Analyzing Micro-Macro Structures in a Financial Market
via Agent-Based Simulation
 

     This paper describes intermediate results on the analysis of micro- and macro- structures of a financial market.  Instead of using conventional behavioral financial models, which consist of homogeneous decision making agents, we employ agent-based simulation approaches to the analysis.
     Our simulation model is characterized by
 
        1) rational and non-rational agents with decision making strategies about trading the assets of either individual stocks or  riskless assets;
        2) an artificial market, in which the benefits or losses will occur based on the Brownian motion, and each agent trades its asset based on its benefits/losses and past pricing information¥cite{Takahashi}.
    
    Using the model, the objective of the research is to investigate the effects of 1) the value at risk (VaR), 2)
the concepts of portfolio insurance, and 3) the effects of herding behaviors among agents with decision making strategies.
     The simulation model have shown that 1) conventional risk management techniques in the literature are effective in the usual cases, 2) dynamics of asset pricing techniques so far are coincide with the theoretical results, however, 3) the market prices would become too worse compared with the theoretical ones, if i) risk management strategies would be too sensitive, or ii) there would exist so many investors with herding characteristics.  The results implies that the agent-based approach is promising when the assumptions of the analysis are realistic and/or complex.


Takao Terano
University of Tsukuba
Graduate School of Systems Management
terano@gssm.otsuka.tsukuba.ac.jp 

Hiroshi Takahashi
University of tsukuba, Tokyo
Graduate School of Systems Management
taishi@rr.iij4u.or.jp


Human- and Software-Agent Integration for Analyzing Business Decision Making


    This research reports an application of the agent approach to business games.  We will discuss the topic of the analyzing business decision making procedure by using Human- and Software-Agent integration in business game domains.
      The agent approach is attracting attention in social sciences from the viewpoint of cooperation of both software and human players and agents. A business game is a kind of the gaming simulation which uncovers the decision making procedures executed by the plural participant players who have a common purpose is to solve a specific problem [Duke 1974].
      In order to achieve the purpose, we have developed a game construction toolkit, which consists of a simple Business Model Description Language (BMDL), Agent Rules written in Ruby language [Matsumoto 1996], and their Business Model Development System (BMDS) [Fujimori 1999] [Terano 1999].  Furthermore, we have linked the software agents with machine learning programs by inter process communication.
      The developed simulators can be used by both human users and software agents in the WWW environment. Through the educational experience and intensive computer experiments, we have found the decision making procedures to a specific business model.
      This research describes the background and motivation, basic principles, the architecture and implementation of BMDL/Agent Rules/BMDS, some results of experiments as current detail, and learning software agent as a new function.
       The main contribution of the research is 1) to propose a general architecture for the human- and software-agent integration approach to analyzing of decision making procedures for an arbitrary business model by learning software agents, and 2) to demonstrate the effectiveness of human players and learning software agents to a business simulator by exemplifying the novel business simulation toolkit.
      This paper is organized as follows:

1. To discuss the background and motivation of the research.
2. To explain the basic architecture.
3. To explain the implementation of the agent system.
4. To explain the implementation of learning system
    [Butz 2000] [Takadama 2000]
5. To describe the experimental setup and the results.
6. To give some concluding remarks and future work.

      In conclusion, by using learning software agent and human players, we have shown the effectiveness to analyze the decision making principles in a human complex system.
      The future work includes to uncover how mutual learning by plural human and software agents grows up global beneficial systems in business environment.

[Butz 2000] Martin V. Butz, Stewart W. Wilson: An Algorithmic Description of
    XCS,
    ftp://ftp-illigal.ge.uiuc.edu/pub/papers/IlliGALs/2000017.ps.Z, (2000).
[Duke 1974] Richard D. Duke, GAMING: THE FUTURE'S LANGUAGE, Sage Publication,
    Inc. (1974)
[Fujimori 1999] Fujimori, H., Kuno, Y., Shirai, H., Suzuki, H., Terano, T.:
    Alexander Islands: GSSM Tiny Business Simulator on the WWW. S. Morgan,
    D. Page (eds.): Developments in Business Simulation and Experiential
    Learning, Vol. 26, (Proc. ABSEL'99), pp. 224-225.
[Matsumoto 1996] Yukihiro Matsumoto: http://www.ruby-lang.org/
[Takadama 2000] Keiki Takadama, Takao Terano, Katsunori Shimohara:
    Learning Classifier Systems Meet Multiagent Environments, Advances in
    Learning Classifier Systems, Advances in Learning Classifier Systems
    (IWLCS 2000), pp. 192-210,
    (2000).
[Terano 1999] Terano, T., Suzuki, H., Kuno, Y., Fujimori, H., Shirai, H.,
    Nishio, C., Ogura, N., Takahashi, M.: Understanding Your Business
    through Home-Made Simulator Development. Developments in Business
    Simulation and Experiential Learning, Vol. 26, (Proc. ABSEL'99),
    pp. 65-71.


