Agent-Based Modeling’s Open Methodology Approach:Simulation, Reflexivity, and Abduction

Please cite the paper as:
John B. Davis, (2017), Agent-Based Modeling’s Open Methodology Approach:Simulation, Reflexivity, and Abduction, World Economics Association (WEA) Conferences, No. 2 2017, Economic Philosophy: Complexities in Economics, 2nd October to 7th December 2017


This paper argues that agent-based modeling’s innovations in method developed in terms of simulation techniques also involve an innovation in economic methodology. It shows how Epstein’s generative science conception departs from conventional methodological reasoning, and employs what I term an open rather than closed approach to economic methodology associated with the roles that reflexivity, counterfactual reasoning, and abduction play in ABM. Central to this idea is that improvements in how we know something, a matter of method, determine whether we know something, a matter of methodology. The paper links this alternative view of economics and economic methodology to a social science model of economics and contrasts this with standard economics’ natural science model of economics. The paper discusses what this methodological understanding implies about the concept of emergence.

Keywords: , , , , ,

Recent comments


5 comment

  • Frederico Botafogo says:

    Prof. Davis, I read your paper with great interest. Allow me the following comments:

    1. You introduce the dichotomy opposing the natural versus the social approaches to modelling science. I wish to note that even in the natural realm there are instances where the development of scientific knowledge can alter the principles or laws that explain how the world works. I am thinking of meteorology, the study of earthquakes, or the study of volcanoes. The reason why this is so is not because nature’s behaviour would change further to our scientific progress, obviously, but rather because we do not have at this stage a clear understanding how to frame appropriate principles or laws governing the weather or the volcanoes.

    I am not sure of what follows, but it seems to me the foregoing view is consistent with one of your final arguments, namely that epistemology should be segregated from ontology. At the end of the day, the term science is applicable to propositions that we deem relevant when explaining the world. If we consider we can ‘explain’ the weather by means of simulations, then that qualifies as science even though we may not be able to predict when a hurricane will start or what path it will follow.

    2. I am the author of the working paper which follows yours. Therein I claim that accounting provides a procedure to frame economic data such that this is done in ways consistent with the complexity view of economics. Indeed, the recording of transactions by means of double-entry bookkeeping (DEB) is a procedure that is outside both the inductive and deductive approaches to reality. Accountants use DEB as a process and a practice when framing data, particularly cost assignments. For example, there are several ways to allocate cost, say by volume or based on activities; whatever the case, costs satisfy the DEB requirement. This fits well with your saying that ‘[i]t is important to note, then, what is distinctive about reasoning in reflexivity terms in this type of scientific explanation process. [Cost allocation] methods construct artificial worlds that [represent] real worlds, and consequently they compare what could be the case in terms of how that artificial world is microspecified with what could be the case in resulting macrostructural terms were the real world to closely resemble the artificial one. That is, reasoning in reflexivity terms involves counterfactual reasoning, …’ Appreciate that costs allocations have an impact on product profitability and the viability of profit centres; thus, they affect how managers decide what and how much to produce. In other words, the accounting procedure being chosen by accountants changes the world.

    3. I now consider the idea of equilibrium as a means for analytical tractability. I trust the notation I am introducing can make an impactful contribution to complexity in economics. Agents are represented in my framework by means of linear operators, not by means of utility or production functions. Thus, the whole approach of general equilibrium is no longer applicable. Instead of equilibrium, along with all the mathematical apparatus that goes with it, the constraint in the economic system now being applicable is that each and every agent’s balance sheet must balance.

    I cannot elaborate on this here (this is simply a commentary) but appreciate that all agents in the economic system can be required to satisfy the condition that their assets’ values equal their liabilities’. Further, in every transaction involving two parties, to one’s asset corresponds the other’s liability. This creates a systemic constraint. In a setting of perfect information, I claim equilibrium can be deducted from this requirement. In a setting of uncertainty, this provides the minimal condition for running simulations.

    Incidentally, I am unaware of complexity simulations that impose the DEB constraint.

    To conclude this third comment, notice that what-if questions can be addressed by using ERPs systems to run alternative scenarios, say using different allocation procedures to check the impact on the profitability of different products and profit centres and subsequently to consider alternative strategies. Accounting managers, however, still have a lot to develop for the sake of advancing those simulations.

    4. My final comment concerns your final paragraph in section 4:
    ‘Of course, it could still be the case … that there exist deep underlying principles and laws governing behavior in economic life …’ Although economists do not seem to be aware, DEB has been around for 500 years and thus provide one stable constraint to framing economic data. To the least, economists should be giving a bit more attention to the underlying economic meaning to be associated with DEB.

