Computational Economic Models: Agent-based Computational Economic model
The modeling of statistically and analytically defined economic problems is often carried out using advanced computer softwares based on economic models. An example of such models is the Agent – based computational analysis model (ACEs), which considers economic processes as changing systems of agents that interact (Reid par. 9).This method can be used to study all the economic systems, people, social groupings, and physical systems, such as the transport system, and biological systems. The use of ACEs has gained tremendous development in the last few years due to the various advantages it offers over the conventional economic models. One of the advantages is the ability to consider entire systems’ evolution over time and space without interference.
Conventional computation in economics is used for analyzing empirical data and calculating equilibria in conventional economic models. The key tools used in computation are standard analytical and mathematical tools, which aid in the solution of non-linear programming problems and problems in optimization. ACE on the other hand, focuses on computer models for the solution of complex dynamic systems for analyzing alternative theories for economic behavior. Consequently, there has been increasing application of ACEs at the expense of the conventional computational methods (Judd 2).
ACE researchers are often concerned with understanding why some regularity in the global economics are observed to be changing in a decentralized manner despite the absence of structured top-down controls. They also aim at using ACE frameworks as computational laboratories for the study and testing of socio-economic structures, and their effects on both social welfare and individual behavior (Tesfatsion 2001 286).
The purpose of this paper is to determine the weaknesses of conventional models of economic analysis, and thus effectively describe the use of and advantages associated with the application of ACEs to the statistical and analytical description of economic systems. This is intended to be achieved through a study of relevant literature.
Limitations of Standard Models
The main reason why researchers shift their focus from conventional economic models to the use of ACEs can best be described as the dissatisfaction with the conventional methods. This is due to ineffectiveness of the methods in handling complex systems that are dynamic in nature, as they are fairly simplified. For instance, conventional methods base their analyses on theories, which are generally indefinite in scope. Consequently, simple assumptions have to be made to achieve substantive results. In making these assumptions, sometimes the elements that have to be sacrificed are of first order importance to the economic problems in question. On the other hand, assumptions may also be strict hence forcing out some important features of an economic system during analysis (Judd 3).
Secondly, conventional economic models are simplistic in the nature of their applications in that they engage simple models with the aim of finding the causes of some economic issues, or explaining arguments. This common framework is based on the tractability of the model in question. In so doing, the ability of systems to interact within a given economic framework is not taken into consideration. In essence, it ignores the possible multi-dimensionality in the solution to existing economic problems (Judd 5).
Moreover, conventional computational models tend to exaggerate the researcher’s foresight and rationality. This is because in application, assumptions made are subject to the researcher’s bias and thus affect the results obtained. Over simplification and outrageous assumptions may characterize the use of conventional computational methods. This bias is eliminated in the use of ACEs since the model uses already tested and confirmed mathematical and analytical tools for the confirmation of suggested theories (Varian 477).
In addition, the importance of heterogeneity may be understated since the rationality of the expectations and the choice of representative agents remain the simplest problems to be solved using the design computational models. Consequently, the variety in heterogeneous systems poses a significant problem in using conventional models for the solution of complex system. This finally leads to poor results based on poor strategies that are founded on simple problems.
It is also difficult to use standard computational economics techniques to study local interactions due to their generalization of interactions to the macro level. Most interactions that can be solved using conventional computational economics models are characterized by price interactions at the macro level of the economy. Thus, solving highly interactive systems cannot be achieved directly using these methods (Judd 5).
The implication of the weaknesses associated with conventional methods is that there are high chances of misguided decision-making based on a limited information scope. In addition to this, some aspects, which might be of paramount importance, are ignored in the obtaining of a solution to economic problems based on limited models. Because of these limitations, economists tend to shift in application of computational methods to the adoption of models, such as ACE that reduce the impacts of these shortcomings. For instance in finance, there has been increasing use of agent based computational finance, which is analogous to the agent based computational economics (LeBaron 686).
Advantages of ACEs
The ability of ACEs to eliminate the weaknesses of conventional computational models has made them to be used in the field of economics increasingly. Although ACES also have their own limitations, the associated advantages outperform the conventional methods. Some of the advantages of using ACEs are outlined below:
Model diversity – the foundational principles of Agent – based computational economics are applicable in a variety of systems. For instance, the financial systems have a similar model defined as agent-based computational finance, which is analogous to ACE. Besides this, the model also attracts the application to a variety of systems, such as individuals, entire physical systems, for instance, the transport or health system, and social networks. This diversity of ACE models makes them suitable to the solution of varied economic model problems of both statistical and analytical origin. ACEs are applied in biological systems, social systems, and business, network, and technology systems. In each case, the application differs depending on the local problem definition (Chen and others 2).
Heterogeneity – in the solution of single problems, the complexity often associated with economic systems seldom require oversimplification in order for ACEs to be applied. It is in the domain of ACEs to solve complex problems with a dynamic nature as opposed to the conventional methods, which mainly solve simple problems, most of which are static in nature. Intensive computations – it is possible to carry out intensive computations using ACEs. Consequently, the challenge of local interactions, which is normally associated with conventional methods is eliminated. Besides the global interactions, which can be handled by other economics computational models, ACEs also adapt to the solution of highly individualized problems that associate themselves with economic systems. For instance, while the issue of pricing may be a global challenge to the transport system, technological use may be highly individualized to the specific agent in question (Chen and others 3).
