AdobestockWith the publication of “The Oxford Handbook of Agent-Based Computational Management Science” in February 2026, Friederike Wall, Shu-Heng Chen, and Stephan Leitner present the first comprehensive handbook on agent-based modelling in management science. In this interview, Wall and Leitner explain how they model individual behaviour through agents and how this approach enables them to explain and anticipate the behaviour of entire systems.
Why is it worthwhile to model agents and simulate their behaviour and its consequences?
Friederike Wall: In management science, some fields focus primarily on the micro level – that is, on the behaviour of individuals or collectives of few people – while others concentrate on the macro level, examining organisations and firms as a whole. The respective research cultures and methodologies differ considerably, which has led to a degree of separation between scientific communities. Agent-based modelling and simulation can help bridge this divide.
Stephan Leitner: This distinction between micro and macro levels is not unique to management science; it can be observed across many disciplines. In psychology, for example, some scholars study individual behaviour, while others focus on groups. Agent-based modelling and simulation aim to aggregate individual behaviour in order to observe and analyse outcomes at group level. That is the core idea.
In management science, the primary interest lies in the behaviour of a system – of a firm, for instance – is that correct?
Stephan Leitner: Yes, but in practice one can influence only the behaviour of individuals. Any intervention therefore alters the behaviour of multiple individuals, and with this method we can estimate the resulting effects at the macro level – for example, at the level of the firm. These effects are rarely linear; rather, they involve complex interdependencies. Social norms provide a useful illustration: when one person changes their behaviour, this affects others’ perceptions of what is socially acceptable, which in turn triggers further adjustments.
You mentioned that individual behaviour is aggregated. Who performs this aggregation?
Friederike Wall: The aggregation is carried out by the simulations themselves. This marks a fundamental difference from more traditional economic models. In such models, it is often assumed that the average of a population of economic actors behaves like a single actor – the so-called representative agent. Aggregation is then straightforward: put somewhat simplistically, one merely multiplies the outcome for one agent by the number of actors. Agent-based modelling and simulation depart from this simplified view. We consider heterogeneous agents who interact with one another in diverse ways within the simulation. As modellers – indeed, as experimenters – we then observe system-level variables, such as the extent and speed with which improvements in overall performance emerge.
How do you construct these heterogeneous agents and define their characteristics?
Stephan Leitner: We develop our agents on the basis of established research findings, for example from behavioural research.
Friederike Wall: Precisely. We draw on insights that enable us to incorporate bounded rationality into our models. In the chapter by Shu-Heng Chen, co-editor of the Handbook, he explains in detail how we derive our behavioural assumptions. At the same time, it is important to remember that any model, however sophisticated, remains a simplification of reality.
Suppose you are modelling how specific incentives affect managers within a firm. How do you implement heterogeneity, given that managers differ in their behaviour?
Friederike Wall: Much depends on the theoretical perspective one adopts. A microeconomist might argue that each actor should be assigned a specific utility function. Typically, this function reflects an interest in income and – often disproportionately increasing – cost of effort. Since actors are assumed to behave as individual utility maximisers, the utility function is mathematically maximised. We proceed differently. We assume bounded rationality, which may vary from one manager to another. Our models therefore incorporate not only what matters to the agent – such as income and effort – but also the information available to them, their expectations about the future, and potential cognitive biases. We assume limited information and recognise that agents are not aware of all possible courses of action; instead, they must search for and discover these options gradually.
Stephan Leitner: Learning is another key element in our models. Yet even if all agents follow the same learning rule, outcomes may still differ. Consider expectations about others’ behaviour: these evolve over time and already vary between individuals. Even with identical learning mechanisms, the results will therefore diverge. In the context of social norms, for example, individuals observe what their colleagues do and infer from this what is socially acceptable. If they observe different behaviours, they will draw different conclusions.
You frequently refer to other disciplines. How did you, as management scholars, come to adopt this method?
Friederike Wall: In the field of Management Control, I had long been uneasy with overly simplified behavioural assumptions. Management Control ultimately aims to ensure the success of the organisation as a whole, which requires influencing individual behaviour. I then encountered agent-based modelling and simulation. In spring 2008, I succeeded in running my first model and presented it in my hearing for the professorship in Klagenfurt in July of that year. My collaboration with Stephan Leitner began in autumn 2009.
Stephan Leitner: We are interested in explaining how patterns emerge within organisations. For me, this is one of the most intellectually compelling questions in management science. At the time, this approach was entirely new in Management Control and therefore particularly intriguing.
Since when has agent-based modelling and simulation existed?
Stephan Leitner: The term has been in use since the 1980s, although the underlying concept is much older.
Friederike Wall: A well-known example is the segregation model developed by Nobel laureate Thomas Schelling and published in the American Economic Review in 1969. He explored how urban areas emerge in which people with similar characteristics cluster together, even when their preference for living among similar others is relatively weak. This was one of the earliest agent-based models.
