We propose agent based modeling and simulation as a new approach to decision support in the payments domain, which will assist banks in both strategic and ...
ISBN: 972-8924-13-5 © 2006 IADIS
MODELING CONSUMER BEHAVIOUR TOWARDS PAYMENT SYSTEM SELECTION USING MULTIAGENT BASED SIMULATION George Rigopoulos, Konstantinos D. Patlitzianas, Nikolaos V. Karadimas National Technical University of Athens School of Electrical & Computer Engineering
ABSTRACT Payments market is undergoing critical changes, deriving from technology and legislative regulations. Banks face increasing competition from new players in the market and need to adapt their decisions according to the new framework. Retail payments, a major service that banks provide to their customers, and contribute to their revenues, are more subject to digital evolution. Both strategic and tactical decisions referring to payment systems require appropriate tools that integrate knowledge from complementary scientific disciplines, and provide intelligent assistance to decision makers. Multi-agent modeling and simulation is becoming popular in various fields and especially in social and economic studies, as it provides a bottom-up approach to domains with high complexity. This paper introduces an architecture towards a decision support system based on multi-agent based simulation (MABS-DSS). The proposed architecture is oriented towards the retail digital payments domain, but may be extended to relevant domains with similar characteristics. The paper demonstrates the initial architecture of the proposed MABS-DSS, which utilizes multi-agent based simulation, as a valid approach for decision support in domains with network effects. KEYWORDS Multiagent simulation, decision support, payment systems.
1. INTRODUCTION Decision makers in the retail digital payments market are facing new challenges. Emerging payment methods and instruments are competing with traditional cash. Evolution of mobile devices is expanding needs to mobile retail payments, and future technologies may also impose new needs in payments (Markose 2000). Decision challenges are both strategic and tactical for payment service provides and especially banks. Strategic decisions deal with market entrance or exit, appropriate positioning to the market of payments (Chakravorti 2003) and selection of right portfolio mix of payment instruments and methods. Tactical decisions refer to selection of appropriate marketing mix and operational issues of payment systems. From past experience we may notice that payments innovations are not always adopted from consumers in retail payment market (Chakravorti 2000; Plouffe et al 2000). Particularities of payments market like network effects (McAndrews 1998; Chakravorti and Roson 2004) impose new needs in decision support. Traditional decision support in the domain is based mostly on historical data analysis and marketing research, where patterns of consumer’s behavior are extracted, used to forecast adoption of payment innovations. We propose agent based modeling and simulation as a new approach to decision support in the payments domain, which will assist banks in both strategic and tactical decisions concerning existing payment systems or planned ones. Our effort is towards a decision support system based on multi-agent modeling and simulation (MABS-DSS), which is going to be integrated into an existing payment infrastructure of a bank or payment service provider. Actors of the domain are modeled as agents in a multi-agent platform and simulations of the market are executed. A core DSS module is responsible for setting the agent’s behavior rules and coordinating the market simulation scenarios. It provides a friendly user interface for the decision maker and is responsible for the integration with the payment infrastructure. Experts’ knowledge, past
242
IADIS Virtual Multi Conference on Computer Science and Information Systems 2006
payment data and marketing research data are contributing in the formation of the model and calibration of simulations, in order to be validated against real data. This paper’s aim is to present an overview of the proposed architecture and relevant technology selection, of our work in progress. Though the MABS-DSS is based on payments market and especially banks, it may be extended to domains with similar characteristics. In the remainder of the paper we provide some background and related work in the usage of agent based simulation in decision aid, introduce the proposed architecture and technology, and conclude with a discussion of future directions of this work.