ACE also views economies as evolving systems in which agents play different roles. In application of ACE, it is thus possible to predict potential outcomes of varying actions with regards to specific agents. For instance, through computational modeling based on ACE, it is possible to predict the impact of extracting a single unit from a given transport system on the overall pricing strategy of the remaining agents. The dynamics involved in modeling and simulation using ACE are highly specialized, and work with the least amount of error (Tesfatsion 282).
The use of powerful computational tools – the tools used by ACE models are new and powerful enough to aid in the solution of the inherent economic problems, particularly those cited as requiring continued research. Some of the tools used in ACE modeling are based on object-oriented programming, which makes it possible to simulate a variety of complex and dynamic systems, a fete, which cannot be achieved through the use of standard computational procedures. This has made it possible for ACE models to base their computational analyses on several interacting agents as well as on single agents with high levels of local interactions. By computational means, economic worlds can be constructed with a population of heterogeneous agents whose interactions with the environment and with other agents are determined by internal behaviors, internalized social norms, and based on the agent experience. In comparison to conventionally modeled agents, these have higher cognitive strengths and levels of system autonomy (Wilmans 16).
ACE models are also dynamic as the systems, which are used to model – it is possible to model frequently changing systems using ACE models, which permit a wide range of agent interactions and behaviors. In this scenario, interactions and cooperative associations take a centre stage together with quantity and price relationships.
The behaviors of various agents are adapted to the constant agent – agent interactions as well as the agent-environment interactions. In such systems, the economic world undergoes a series of self-organizations based on the observed agent behaviors through the intricate interactions between agents and other agents as well as between agents and the environment. ACE models can effectively model and simulate these self-organizing interactions to enable effective planning and prediction of volatile market systems (Tesfatsion 282).
ACE models are participatory and can influence decision making through the inclusion of the studied agents into the simulation exercises. As opposed to conventional models, which are in no position to predict the outcomes of potential agent actions, ACE models provide the opportunity for researchers to engage the agents in their study. This makes it possible for studies to influence agents in their decision-making due to its predictive nature. For instance, still taking the case of the transport industry, it is possible for the participants in the transport industry to observe the potential effects of the withdrawal of one agent from the system, and to decide based on the simulated outcomes (Judd 6).
In addition to the already mentioned advantages, the use of ACEs also represents economic systems as subject to natural selection pressures. In this process, the agents are in a constant state of experimentation, with new rules guiding behavior in every instance. The behavior of the agents is induced through constant interactions hence leading to a co-evolution of studied agents.
ACE permits the growth of economic systems across a given time-line. This is because systems can be observed across time and space without influencing them in any way, that is, the observer is passive, and the systems act independently. The results of any simulation process are independent of the researcher’s bias since all operations are carried out by computational means. Consideration of the incessant system behavior is made while not trying to interfere with the system of study in any way. The modeler has the duty of setting only the initial conditions. The rest of the agent behaviors will be determined by interactions between agents (Tesfatsion 2).
It is also possible to align the ACE models to various models of different disciplines. This makes it possible to consider several factors in the statistical analyses associated with ACE modeling. The comparison of data across different fields also enables the computational intensity that is associated with ACE models to take root. In addition to this, it further eliminates the need for over simplification of issues and the use of too many assumptions. As stated previously, one limitation inherent with conventional models is their inability to achieve substantive solutions based on information constrained by the placed assumptions (Tesfatsion 283).
ACE is a methodology that incorporates concepts adopted from several disciplines and tools. Such tools include evolutionary economics, computer science, and cognitive science in such a way that permits developments like the positive grounding of theories associated with economics in the interaction and thinking of independent agents. Other developments include testing, and implementation of the studied theories through continuous computational experimentation, statistical analyses, and comparison of findings with those obtained from analytical studies, and the formulation and testing of the integrated socio-economic theories. The theories are compatible with data and theory obtained from relevant fields of social science (Varian 481).
Limitations of ACEs
Despite being very effective in doing away with the limitations associated with conventional methods, ACEs still have some limitations in their applications.
Critics of intensive computational analyses assert that one challenge of using computationally intensive research models is that they result in examples instead of theorems. Although this is true to a large extent, those examples aid in making quick decisions that can make or break the future. An example is in the future investment in the stock market, which can be made based on computational predictions. On the other hand, theorems examine a range of examples and make signature statements from them. In the applications of theorems, assumptions are often made in order to comply with the tractability demands associated with research work. In making these assumptions, several interesting phenomena may be missed. The assumptions take the form of functional specifications, such as demand curves of a linear nature, which are often observed in the oligopoly theory. They may also be qualitative in nature, such as the informational assumptions used in expectations analyses. The robustness and relevance of examples used in analysis is of more essence than the number of examples used since computational analysis allows one to make more complex and realistic assumptions that theories cannot touch (Marks and Vriend 117).