Are there still areas in which ground-breaking discoveries might be made?
Friederike Wall: Certainly. In economics, many open questions remain – for example in financial markets, such as the phenomenon of so-called bubble formation. There are also pressing issues in climate policy: how can individuals be encouraged to adopt more climate-friendly lifestyles and adjust their behaviour accordingly?
Stephan Leitner: Political opinion formation is another promising field. During the COVID-19 pandemic, numerous agent-based models were used to estimate and anticipate human behaviour.
How is the theoretical framework of the method evolving? Are new elements emerging?
Stephan Leitner: Artificial intelligence opens up new possibilities. As we have explained, modelling agents still involves substantial manual work. In principle, artificial intelligences could themselves be defined as agents. Such agents could communicate meaningfully with one another on the basis of available data. This is an area characterised by rapid development.
Do you see your research being applied in practice, whether in politics or in business?
Stephan Leitner: Yes, there is considerable interest, particularly among firms.
Friederike Wall: While not necessarily our own specific research, the method itself is widely used and well regarded in policy advisory contexts. Experiments conducted in simulated environments are more manageable and less consequential than those carried out in real-world systems – and are therefore used frequently. They allow policymakers and organisations to test various levers and assess potential consequences before implementing measures in practice.
.flex_column.av-1ihi5w47-e0213dbe1f0ee903df295a91ff6b97de{
border-radius:0px 0px 0px 0px;
padding:25px 25px 25px 25px;
}
#top .av-special-heading.av-mljg76qp-78a360808941763f8e23c38c776d9c29{
margin:0px 0px 0px 0px;
padding-bottom:0;
}
body .av-special-heading.av-mljg76qp-78a360808941763f8e23c38c776d9c29 .av-special-heading-tag .heading-char{
font-size:25px;
}
.av-special-heading.av-mljg76qp-78a360808941763f8e23c38c776d9c29 .av-subheading{
font-size:15px;
}
About the handbook
Friederike Wall, Shu-Heng Chen (Professor of Economics at National Chengchi University in Taipei, Taiwan) and Stephan Leitner worked for five years on the Oxford Handbook of Agent-Based Computational Management Science, published in February 2026. The volume brings together contributions from scholars across almost all continents on agent-based modelling and simulation and constitutes – at least within management science – the first work of its kind.
.flex_column.av-1by76h1z-e076575402d60f466cb5ca6b25cdb000{
border-radius:0px 0px 0px 0px;
padding:15px 15px 15px 15px;
}
#top .hr.hr-invisible.av-4fagb6f-35579c49457f2faeab343210f22516c4{
height:25px;
}
.avia-image-container.av-mljg6msa-c7ed41dcd02272dd63d38349cc2c24d5 img.avia_image{
box-shadow:none;
}
.avia-image-container.av-mljg6msa-c7ed41dcd02272dd63d38349cc2c24d5 .av-image-caption-overlay-center{
color:#ffffff;
}
#top .hr.hr-invisible.av-mljg7lvt-bd08cfa550760b754e62f595d8ebced8{
height:50px;
}
.flex_column.av-wz4yi3b-448542c8803ebd32faa1d9cbc3f4e66f{
border-radius:0px 0px 0px 0px;
padding:25px 25px 25px 25px;
}
#top .av-special-heading.av-mljg94a7-4a2a6b5b058c7d3a3dd07550b94331a4{
margin:0px 0px 0px 0px;
padding-bottom:0;
}
body .av-special-heading.av-mljg94a7-4a2a6b5b058c7d3a3dd07550b94331a4 .av-special-heading-tag .heading-char{
font-size:25px;
}
.av-special-heading.av-mljg94a7-4a2a6b5b058c7d3a3dd07550b94331a4 .av-subheading{
font-size:15px;
}
About the authors
Friederike Wall studied Business Administration at the Georg-August University of Göttingen and completed her doctorate in 1991 at its Institute of Information Systems. In 1996, she obtained her venia legendi in Business Administration at the University of Hamburg and subsequently accepted a professorship at the Private University of Witten/Herdecke, where she held the Dr Werner Jackstädt Endowed Chair of Business Administration, specialising in Management Control and Information Management. In April 2009, she joined the University of Klagenfurt, where she currently teaches and conducts research in Management Control and Strategic Management. Since 2013, Friederike Wall has been an elected member of the Academia Europaea (section ‘Economics, Business & Management Sciences’).
Stephan Leitner has been Associate Professor at the Department of Accounting, Financial Management, and Governance in the field of Management Control and Strategic Management since 2018. He studied Applied Business Administration at the University of Klagenfurt and completed his doctorate in Social and Economic Sciences there in 2012. In 2018, he was granted the venia docendi in Business Administration at the University of Klagenfurt.
Der Beitrag “We seek to aggregate the behaviour of individuals in order to predict the behaviour of organisations or networks.” erschien zuerst auf University of Klagenfurt.