Exploratory Model Building Toward Agent-Based Economics Takashi Iba Masaharu Hirokane Hirokane Kawakami Yoshiyasu Takefuji Heizo Takenaka [email protected].

Exploratory Model BuildingToward Agent-Based Economics

Takashi IbaMasaharu HirokaneHirokane KawakamiYoshiyasu Takefuji

Heizo Takenaka

[email protected]://www.boxed-economy.org/

Post-Office Box 17, Fujisawa Keio-mae Post Office:4250-1 Endo, Fujisawa, Kanagawa, Japan, zip 252-0816

Keio University, Japan

Simulation is …

How Building Reality?

Possible World

FACT simulation

Model andInitial condition

result

Incremental Development of the simulation model

Predictionor Explanation

ApplicationFeedback

Application

Feedback

Predictionor Explanation

Incremental D

evelopment

of the si

mulation m

odel

From the macroscopic viewpoint,

Exploratory Model Building

• Model Building

• Model Specification– Specification of initial settings

• Parameter Estimation

• Model Validation

From the microscopic viewpoint,

• Model Building

•cognitive science•social psychology•consumer behavior•marketing science•management science•artificial intelligence

•reinforcement learning•neural network

algorithm modelcomponent

meta-algorithmModel component

adaptivecomponent

agent modelcomponent

Phase 1Phase 1

There are some alternative models.

Phase 1Phase 1

• Model Building

•communication studies•social network analysis•Sociology•distributed artificial intelligence

agent modelcomponent

relation modelcomponent

composedsocial model

Phase 1Phase 1

• Model Specification

Phase 2Phase 2

Inductive Parameter Estimation

• Parameter Estimation

Phase 3Phase 3

Deductive Parameter Estimation

Deductive Estimation(for Parameter or Model)

Phase 3Phase 3

• Model Validation

• Comparison with observed data

• Comparison with other economic theories

• Comparison with empirical tendencies

Phase 4Phase 4

observeddata

other economictheories

empiricaltendencies

Parameter Estimation and Model Validation

Phase 4Phase 4Phase 3Phase 3

Total Flow of Exploratory Model Building

AutomaticModel Estimation System

parameter rangeand increments

social modeldesigns

data set forvalidation

componentmodel designs

modelcomponent

data set

Permissioncondition

simulation user

inputs

automatic modelestimation system

outputs

a=20b=5.34x=0.02

a=98C=12000Z>12345

some combinations ofmodel settings

application

For success with social simulation

“Neo-Instrumentalism”

• Instrumentalism+

• Extensibility of model• Reality of the assumptions

Open the Black Box !

Boxed Economy Model

= miniature of the entire economy