Modeling of Market Behavior in Different Scales Using Multi-agent Simulation of Participants' Cognitive Behavior

Staff - Faculty of Informatics

Start date: 18 April 2012

End date: 19 April 2012

You are cordially invited to attend the PhD Dissertation Defense of Maryam ESMAEILI on Wednesday, April 18th 2012 at 15h30 in room A32 (Red building)

Abstract:
Agent-based modeling and computer simulation has received a large amount of attention by many researchers in different fields such as economics and sociology in the last decade. This has led this technique to become one of the mainstreams of complex systems modeling. Studying and modeling the complex social systems need different researchers from different fields such as computer science, social science, modeling etc. to work together and therefore making multidisciplinary research groups in order to manage studying of such complex systems.

In agent-based modeling approach, the models are classified concerning the aim of the research. For instance, some researchers are trying to present a generic model in order to explain a complex phenomenon at a very abstract level. This is done to prove a theory or support a hypothesis. These types of agent-based models are called theoretical agent-based models. On the other hand, some researchers try to produce expressive models in which using empirical data a specific case study is modeled and some facts are described. These types of models are called empirically grounded agent-based models. In such models some issues such as empirical initialization, the limitations of data collection, empirical validation or the role of data in the design must be addressed. So we need to propose a methodology that considers the key role of empirical data throughout all the modeling stages. Following this methodology, we have advocated that when there are available data from the observation of the real phenomenon, the modeling and simulation process involves additional stages to address the mentioned issues. Through the proposed methodology, we show how to use the empirical data in the modeling to bring the results closer to the real phenomenon under study, while following the theory in the whole process (Chapter 2).

The proposed methodology has been supported technically by implementation of an integrated (micro/macro) simulation model. The model has been structured in modules, which let us to enable or disable some modules to explore the model space. Following the exposed perspective, we have developed a case study to test and validate the application of the proposed methodology and framework. This case study addresses the complex issue of aggregated phenomena in housing market emerged from decisions of individual agents at the micro-level, space and the macro-economic indicators at the macro-level. Specifically, this study seeks to identify traceable connections between micro and macroeconomic scales exploring a city in southern part of Switzerland. Two different types of aggregated phenomena, which have been studied in this thesis, are the price dynamics of housing and distribution of different ethnic groups across the city of Lugano. Although the application is specific, the model is flexible and can be used in many other cases (Chapter 3).

Two different models are recognizable in this work, which are a micro-model (Chapters 3 and 5) and a macro-model (Chapter 6). Both models are theoretically supported and have been developed within a modular agent framework, designed in incremental layers. The developed model is validated from a quantitative macro point of view (empirical validation), quantitative micro point of view (internal validation) and a qualitative macro point of view (theoretical validation) (Chapter 7). We have used different technics of artificial intelligence in the model to test the adaptability of the framework and applicability of those AI technics in social simulation. Type-1 and Type-2 fuzzy logic systems have been used intensively in order to perform the reasoning process in the software agents (Chapter 4).

Using empirically-grounded agent-based modeling technic, we have addressed some issues like connecting the micro- and macro- scale phenomena in housing market, connecting the economic and cellular spatial simulation models, connecting the real and survey data to the model for initialization and validation and connecting the conventional econometric model to the developed integrated model, we have addressed some challenging issues in modeling of complex phenomena. Being in a multi-disciplinary research group, the model has been modified calibrated and validated using the empirical data and validated econometric results.

Dissertation Committee:

  • Prof. Luca Maria Gambardella, SUPSI, Switzerland (Research Advisor)
  • Dr. Paolo Giordano, University of Vienna, Austria (Research co-Advisor)
  • Dr. Alberto Vancheri, SUPSI, Switzerland (Research co-Advisor)
  • Prof. Fabio Crestani, Università della Svizzera italiana, Switzerland (Internal Member)
  • Prof. Rolf Krause, Università della Svizzera italiana, Switzerland (Internal Member)
  • Prof. Boi Faltings, EPFL, École Polytechnique Fédérale de Lausanne, Switzerland (External Member)
  • Prof. Maria Fasli, University of Essex, United Kingdom (External Member)
  • Prof. Roland W. Scholz, Swiss Federal Institute of Technology Zurich, Switzerland (External Member)