficult to learn and build platform
  • Advantages
    • Permits systematic study of behavior and markets
    • Can look at interesting problems and get quick feedback

The presentation can be found here.