CCSS School on Computational Social Science: Agent-Based Models (ABM) in Economics

15:00～15:10　開会挨拶 / Opening Remarks

Takashi KAMIHIGASHI（Director, Center for Computational Social Science, Kobe University）

Session 1

15:10～15:50

論題
Topic

Introduction to ABM

報告者
Presenter

Corrado DI GUILMI (Economics Discipline Group, University of Technology Sydney)

概要
Abstract

The talk will provide a broad overview of agent-based modeling, from its theoretical foundations, with particular focus on complexity, to the applied aspects of computer coding. Agent-based modeling is rooted in complex system theory, implying a fundamentally different approach with respect to the reductionism that underpins mainstream economic modeling. The application of complexity in economics faces methodological issues that most recent theoretical developments have attempted to address.

15:50～16:30

論題
Topic

Agent Based Model on Actual Supply-chain Network

報告者
Presenter

Hiroyasu INOUE (Graduate School of Information Science, University of Hyogo)

概要
Abstract

Although supply chain disruption is a threat for our economy and society, it is almost unpredictable. The reason of the unpredictability comes from the complexity of the supply chain: products/services have been highly developed and a lot of firms are involed into them. This study focuses on actual supply chains, or supplier--customer networks of firms and applies an agent-based model to the actual supply chains of nearly one million firms and five million relationships in Japan to examine how production losses caused by disasters and pandemics. We estimate the economic losses of the 2011 Great East Japan earthquake and employ the same model to predict the loss of the Nankai earthquake, a mega earthquake predicted to hit major industrial cities in Japan in the near future. In addition, we replicate the economic behavior of the 1st wave of the pandemic in 2020.

16:30～17:10

論題
Topic

Deep Neural Networks in an ABM Virtual Experiment

報告者
Presenter

David GOLDBAUM (Economics Discipline Group, University of Technology Sydney)

概要
Abstract

We use machine learning to model subject decisions in an experiment based on a dynamic game. In this game, players appropriately use prior experiences to inform current decisions. A structural empirical model provides insight into how decisions are informed from prior experiences, but we determine the model to be insufficiently rich to capture decisions well enough to then conduct simulations. Machine learning achieve substantial gains in performance. The role of interactions in determining strategy lends itself to analysis with agent-based modeling. We use the estimated neural network model to run counterfactual virtual simulations that provide information on how well the populations played the game.

17:10～17:30 休憩　Short Break

Session 2

17:30～18:10

論題
Topic

Approximate Bayesian Inference for Agent-Based Models in Economics: A Case Study

報告者
Presenter

Thomas LUX (Faculty of Business, Economics and Social Science, Christian-Albrechts-Universität zu Kiel)

概要
Abstract

Estimation of agent-based models in economics and finance confronts researchers with a number of challenges. Typically, the complex structures of such models do not allow to derive closed-form likelihood functions so that either numerical approximations to the likelihood or moment-based estimators have to be used for parameter inference. However, all these approaches suffer from extremely high computational demands as they typically work with simulations (of the agent-based model) embedded in (Monte Carlo) simulations conducted for the purpose of parameter identification. One approach that is very generally applicable and that has the potential of alleviating the computational burden is Approximate Bayesian Computation (ABC).While popular in other areas of agent-based modelling, it seems not to have been used so far in economics and finance. This paper provides an introduction to this methodology and demonstrates its potential with the example of a well-studied model of speculative dynamics. As it turns out, ABC appears to make more efficient use of moment-based information than frequentist SMM (Simulated Method of Moments), and it can be used for sample sizes of an order far beyond the reach of numerical likelihood methods.

18:10～18:50

論題
Topic

An Introduction to the Frequentist Approach to ABM Estimation

報告者
Presenter

Leonardo BARGIGLI (Department of Economics and Management, University of Florence)

概要
Abstract

In the presentation, first of all, I will try to motivate the opportunity to proceed with the estimation of agent-based models. Then I will focus on various problems related to the frequentist estimation of ABMs. Among these, the main ones concern the computational trade-offs of a frequentist approach based on simulations and the optimal design of the latter; the sensitivity analysis of agent-based models and the identification of their parameters; the approximation techniques which can reduce the computational burden of the estimation.

18:50～19:30

論題
Topic

Experiments and ABM

報告者
Presenter

Annarita COLASANTE (Università degli studi di Roma Unitelma Sapienza)

概要
Abstract

Experimental methodology is spreading in the last decades since it is an effective way to collect data and observe individual behaviors. Agent Based Modeling is increasingly been used to model macro phenomena by using a bottom up approach. These two methodologies are complementary: on the one hand, results collected by experiments are useful to identify micro behavior that could be then implemented for the ABM microfundation; on the other hand, results from ABM may be deployed to validate experimental results. The presentation, after a brief introduction about experimental methodology, aims at presenting how experimental data could be implemented to both calibrate and validate ABM.

19:30～20:10

論題
Topic

Inequality and Monetary Policy with ABM

報告者
Presenter

Alberto RUSSO (Department of Economics, Universitat Jaume I)

概要
Abstract

During the last two decades, there has been a blossoming of agent-based applications to macroeconomic issues. There is a variety of macro ABMs, from simple ones to large-scale multi-country models. This approach represents an alternative to DSGE models for both policy analysis and forecasting. While large-scale calibrated models could be required for an accurate forecasting, small-scale models are very useful for reproducing relevant stylized facts as well as for policy analysis. We design a minimal macro ABM to study how income inequality affects monetary policy through the inequality-household debt channel. We find that when inequality meets financial liberalization, a leaning against-the-wind strategy can preserve financial stability at the cost of high unemployment, whereas an accommodative strategy, i.e. lowering the policy rate, can dampen the fall of aggregate demand at the cost of larger financial leverage (as under the 'Greenspan put'). We conclude that inequality may constrain the central bank, even when it is not explicitly targeted.

20:10～　閉会挨拶 / Closing Remarks

Masahiko SHIBAMOTO (Center for Computational Social Science, Kobe University)