2. BACKGROUND AND RELEVANT WORK 2.1 Agent Based Simulation Agent based modeling and simulation has been lately used in various scientific areas that study complex systems (Jennings and Wooldridge 1998). The term simulation means the process of building a computerbased experiment written in a computer language. “Simulation is the process of designing a model of a real system and conducting experiments with this model for the purpose of understanding the behaviour of the system and/or evaluating various strategies for the operation of the system.” The initiation of simulation is a target system from real life that provides re-searcher some preliminary empirical data and findings, enough to build an abstraction or hypothesis that reflects the behaviour of the real system. Abstraction is not always complete or valid, so according to scientific method an experiment should be executed to validate abstraction against reality. In cases where lab experiments are feasible the validation process is based on experimental findings. However there are research areas such as social or economic domains, where lab experiments are not feasible. For such domains simulation provides a useful method for validation against empirical data. A simulation model is a computer program that is constructed based on the abstraction, and executed in the context of a computer environment in order to support or not the hypotheses of the abstraction. Validation refers to the methodology of testing simulation results with empirical data in order to validate the abstraction hypothesis of simulation (Boero and Flaminio 2004). Epistemological questions regarding the validity of simulation against lab experiments as a method for scientific knowledge have been raised during last years where agents have been used extensively. However, in social domain and decision support processes simulation provides a valid method for understanding macro patterns of the target domain that emerge from micro interactions between individuals (Boero and Flaminio 2004). Jager (2000, 2003) also argues towards using multi agent simulation as appropriate methodology for the study of social phenomena that is not possible to be studied in a lab, and proposes a meta-model of behaviour for developing agent rules that may be used to formulate agent behaviour in an agent-based simulation.
2.2 Works Using Agents in Decision Support Agent based modelling and simulation has been lately used in various scientific areas that study complex systems (Chakravorti 2003). The term simulation means the process of building a computer based experiment written in a computer language and conducting experiments with this model for the purpose of understanding the behaviour of the system and/or evaluating various strategies for the operation of the system. Several researchers incorporate agent technology for intelligent decision support in a target domain. Mainly, they use agent technology to model actors of the domain as intelligent agents and simulate the domain. Following we mention some works that use multi-agent modelling and simulations in the form of modelling a market segment and simulating actors’ behaviour, to assist decision making in relevant domains Agent technology is used in Matsatsinis et al (2003), an agent based system for decision making in product penetration strategy selection. It is based on MARKEX (Matsatsinis and Siskos 1999) and uses agent technology to replace human based tasks with agent based ones under the supervision of decision maker. Kyrylov and Bonanni (2004) develop an agent based simulation model for the telecommunications market as a strategy game to support decision-making. He models consumers and services’ providers as agents and simulates their interaction in a residential area in order to minimize human involvement in the game playing.
243
ISBN: 972-8924-13-5 © 2006 IADIS
Robertson models a network of banks and consumers in a multi-agent based simulation (Robertson 2003). The work focuses on bank strategic decision support through the study of agents’ interactions that come out from the simulations. Schwaiger proposes SimMarket (Schwaiger and Stahmer 2003), a multi-agent system that models a supermarket and simulates consumer behaviour supporting decision-making in this market. Ben Said et al propose CUBES (Ben Said et al 2001), a framework that integrates knowledge from several different research areas concerning consumer behaviour such as marketing, sociology, psychology and economics, and a multi-agent simulation of consumers in order to study market phenomena. SIMSEG project (Buchta and Mazanec 2001) utilizes agent technology in an artificial consumer market environment to support market positioning and segmentation decisions. Finally, Wohltorf proposes a DSS for the study of 3G mobile services’ introduction in the mobile communication services market (Wohltorf and Albayrak 2003). He models actors of the domain as agents based on previous analysis of the domain and simulates the market providing support for business model evaluation. Our proposed DSS differentiates by focusing in the payments market, which is characterized by network effects and strong competition, and by using MABS as a core component of the DSS. Our effort is towards developing an intelligent tool for decision makers to analyze the market and adopt appropriate decisions regarding payment systems.
