But the economic models they relied on fell short in this volatile reality. These models assume a rational, stable and predictable world - while real life, especially in times of crisis, is rarely so orderly.

Researchers at the University of Amsterdam (UvA), working closely with several central banks, developed an alternative. Their so-called Agent-Based Models integrate a more realistic behavioural perspective and therefore offer better insight into how the economy evolves. Especially during crises and when forecasting inflation, these models outperform traditional approaches. And that matters: the better central banks understand crisis dynamics, the more targeted and effective their interventions can be. This research helps societies safeguard themselves against future economic shocks — from pandemics to housing bubbles and geopolitical tensions.

From idea to application

Agent-Based Models are central to the work of Cars Hommes, Professor of Economic Dynamics, and his team at the Center for Nonlinear Dynamics in Economics and Finance (CeNDEF). This research centre, part of the Amsterdam School of Economics, studies complex economic systems from a behavioural perspective. A key concept in behavioural economics is bounded rationality: people do not behave fully rationally. They have limited information, they respond emotionally, and they adjust their expectations based on the behaviour of others. But how do you model this in a realistic way? 

In the 1990s, Hommes and American economist William Brock developed a theory of heterogeneous expectations. Instead of assuming that all economic participants (agents) form the same forecasts, the theory recognises that thousands of agents can hold different beliefs — and can switch strategies when conditions change. Building on these insights, CeNDEF developed one of the first theoretical Agent-Based Models for the financial sector. 

A theoretical model only gains value once it is tested against real human behaviour. Between 1998 and 2008, CeNDEF conducted a series of laboratory experiments. These confirmed that the behavioural patterns predicted by the theory actually emerged in practice. Outside the lab, during the 2008 financial crisis, this new type of model also proved its worth. Compared with traditional models, Agent-Based Models were better able to explain shocks in stock prices, housing prices and inflation. 

From the lab to policy 

For a model to be useful for policy, it must do more than hold theoretically - it must be able to process vast amounts of data and simulate the behaviour of thousands of economic agents. Collaboration with Austrian physicist and economist Sebastian Poledna, supported by European research funding, made this possible. Together they developed a detailed Agent-Based Model of the Austrian economy, based on extensive micro- and macro-data on households, firms, and banks. 

CeNDEF’s work began gaining momentum. Central banks increasingly recognised the potential of this new modelling approach. In collaboration with the European Central Bank and De Nederlandsche Bank (DNB), Hommes explored practical applications of Agent-Based Models. Among the results was a 2019 'heatmap' developed with DNB to detect housing market fluctuations earlier and more reliably.

The COVID-19 crisis as a testing ground 

The real breakthrough into policymaking came in Canada. The DNB housing-market study drew the attention of the Bank of Canada and sparked a collaboration. Hommes and researchers from the Canadian central bank developed a new model tailored to the Canadian economy: the Canadian Behavioural Agent-Based Model (CANVAS). The model was then added to the Bank of Canada’s official modelling suite — making CANVAS the first large-scale macroeconomic Agent-Based Model actively used by a central bank.

During the economic aftermath of the pandemic, CANVAS proved its value. When inflation rose sharply, the model revealed how cost increases spread unevenly across sectors — an insight traditional models failed to capture. The Bank of Canada now uses CANVAS structurally: not as a replacement for existing models, but as an indispensable complement that provides a richer and more realistic understanding of inflation dynamics. 'CANVAS provided us with additional insights into post-pandemic inflation dynamics. It enabled us to analyse economic developments through various behavioral mechanisms - something traditional models could not offer', says Bank of Canada Senior Policy Director Yang Zhang.

Gaining control in an unpredictable world 

CANVAS was published in 2025 in the Journal of Economic Dynamics and Control. But its greatest value lies beyond academia: better-informed central banks can prepare more effectively for economic shocks. When monetary policy reflects the true complexity of the economy, people notice the difference - in the form of more stable prices, fewer disruptions in key sectors, and better employment prospects. 

Central banks including De Nederlandsche Bank and the Bank of England are now taking concrete steps to expand their use of this modelling approach. According to Hommes, it is essential that all central banks eventually develop mature Agent-Based Models. Only then can they respond quickly and appropriately to economic disturbances - a necessity in a world where crises emerge faster and more unpredictably than ever. 

Would you like to know more?

Betrokken Impact Centre: Center for Nonlinear Dynamics in Economics and Finance (CeNDEF)

Publications and reports

• CANVAS: A Canadian behavioral agent based model for monetary policy. C. Hommes, M. He, S. Poledna, M. Siqueira en Y. Zhang (2025). Journal of Economic Dynamics and Control, 172.
• Economic forecasting with an agentbased model. S. Poledna, M.G. Miess, C. Hommes en K. Rabitsch (2023). European Economic Review, 151.

Other sources

• Cars Hommes: Why markets – and AI – behave like humans (2025, 3 november). Harry Geels, Financial Investigator.