Another criticism comes in the argument that numerical solutions often have errors. In this assertion, theoretical researches often fail to consider the fact that theories also often pose challenges in the solution of complex problems that require in-depth analyses. By focusing on only tractable cases, theoretical models also result in ineffective solutions due to specification errors. While numerical errors can be reduced during computations, specification errors are more difficult to resolve. Consequently, in deciding whether to use ACE models or conventional models, researches are faced with the determination of a trade -off between solvable numerical errors and specification errors. In comparison to conventional methods, computationally intensive models provide an opportunity to obtain solutions to more realistic problems. It comes down to the question as to whether one should have the right answer to a wrong question or an erroneous answer to the right question (Wilmans 28).
It is also proposed that like other computational models, ACE models tend to offer only a few insights if any. However, this claim remains true only in the event that the applied model has been used in the solution of a problem based on a single set of initial conditions. This is understandable since a given set of conditions often results in a single set of solutions no matter how many times an analysis is carried out. To address this issue, sensitivity analysis is always carried out, where different sets of varying parameters are used to compile a range of outcomes that can be applied in predicting potential outcomes from an input variation (Judd 7).
Despite the possible reduction in the impact of limitations associated with ACEs, they remain vulnerable to critic due to changes in strategy without the change in preferences. In this argument, the use of ACEs in the solution of economic problems is still carried out with a foundation of principles that were initially used in the application of conventional methods. In other words, the use of ACE is the computation of standard procedures that result in the same outcomes as those of conventional methods only without the specification errors.
In addition to this, computational analyses exaggerate errors in individual decision making due to the high levels of accuracy associated with computational procedures. This makes researchers subject to high pressure levels with the need to be accurate being essential to performance.
Issues in ACE Model Use
The most pertinent issue in the application of computationally intensive models of research in economics lies in the difficulty involved in communicating results obtained from statistical analyses. In theoretical research, results are easily communicated through a statement of a theory. On the other hand, although it may be possible to give examples from research results, the information from computational analyses cannot be generalized into a theory (Chen and others 3). This makes it significantly hard to present results.
The concept of Computational Economic models is one of varying designs. The insurgence of a variety of computationally intensive models has seen the rise of ACE models as the most frequently applied in the field of economics. This frequent application has been prompted by the ability of ACEs to provide satisfaction in areas, such as concentration on tractable systems and the oversimplification of assumptions, which often lead to specification errors in conventional models. Besides this, ACE models have been discovered to offer various advantages, such as heterogeneity, diversity, and dynamic operations, which improve results obtained from ACE, based analyses. On the other hand, critics of ACE models blame them for lack of theoretical result presentation and for numerical errors. These are however insignificant considering that they also offer much needed insight and eliminate the occurrence of specification errors. The main issues related to the use of ACE models is therefore the difficulty involved in presenting results, and the fact that they are vulnerable to criticism due to their concentration on strategy change rather than the preference for results.
Chen, Shu-Heng, Chia-Ling Chang and Ye-Rong Du. Agent Based Economic Models and Econometrics. (n.d). Retrieved from http://www.bwl.uni-kiel.de/gwrp/econophysics/abstracts/chen.pdf
Judd, Kenneth. Computationally Intensive Analyses in Economics Handbook of Computational Economics 11. (2005). Retrieved from http://bucky.stanford.edu
LeBaron, Blake. Agent Based Computational Finance: Suggested Readings and Early Research. Journal of Economic Dynamics and Control 24 (2000): 679 – 702. Retrieved from http://www.eecs.harvard.edu/~parkes/cs286r/spring08/reading6/lebaron00agent.pdf
Marks, Robert and Nicolaas Vriend. The special issue: agent-based computational economics—overview. The Knowledge Engineering Review 27. 2(2012): 115-122. Retrieved from http://webspace.qmul.ac.uk/nvriend/pub/ker.pdf
Reid, Stuart. Introduction to agent-based computational economic models (2014). Retrieved from http://www.stuartreid.co.za/agent-based-computational-economic-models/
Tesfatsion, Leigh. Agent Based Computational Economics. Working Paper No. 1. 2003. Retrieved from http://www2.econ.iastate.edu/tesfatsi/acewp1.pdf
Tesfatsion, Leigh. Introduction to the special issue on agent-based computational economics. Journal of Economic Dynamics & Control 25 (2001): 281-293. Retrieved from http://econ2.econ.iastate.edu/tesfatsi/ACEIntroSpecialIssue.JEDC2001.LT.pdf
Varian, Hal. Computational Economics and Finance: Modeling and Analysis with Mathematica. The American Mathematical Monthly 105. 5(1998):477 – 481. Retrieved from http://www.jstor.org/discover/10.2307/3109830?sid=21105498364293&uid=3738336&uid=2&uid=2129&uid=70&uid=4
Wilmans, Jason. Application of Agent-Based Computational Economics models to the U.S. banking market. Thesis. Hamburg university of Applied Sciences. 2013. Retrieved from http://edoc.sub.uni-hamburg.de/haw/volltexte/2014/2211/pdf/BSc_JTW.pd