The recent financial crisis sparked a debate on the usefulness of the neoclassical economic paradigm to understand and predict financial crises. The depression of the thirties was also a time of re-thinking macroeconomic modelling and marked the rise of Keynesianism in economics. Bookstaber’s The End of Theory is a frontal attack on the usefulness of the current economic paradigm based on market efficiency and rational expectations to predict financial crises. He argues for a radical change in economic modelling with agent-based models (ABM) replacing the current neoclassical model and its many versions.
To their critics, standard economic models represent over-simplifications of reality, the result of the assumption that the economy is populated by fully rational agents who have full information and perfect knowledge of the characteristics of the economy. Agents do not interact except through markets or affect their environment. The economy and financial markets are stable, tending to return to equilibrium after a perturbation. An external shock could create a boom or a bust but the economic system is self-correcting, always returning to the steady state. Even Dynamic Stochastic Equilibrium Models (DSGE), like the model used by the ECB, that incorporate frictions that could slow adjustment to exogenous shocks did nor foresee the 2008 crisis nor its implications for the economy and the financial system.
How badly did mainstream economics do during the 2008 crises? To critics, such as prominent economists Krugman, Stiglitz and Shiller, standard economic models based on rational expectations and market clearing did not foresee the crisis and when the crisis erupted they disagreed on the appropriate policy response. Not all economists agree that economic models failed during the crisis. Lucas, a central figure in the development of new classical economics, in an article published in the Economist titled ‘In defence of the dismal science’ rebuts the criticisms and provides a spirited defence of mainstream economics and finance theories.
Bookstaber’s book provides ‘a nontechnical introduction to agent-based modelling, an alternative to neoclassical economics that shows great promise in predicting crises, averting them, and thus helping us recover from them’. ABMs are the response to a world that is characterized by emergent phenomena, non-ergodicity, radical uncertainty and computational irreducibility (the four horsemen of the econopalypse).
Chapters three to seven (section 2 of the book) provide an intuitive introduction to the four horsemen using examples from physics, biology and literacy. Emergent phenomena are the result of interactions of individually sensible agents who in aggregate might produce complex and non-linear behaviour. Stampedes in Mecca, a school of fish, the flight of a flock of birds and traffic are good examples of how agents operating under simple rules can create very complex group dynamics—emergent phenomena. Modeling such phenomena, once we know the simple rules agents use, is only possible using computer simulation.
If a process is ergodic, data samples from the past are equivalent to data samples from the future. For an ergodic process history does not matter. According to Samuelson the ergodic hypothesis is the sine qua non of scientific investigation in economics. If economic variables were stationary ergodic, then historical data could be regarded as observations from the same distribution and could be used to calculate probabilities from the distribution if the sample was sufficiently long. Bookstaber argues that economic processes are non-ergodic processes because interactions change us and context matters.
In a world of radical uncertainty, better known in finance as Knightian uncertainty, there is no basis that could be used to calculate probabilities.
How do economic agents make decisions in a world of radical uncertainty? Would they be consistent and rational as standard economic models assume or would they use simple decision-making rules, heuristics? Bookstaber presents many examples from different species where in a world of radical uncertainty decision-making rules will ignore available information and be suboptimal. Simple rules of thumb, heuristics, are coarse but robust to radical uncertainty.
Computationally, irreducibility means that the only way to solve a problem and determine the future outcome is through simulation. Neoclassical economics assumes that the economy is a computationally reducible system described by mathematical formulas. Bookstaber argues that ‘computational irreducibility is the norm in real world dynamical systems’ and provides as examples the three-body problem in physics and Conways’s game of life. If the world is characterized by the four horsemen, then deductive mathematical reasoning, the basis of neoclassical economics, cannot be used to model the complexity created by interacting agents.
In section 3 of the book the author spells out the implications of computational irreducibility, emergent phenomena, non-ergodicity and radical uncertainty for the dominant economic models and in particular their ability to predict economic crises. If you agree with the above then you should refute: the use of mathematics, the assumption that all agents are rational utility maximisers, the notion of stable preferences and the ability to calculate probabilities about the future state of the world. In other words you should reject the current neoclassical economic model – the end of theory. And what do you put in its place? Agent-based models.
ABMs simulate the actions and interactions of individual participants of the financial system. The basic ingredients of an ABM are heterogenous agents who observe and react to changes in the environment and interact with each other. The world is an interconnected dynamic system of individual agents who use heuristics rather than optimization in a world of radical uncertainty. The result is a dynamic system producing dynamics which might be computationally irreducible. A financial crisis is the result of economic agents following simple rules but producing an aggregate chaotic behaviour.