3. PROPOSED ARCHITECTURE Our methodological approach for building the MABS component is similar to Niehaves’ (2004) that defines a flow towards developing a simulation. We refined it by combining concepts from Drogoul et al (2002), who propose a methodology for multi-agent simulation creation that is based on collaboration between roles Initially, the problem from a real domain should be defined and an extensive analysis of the domain should be executed. This is necessary in order to check if simulation is appropriate for the domain and to define the subset of empirical data that will be used in validation. In the next step we move to the simulation environment and start building the abstraction of the real target system. Next, we build the model, run the simulation and obtain the set of results data. Analysis of data and validation towards empirical data is the next step. If the model is acceptable the model is deployed. The proposed system and architecture focuses on payments market where banks and payment service providers operate. It is built on top of the payment infrastructure of such large actors and especially banks. A generic bank’s payment infrastructure for retail payments is a heavy real time processing system that is consisted of the host environment, the acquiring layer and the channel layer. The host system is utilizing the core banking processes and is responsible for the entire payment functionality, the acquiring layer is responsible for approving or not the incoming payment requests, and the channel layer is handling the various payment channels such as Internet or mobile channels The proposed MABS-DSS architecture is distributed and multi-tiered in order to be able to capture all the necessary input from remote data sources, and integrate necessary modules from real payments infrastructure. It is consisted of three layers: data layer, application layer and presentation layer. Data layer is comprised of database, knowledge base and model base components. Database component of data layer contains real payments data as well as historical data derived from existing payment systems. It is a part of the payment system infrastructure and links the DSS with the real payments environment. Knowledge base component contains statistical and marketing data, mostly derived from marketing research and experts’ knowledge, by using statistical software. Model base component contains models for the initiation of agents’ behaviour, derived from the knowledge base component (Figure 1). Since MABS-DSS is in close relation with payments infrastructure, specific limitations were considered. The most important one was that MABS-DSS operation should not in any way affect bank’s payment infrastructure performance, which is heavy used in real time. The databases of MABS-DSS that query data from host environment are scheduled to run the tasks in non-stress hours and use appropriate optimum design to reduce overhead on host. Layered structure was selected in order to provide operational benefits, since it may easily be integrated into existing infrastructure. In addition, technology was selected according to compatibility and performance criteria.
244
IADIS Virtual Multi Conference on Computer Science and Information Systems 2006
MABS-DSS architecture
Presentation layer
Client
MAS Engine
Application layer
Data layer
Core DSS Module
Model Base
Database
Knowledge Base
Integration with payments infrastructure Host Environment
Figure 1: MABS-DSS overall architecture
Technology selection was based on specific requirements imposed by the payments infrastructure. Major issues were compatibility with existing platforms and applications, almost zero impact on payment systems’ real time performance, ability to use existing hardware and operating environment. For the above reasons we selected Java as the platform for the MABS-DSS, and Oracle RDBMS for the databases components. More detailed, in order to be easily integrated into existing banks’ infrastructure and be web deployed, Java 2Enterprize Edition (J2EE) was selected for the building of the DSS module. Java 2Enterprise Edition is an architecture for building server-side deployments in the Java programming language and may be used to build traditional web sites, software components, or packaged applications. For the MAS engine and simulation development Java/SWARM platform was selected. Swarm is a Java library which implements agent-based technology and can be used to model multi-agent simulation of complex systems. Oracle RDBMS was selected for the databases and JDBC interface is used for connection between core DSS module and databases. Finally, XML interface is used between clients and DSS module. Selected technology provides a flexible solution to MABS-DSS architecture that may be easily integrated into existing payment infrastructure without impact in the overall performance. The system focuses on assisting decision makers by providing them with appropriate result for strategic and tactical decisions concerning payment systems. This task is accomplished from the core DSS module in connection with the MAS engine that produces simulations of the market according to scenarios defined by the decision makers. Both MAS engine and core DSS module incorporate knowledge derived from various sources and stored in the databases components. More detailed, the databases contain: 1. Market data from marketing research processed through statistical software and derived indicators for existing or planned payment systems. 2. Experts’ knowledge and critical factors derived form experts’ interviews. 3. Decision and adoption models contributed from social research. 4. Real payments data, in various formats coming from bank’s payments infrastructure. All the above contribute, in agent behaviour modeling by defining agent rules that affect simulation results.
4. CONCLUSION This paper demonstrates an initial architecture for a decision support system based on multi-agent based simulation for the domain of digital payments. We are in the initial steps of implementing the system, refining the architecture and model, in collaboration with experts from banking sector. However, multi-agent modeling and simulation integration into a DSS proves to be a valid approach for decision support in
245
ISBN: 972-8924-13-5 © 2006 IADIS
domains characterized by network effects. Future work includes full development of the system, refinement of the architecture and calibration of the MABS engine based on empirical data.