Masato Kobayashi
University of Tsukuba
Graduate School of Systems Management
masato@gssm.otsuka.tsukuba.ac.jp

Takao Terano
University of Tsukuba
Graduate School of Systems Management
terano@gssm.otsuka.tsukuba.ac.jp
 

Money, Competition, or Policy, How to Motivate Young Students:
Agent-Based Modeling


    In this research, the causal relationship between student motivation for learning and the educational policy is analyzed by Agent-Based Modeling. This research proposes a novel application in the social science and pedagogy by Agent-Based Approach [Terano 1998].  We describe that the influence of the educational policies upon student motivation for learning changes seriously by disparities in the social stratum.  The social stratum means that amount of investment in education with wealth gap, and strength
of motivation.  The scenario of the model in an artificial society [Epstein 1996] where agents act is described below.
    There are three kinds of agents with different motivation for learning in an artificial society.  The agents hybridize when they reach fixed age, and have offspring.  Agents will receive sometime challenging tasks, and obtain rewards in proportion to their level of motivation, and finally, they will advance their level of academic achievement.  And then agents will be motivated to advance to higher education if agents attain the some amount of scholastic proficiency.  However, as educational expenses are necessary to gain the task, the level of  academic achievement varies in their income. Moreover, in this society, the educational policies, such as increasing or decreasing competitive entrance examinations between agents, and depending on their motivation for learning, are introduced.
    The educational policy in this case mainly involves increasing or decreasing the contents of study, or changing the level of difficulty of study by changing course curriculum.  On the other hand, agents as the students receive income from their parents and the influence of parent’s motivation for educating children, thus agents already have specific motivations for learning under the given conditions.
     Motivation for learning is, as a rule, based on the characteristic actions of Atkinson’s theory, or Achievement motivation [Atkinson 1974].  Namely, that the motivation for learning is defined by the resultant tendencies of the motive to achieve success (Ms) and motive to avoid failure (MAF).  The orders of these achievement tendencies are assumed to be as follows:

        Ms> MAF,  Ms= MAF,  Ms< MAF

    Therefore, the level of student’s motivation for learning corresponds to the orders of the achievement tendencies, and task preference and the amount of accomplishment differ by depending on the agents.  In addition, the task has the probability of success ( Ps) in proportion to its difficulty.  So, agents have the score of  tendency of achievement motivation (TA) by the following formula, and moves to the acquisition of the task.

        TA = ( Ms -MAF) _*__ _[Ps _*__ ( _1_ _- Ps)]

    The result of the research shows that in this artificial society, the level of achievement for learning and the motivation to advance to the next stage of education will change with each generation.  The conclusion of our research shows how the ratio of students who progress to the next stage of
education changes with the educational policy.  Namely, if the educational policy which gives the students latitude is introduced to the artificial society, the incentive from outside will become weak, and then the following results are obtained.

        1. In students belonging to the high rank, their motivation for learning decreases lightly.
        2. In students belonging to the less than the middle rank, their motivation for learning decreases seriously.  The lower their ranks are, the lower their motivation for learning is.
        3. In the results of the above, in the artificial society, the motivation for learning is bipolarized. Whole level of achievement for learning also decreases.

    On the other hand, if the educational policy which promotes competitive entrance examinations are introduced to the artificial society, the following results are obtained.

         1. In students belonging to the high rank, their motivation for learning increases.
         2. In students belonging to the less than the middle rank, their motivation for learning also increases. However, if the competitive entrance examinations are too seriously, some students drop out from the competitions, and then they are not interested in gaining the task.
         3. In the results of the above, whole level of achievement for learning increases more than it was.

    Future work includes exploring the various factors about the next stage of education.  It is because, in general, the next stage of education, especially the ratio of students who go on to a university or a college, depends on the policy of the country.  However, it is quite interesting that the rate of the entering university does not always correspond to the national power and the economic growth of that country.

REFERENCES
[Atkinson 1974] J.W.Atkinson, J.O.Raynor: Motivation and Achievement,
Winston & Sons, 1974
[Epstein 1996] J.Epstein, R.Axtell: Growing Artificual Societies, The MIT
Press, 1996
[Terano 1998 ] T. Terano, S. Kurahashi, U. Minami: TRURL: Artificial World
for Social Interaction Studies.  Proc. 6th Int. Conf. on Artificial Life
(ALIFE VI), pp. 326-335, 1998.


Atsuko Arai
University of Tsukuba
arai@gssm.otsuka.tsukuba.ac.jp

Takao Terano
University of Tsukuba
terano@gssm.otsuka.tsukuba.ac.jp