ABMs provide a very flexible framework in which to study the behaviour of an economic system characterized by contagions, boom and busts, heterogenous agents using heuristics for decision-making, asymmetric information and other features that neoclassical economics find hard to explain and model. However, the flexibility has costs. ABMs reliance on simulation makes them complex and difficult to analyse. ABM modellers can choose from an infinite number of rules to characterize agent behaviour. That may provide many degrees of freedom but it also raises the issue of model falsifiability. Does the model ‘fit’ reality because of the particular choice of agent decision rules? A narrowing down of decision-making rules, perhaps along the lines suggested in behavioural economics, would certainly be beneficial in this respect. Another concern with ABMs is their stability to changes in assumptions. For example, are the model’s dynamics robust to the addition of new agents?
The last part of the book (section 4) applies the ideas of agent-based modelling to the financial system and the 2008 crisis. The author however makes clear at the start that ‘I am not going to be advocating for a specific model, laid out, parameterised and solved’. The reader would be clearly disappointed.
The agent-based model starts with a description of the financial agents who populate the financial system – where they operate, their environment, heuristics and dynamics. Figure 11 of the book shows a map of the financial system, the flow of funds, collateral and its main agents (players): bank/dealers, hedge funds, institutional investors and corporate treasuries. The system is complex and multilayer. The 2008 crisis exposed the fault lines of such a system: ‘the propagation of risk, the path a shock takes and the value of integration versus segregation of the functions of various agents or nodes, all have a different and richer nature as we move to a multilayer view of the financial system’. The blackout that happened during the White Night, Notte Biance, in Rome in 2003 provides an excellent example of ‘large-scale failure of a multilayer network’.
In the book’s conclusions the author makes clear the objectives of ABM: ‘We need agent-based economics to create a weather service for the financial markets. We need to forecast the tropical storms of financial dislocation. Will the storm turn into a hurricane? What path will take? How bad will things get?’. These are very ambitious goals.
The failure of current economic models to predict the 2008 crisis provided an impetus for a re-assessment of the dominant economic paradigm. Proponents of agent-based modelling are critical of the building blocks of the standard macroeconomic models and in particular the assumptions of a representative agent with rational expectations (agents know the underlying model of the economy and use all available information efficiently). Both criticisms are not new. Behavioural economics in particular is based on the assumption that agents are different and make inefficient use of information. Economists, since Freidman, argue that models should not be judged by the realism of their assumptions but on how accurately they predict the real world.
Not all economic research adopts the standard neoclassical assumptions, macroeconomic modelling has evolved considerably over time. In large part this is due to the empirical failure of the models, not to the plausibility of their assumptions. The second generation of DSGE models incorporates heterogeneous expectations, incomplete markets, habit persistence, sticky prices and wages and incomplete information. Inclusion of financial frictions in more recent DSGE models has improved significantly their ability to explain the behaviour of the economy after the 2008 crisis. More work needs to be done to incorporate simultaneously pricing, financial and labour frictions and more sophisticated and realistic models of the financial system. This represents a rich research program which could produce a better economic model. That it has not happened so far means that these models are not thought to be better than the model they are trying to replace. ABMs represent an alternative approach to economic modelling possibly better tailored to model complex systems.
Agent-based models are used widely in the social and natural sciences. There is less usage in economics. Cheap computing power, big data and the disappointing performance of mainstream economic modelling, could provide the necessary impetus for wider acceptance within the economic profession. If not, ABMs might still be a promising alternative to standard models in periods characterised by instability and complexity. Richard Bookstaber’s book is an excellent treatise, in simple non-mathematical terms, of an alternative approach to the neoclassical economic paradigm that should be read by risk managers and policy officials.
EDHEC Business School, London
© 2018, Nikolaos Tessaromatis
Additional information
Notes on contributors
Nikolaos Tessaromatis
Nikolaos Tessaromatis is a Professor of Finance at EDHEC Business School. Prior to joining EDHEC Business School Dr Tessaromatis was CEO and CIO of EDEKT Asset Management, the leading fiduciary manager of Greek pension funds, and Associate Professor of Finance at ALBA Graduate Business School. Before EDEKT, he was Principal and Head of Research and Product Development at Gartmore Investment Management, Associate Director at Nat-West Investment Management, Senior Quantitative Analyst at Hermes Investment Management and lecturer at Warwick Business School.