    5. I noticed a few typos in your manuscript: (i) on page 5, one sentence prior to the last one, the word ‘indeed’ is repeated twice; (ii) on page 8, on the 12th line of the 2nd paragraph, I suggest that you avoid repeating the word ‘emphasize’; (iii) on page 8, 3rd sentence counting backwards from the end, I don’t think the words ‘that is’ between commas should be there; (iv) on page 10, 3rd line, you are missing the article ‘the’ before ‘world works’; (v) on page 15, 2nd paragraph, 4th line, you should replace the first ‘and’ for a ‘the’; (vi) on footnote 9, you should invert the order from ‘the how’ to ‘how the’.

    Yours truly, Frederico Botafogo

  • Yoshinori Shiozawa says:

    Davis forgets to argue at least three big points: (1) question of time scale with regards to Agent-Based Simulation (ABS), (2) the new possibilities opened by ABS, (3) the nature of knowledge we can get from ABS.

    My comment should be short. I explain only the basic points of the questions. All points are treated in my paper: A Guided Tour of the Backside of Agent-Based Simulation, Chapter 1 in Kita, Taniguchi, and Nakajima (Eds.) Realistic Simulation of Financial Markets: Analyzing Market Behavior by the Third Mode of Science, Tokyo, Springer, 2016.

    (1) Question of time scale
    ABS make a part of computer simulation. It opens a new possibility to sciences a new possibility that is comparable to theory and experimentation. Theoretical arguments started in Classic Greek and Alexandrian ages. Major mode of theory making was deep speculation (careful observations and considering). Logic and mathematics was the major tools of verification. Experimentation began at the time of alchemy and became indispensable tool for modern science. It is important to note that it took many centuries until it became an established mode of scientific investigation. Simulation (and ABS in the case of economics) is the third mode of science. We should keep this fact in mind. If not, we may go astray in a jungle of excess information.

    For economics, we can situate SBS in a bit shorter time span. We may think that ABS is at the confluent point of two major currents. Economics started in the 18th century as natural history of social affairs. Conceptual and theoretical reasoning started roughly at the time of Adam Smith and became logical science at the time of Ricardo. Mathematics came to be acknowledged as a tool of economic research at the beginning of the 19th century (e.g. N.S. Canard 1801 Principes d’Economie Politique) but it took at least one century until it became for economics really useful tool of analysis. ABS started at the second half of the 20th century and lies in a burgeoning stage. We have to develop this mode of economics to a stage which is comparable as conceptual (or literal) and mathematical reasoning.

    Davis uses triad of deduction, induction and abduction in his praise of ABS but his argument is not very correct, because even in literal and mathematical analysis, abduction is important part of investigations and not a monopoly of simulation.

    (2) New possibilities opened by ABS
    Davis is not well aware of difference between Computable General Equilibrium (CGE) models and Agent-based Computational Economics (ACE). CGE is a mere application of computer powers to the old scheme of economics of the 20th century. ACE or ABS opens a new field of investigations which were hitherto impossible for economic analysis. Equilibrium was practically the unique possible situation that mathematical analysis can produce a useful knowledge. Differential dynamical system was much less successful than general equilibrium analysis. However, the giants of mathematical economics such as K. Arrow and F. Hahn knew well that economics should be sequential in a true sense. (See Hahn, Equilibrium and Macroeconomics, 1984, p.53, Hahn used the expression “sequential in an essential way”). Importance of sequential analysis was acknowledged early in 1920’s but it was forgotten by the success of equilibrium analysis. (See Kohn, Monetary analysis, the equilibrium method, and Keynes’s General Theory, JPE, 1986)

    ABS and ACE provides a new powerful method to realize truly sequential analysis of economic processes. Davis talks much about emergence and reflexivity but it is necessary to situate new possibility in the history of economics and try to see the new possibilities for the future. Emergence and reflexivity are results we may get from sequential analysis but they are not the target of our research.

    (3) Nature of knowledge obtainable by ABS
    The status of knowledge obtained by simulation is quite ambiguous. People who work in computer simulation often give warns to themselves: Avoid GIGO or Garbage In, Garbage Out. If one knows how to program ABS, it is easy to get some results. One may write a paper on them. But, the knowledge we get from those simulations is quite ambiguous. In ABS, it is easy to implements large number of parameters ad it is often hard to find which of parameters or factors operated in the results.