REFERENCES Balázs, Csik., 2003, Simulation of competitive market situations using intelligent agents, Per. Pol. Soc. Man. Sci., 11/1, 83-93 Ben Said, L., A. Drogoul. and T. Bouron., 2001, “Multi-Agent Based Simulation of Consumer Behaviour: Towards a New Marketing Approach”, MODSIM'2001. Boero, R. and S. Flaminio. 2004. “Empirical Validations of Agent-Based Models in Social Science: The Reality of Modelling and the M of Reality.”, EPOS. Koblenz, Germany. Buchta, C. and J. Mazanec. 2001. “SIMSEG/ACM – A Simulation Environment for Artificial Consumer Markets.” Working Paper Nr. 79. Chakravorti, S. 2000. “Why Has Stored Value Not Caught On?”, Federal Reserve Bank of Chicago, Emerging Issues Series, Working Paper Number S&R 00-6. Chakravorti, S. and E. Kobor. 2003. “Why Invest in Payment Innovations?”, Federal Reserve Bank of Chicago Pubilc Policy Series, EPS-2003-1B. Chakravorti, S. and R. Roson. 2004. “Platform competition in two-sided markets: the case of payment networks.” Working Paper Series WP-04-09, Federal Reserve Bank of Chicago. Drogoul, A. D. Vanbergue. and T. Meurisse. 2002. “Multi-Agent Based Simulation: Where are the Agents ?.” Proceedings of MABS'02, Multi-Agent Based Simulation, Bologna, Italy, LNCS, Springer-Verlag. Gilbert, N. and K. G. Troitzsch. 1999. “Simulation for the Social Scientist.” Open University Press. Jager, W. 2000. “Modeling Consumer Behaviour.” Doctoral thesis, Centre for Environmental and Traffic Psychology, University of Groningen, Groningen. Jager, W. and M. A. Jansen. 2003. “Diffusion Processes in Demographic Transitions: A Prospect on Using Multi Agent Simulation to Explore the Role of Cognitive Strategies and Social Interactions.” In: F.C. Billari & A. Prskawetz (Eds.) Agent Based Computational Demography: Using Simulation to Improve Our Understanding of Demographic Behaviour. Contributions to Economics Series. Springer-Verlag. Jennings, N. R., and M. J. Wooldridge. (Eds) 1998. “Agent Technology: Foundations, Applications, and Markets”, Springer publishing house. Kyrylov, V. and C. Bonanni. 2004. “Modeling Decision Making by Telecommunications Ser-vices Providers in a Strategy Market Game.” In Proceedings of the 2004 Applied Telecommunications Symposium. Markose, S. M. and Jia Loke Yiing. 2000. “Changing trends in payment systems for selected G10 and EU countries.” 1990-1998, Published in International Correspondent Banking Review Yearbook 2000/2001. Matsatsinis, N. P. Moraitis, P. Psomatakis V; and N. Spanoudakis. 2003. “An agent-based system for products penetration strategy selection.” Applied Artificial Intelligence, v17, i10, p901. Matsatsinis, N. F. and Y. Siskos. 1999. “ΜARKEX: An intelligent decision support system for product development decisions.” European Journal of Operational Research 113(2):336–354. McAndrews, J. J. 1998. “Network issues and payment systems”, Credit & Financial Management Review. Niehaves, B. 2004. “Using an Epistemological Framework For Analyzing the Assumptions of Simulation Models.” Workshop Epistemological Perspectives on Simulation EPOS. Koblenz, Germany. Plouffe, C. R. Vandenbosch, M. and Hulland, J. 2000. “Why Smart Cards Have Failed: Looking to Consumer and Merchant Reactions to a New Payment Technology.” International Journal of Bank Marketing, 18:112-123. Robertson, D. A. 2003. “Agent-Based Models of a Banking Network as an Example of a Turbulent Environment: the Deliberate vs. Emergent Strategy Debate Revisited”, Emergence: A Journal of Complexity in Organizations and Management, 5(2), 56-71. Schwaiger, A. and B. Stahmer. 2003. “SimMarket: Multiagent-based customer simulation and decision support for category management”, LECT NOTES ARTIF INT 2831: 74-84. Wohltorf, J. and S. Albayrak. 2003. “An Agent-based Decision Support System for the Introduction of Next Generation Mobile Services”, DAI-Labor, TU Berlin, Working paper.
246
