Agent-Based Modelling: A New Path Forward?
Standard economic models make assumptions that are routinely criticized as unrealistic and turn off many outsiders from taking economics seriously. Agent-based modelling improves on these models by refining the level of analysis down to heterogeneous individuals, thereby incorporating realistic behaviors and psychologies while also allowing for a high level of model complexity. Although there are still issues with the verisimilitude of the heuristics used and the tractability of the model as a whole, agent-based models present an opportunity to dramatically improve economists’ attempts to understand the economy.
Karl Marx famously wrote that “the tradition of all dead generations weighs like a nightmare on the brains of the living” (Marx, 1852). A similar statement can be said for macroeconomics education at the undergraduate level, at least at McGill. In the intermediate macroeconomics sequence, almost all of what we cover concerns academic tools developed before the 1990s, such as the basic IS-LM model and its variants, little of which is or was ever used by real economists (Hendry, 2020). Moreover, much of it rests on assumptions that would strike a casual observer first as tragedy, second as farce. Indeed, we ‘assume one interest rate’, ‘assume perfect information’, and, even worse than that, ‘assume a representative agent’. Luckily, there is a new method to modelling macroeconomic phenomena that does not rely on these ‘top-down’ models with abstruse assumptions: agent-based modelling (ABM). By refining the level of analysis to a set of heterogeneous, interactive individuals, ABM avoids the theoretical pitfalls of standard macroeconomic models while generally outperforming both their ability to explain what actually happens in an economy and, potentially, their quantitative predictive power.
Before continuing, it is pertinent to discuss what exactly ABM entails. On a general level, an ‘agent-based model’ contains ‘agents’ that interact with each other and the environment that they are placed in. While these agents can represent economic agents, such as firms, banks, central banks, workers, and consumers, they need not be (Wang et al., 2018). For instance, ABM in biology can have the ‘agents’ be members of a species. In economic models, the agents can be heterogeneous, such as in age, income, preferences, and so on (Axtell and Farmer, 2022). Moreover, these agents ‘interact’ through prescribed actions, such as buying or selling a stock, consuming heterogeneous goods, or even moving to a new neighborhood (Wang et al., 2018). These actions are made through various decision rules and heuristics founded on psychological experiments in behavioral economics.
It is clear that agent-based models can be quite flexible. But ABM is not just a theoretical option for empirical research, as it has been successfully used in many areas of economics. For instance, the Bank of England developed an agent-based model of the housing market and found that it accurately reproduced the loan to income distribution in the United Kingdom (Turrell, 2016). The Bank of England also found that their ABM approach to trading in corporate bonds accurately reproduced the distribution of daily log-price returns. Indeed, ABM has been used as a computational method in financial economics for at least two decades, such as in the creation of artificial stock markets (Chen et al., 2018)..
The successes of ABM have until recently been restricted to these kinds of recreations of ‘stylized facts’ in the economy. But ABM has made major breakthroughs in empirical forecasting in the past few years. One paper published at the Institute for New Economic Thinking at the University of Oxford, Wiese et al. (2024), used an agent-based model to predict “GDP, inflation, household consumption, government consumption, and investment” for 38 countries. Their model outperformed another forecasting-oriented agent-based model used in Poledna et al. (2023b), which reproduced the economy of Austria, and an AR(1) model. Their model is quite complex and granular: it contains a housing market, credit market, central bank, central government, government entities, households, individuals, and other ‘agents’. Given that ABM for forecasting is still in its infancy, these recent developments are promising for the field.
While all this is impressive, ABM does have a few drawbacks. Besides being currently focused on recreating stylized facts and only recently improving their predictive power, agent-based models are inherently computationally intensive and complex (Bonabeau, 2002). Moreover, their granular and behavioral nature may make them subject to the ‘Lucas Critique’, the idea that individuals may operate differently than they have been in response to new policies or structural changes (Turrell, 2016). The specific decision rules that agents use to interact in these models thus may change in reality in response to certain policies. As advancements in computer science and artificial intelligence increase, the former issue is likely to diminish (Bonebeau, 2002). Nonetheless, ABM still needs to be flexible enough to be as ‘Lucas Critique’-proof as possible, but still granular enough to take advantage of their capabilities for increased realism in economic modelling that leaves no stone unturned. ABM offers a promising new path to both better understand and thereby accurately predict the macro economy.
References
Axtell, R. and Farmer, J.. “Agent-Based Modeling in Economics and Finance: Past, Present, and Future”. INET Oxford Working Paper No. 2022-10. 21st June 2022. https://oms-inet.files.svdcdn.com/staging/files/JEL-v2.0.pdf
Bonabeau, Eric. “Agent-Based Modeling: Methods and Techniques for Simulating Human Systems.” Proceedings of the National Academy of Sciences of the United States of America 99, no. 10 (2002): 7280–87. http://www.jstor.org/stable/3057854
Chen, Shu-Heng, Mak Kaboudan, and Ye-Rong Du (eds), The Oxford Handbook of Computational Economics and Finance, Oxford Handbooks (2018; online edn, Oxford Academic, 5 Feb. 2018), https://doi.org/10.1093/oxfordhb/9780199844371.001.0001, accessed 19 Oct. 2024.
Hendry, David. “A Short History of Macro-econometric Modelling”. Economic Papers 2020-W01, Economics Group, Nuffield College, University of Oxford. https://www.nuffield.ox.ac.uk/economics/Papers/2020/2020W01_MacroHist18.pdf
Marx, Karl. “The Eighteenth Brumaire of Louis Bonaparte”. 1852. https://www.marxists.org/archive/marx/works/1852/18th-brumaire/ch01.htm
Poledna et al. “Economic forecasting with an agent-based model”. European Economic Review, Vol. 151. January 2023. https://www.sciencedirect.com/science/article/pii/S0014292122001891
Turrell, Arthur. “Agent-based modelling: understanding the economy from the bottom-up”. Bank of England Quarterly Bulletin. 2016 Q4. https://www.bankofengland.co.uk/-/media/boe/files/quarterly-bulletin/2016/agent-based-models-understanding-the-economy-from-the-bottom-up.pdf
Wang et al. “Agent-based models in financial markets studies”. Journal of Physics: Conference Series, 1039 012022. 2018. https://iopscience.iop.org/article/10.1088/1742-6596/1039/1/012022/pdf
Wiese et al. “Forecasting Macroeconomic Dynamics using a Calibrated Data-Driven Agent-based Model”. INET Oxford Working Paper No. 2024-06. 27 September 2024. https://oms-inet.files.svdcdn.com/production/files/Forecasting_Macroeconomic_Dynamics_Sep2024_WP-compressed.pdf?dm=1727710642