    Even there are many deficiencies in ABS, I strongly support ABS, because I believe that we are still at the initial stage of ABS. Experimentation became an indispensable mode of research for the modern physics but it required many centuries before we know criteria to adopt the result of an experiment and consider it an established knowledge. Philosophers and methodologists must argue the nature of knowledge we can get from ABS. Only through these long arguments, we can get plausible procedures in order to consider simulation results as a firm knowledge.

  • João Victor Souza da Silva says:

    You assume the capacity of ABM/ACE Models to operate the Economics as social science, complex and non-linear, while the mainstream equilibrium models assumes the scenarios similar to classic physics models. The argument basis to equilibrium models is in Walras and consequently Smith, with the “invisible hand”, but the duality of the agent and the desequilibrium arising from thechnological intesification from division of markets, makes it possible to understand the invisible hand not as a point of equilibrium, but as an emerging process in a “chaotic scenario” of economic growth. Do you believe in the feasibility of simulating a process of structural change in economic growth, based on Smith’s theory of markets, from the complex perspective of the “invisible hand” from these complex computational models?

    I really apreciate your papper.

  • Ping Chen says:

    Davis paper is one of the best at this forum, so that we can further discuss the achievement and limitations of ABS approach. I agree with Davis that computer simulation is a promising learning tool in studying non-linear causality, which is fundamentally different from neoclassical economics. Even we can construct nonlinear model in macro dynamics, we are still hard to compare it with real world macro policy, since theoretical modeling can only deal with few variables in nonlinear dynamics. That is why deterministic simulation in system dynamics and stochastic simulation in ABS/CAS can play a supplementary role in policy studies. However, it is too early to assert that simulation model has more power than fundamental laws than natural science. At this stage, ABS/CAS may generate some interesting PATTERNS that may similar to some social phenomena under some CONDITION, but yet to discover or challenge any fundamental laws such as DEMAND-SUPPLY EQUILIBRIUM or MARGINAL PRICING in economics.
    Let me discuss one example. Davis mentioned the LAW OF SUPPLY AND DEMAND in microeconomics. But he only assumes a linear relation between price and quantity in conventional textbooks. As Becker and Stiglitz observed that both demand and supply curve can be S-SHAPED. Its implied MULTIPLE EQUILIBRIUMS including HERD BEHAVIOR, PERSISTENT POVERTY, and MODERN DISEASE, so that Efficient MARKET hypothesis breaks down along with RATIONAL EXPECTATIONS.
    We can prove that S-shape demand curve could emerge with nonlinear interactions by means of statistical mechanics. The main difference between physics and economics is that there is no definition of “SOCIAL TEMPERATURE” that corresponds to THERMAL EQUILIBRIUM in equilibrium statistical mechanics. That is why statistical mechanics cannot be directly applied to economics. We find out that the intensity of individual interaction can replace social temperature in describing polarized distribution in ISING MODEL, and S-shaped demand curve does exist in collective phenomena such as fashion and imitation in learning. See Chen, Ping. “Imitation, Learning, and Communication: Central or Polarized Patterns in Collective Actions,” in A. Babloyantz ed., Self-Organization, Emerging Properties and Learning, pp. 279-286, Plenum, New York (1991). Also, in P. Chen, Economic Complexity (2010). I suggest that computer simulation based on BAS/CAS could verify this NONLINEAR DEMAND CURVE.
    At this stage, I think complexity economists are eager to embrace new ideas from physics and complexity science, such as EMERGENCE and PHASE TRANSITION, but far to short to apply these ideas in challenging mainstream economic thinking. I will raise some problems for CAS or ABS people for future study.
    RBC model in macroeconomics is the computational simulation based on equilibrium framework in mainstream economics. It is widely used by FED and European central bank in conducting monetary policy before and after financial crisis. Numerically, you can CALIBRATE any parameter or add more variables to fit empirical data, just like econometric models, but fail to identify the source and events of financial crisis. RBC school argue that their model is based on DSGE theory while econometrics model is routine regression without economic theory. Can you guys do better than RBC/DSGE and macro econometrics approach in macro policy?
    Here, we strongly argue that natural science including physics law can be applied to economics as the FIRST PRINCIPLE in model selection. Why? Because Lucas model of monetary neutrality and RBC model are all belong to Frisch model of noise-driven cycles. Frisch model is a perpetual motion machine in nature. Its physics version is called Brownian Oscillator in statistical mechanics. Frisch only made a claim in 1930 but never published his promised paper in Econometrica. Frisch’s claim was rejected by physicists in 1930, three years earlier than Frish. Strangely, Frisch model share the FIRST NOBEL PRIZE in ECONOMICS, plus Lucas, RBC, and Fama, all Nobel Prize winning models were based on Frisch model of noise-driven cycles. We know living organism is based on organized solar energy that is not random energy from heat environment. Where is the energy source for persistent random shocks that could generate PERSISTENT BUSINESS CYCLES in large economy such as US? see. Ping Chen, “The Frisch Model of Business Cycles – a Spurious Doctrine, but a Mysterious Success” (1999), Chapter 12, in P. Chen, Economic Complexity and Equilibrium Illusion: Essays on Market Instability and Macro Vitality, London: Routledge (2010), pp.239-250.
    We can apply non-stationary time series analysis in terms of TIME-FREQUENCY ANALYSIS to directly observe TIME-VARYING BUSINESS CYCLES from macro and finance indexes, which is in a range of 2-10 years with average period of 4-5 years. Its correlation dimension is more than 2, and 70 % of cyclic variance can be explained by nonlinear strange attractor of COLOR CHAOS. We reach similar observation of NBER business cycles with RBC model by means of HP filter, but had complete different theory framework, since our theory have multi-regimes of CLAM (an approximation of linear efficient market) and TURBULENT MARKET (without equilibrium/rational expectations), while DSGE model excludes the possibility of endogenous instability and nonlinear cycles. So far, ABS/CAS approach has yet to challenge mainstream macroeconomic policy while we expect them can do so in near future.
    Game theory is another field that complexity economics may have impact but yet to produce significant result.
    Prigogine’s core idea about evolution is BREAKDOWN OF TIME SYMMETRY, i.e. irreversible process implies TIME ARROW. In real world, all kinds of non-equilibrium phenomena can be characterized by SYMMETRY BREAKING. For example, Hayek’s idea of ROUND-ABOUT PRODUCTION implies asymmetry in life cycle between CONSUMPTION GOODS and INTERMEDIATE GOODS. S-shaped demand and supply curve implies asymmetry between demand and supply in micro. Behavioral economics implies asymmetry between gain and loss in psychology and risk attitude in finance. Power and poverty caused by asymmetrical trade in division of labor and unequal society. However, all game theory model has symmetric setting in rules and choices. Can we believe the existing game theory model including NASH-EQUILIBRIUM is useful for studying nuclear arm race or trade war in the real world?
    The main difficulty in social science is that we cannot conduct large scale and long time process in lab experiments like control experiment in natural science. However, physics also study cosmology and climate dynamics based on physics law of classical mechanics and thermodynamics. Physicists do use computer simulation to narrow choices in model specification. But we do not believe computer simulation could replace basic laws such as CONSERVATION OF ENERGY, THERMODYNAMICS, and UNCERTAINTY PRINCIPLE in quantum mechanics, which are the very foundation of Simon’s BOUNDED RATIONALITY. So-called rational expectations and efficient market simply assume market mechanism is a giant computer with infinite speed and infinite memory, that is impossible since it needs infinite energy flow. Friedman’s positive economics is not valid since prediction cannot test competing econometric models if time series are non-stationary under non-equilibrium conditions.
    In short, ABS/CAS did provide useful tools in computer simulation for economic research, but yet to challenge theoretical foundation and policy making in mainstream economics. I wish complexity scientists may better cooperate with real world economists in addressing contemporary issues, such as financial crisis, budget deficits, trade war, persistent poverty, and infrastructure investment. We need to bring physics idea of emgergence with specific economics issues, such as emergence of industrial cluster, urban-rural structure, military alliance, and answer difficult questions, such as why giant and small business could exist in some industry but not in other industries, and why there are different types of ownership and debt structure that may change over different historical period.
    We may use philosophical insights as VISION of new economic research but not as DEFENSE for limitations of emerging approach. The success or failure in social theory mainly depend on problem-solving rather than methodological debate. Only after our success in solving contemporary problems, we may have better chance to convince the next generation of economists that our complexity/evolutionary approach DOES BETTER than equilibrium/optimization approach in macro policy and micro management.

    Ping Chen, China Institute, Fudan University, Shanghai, China

  • Ping Chen says:

    Dear Editor, there is a time difference between Europe and US. I just send out my comments at 19:11 on Nov. 30, 2017 at US central time in Texas, but appeared in your DISCUSSION FORUM as 1:11 on Dec. 1, 2017 (where?). Can you extend your time permission for this FORUM? This is an interesting example of multiple time scale in evolutionary economics.