1. INTRODUCTION
Over the last 30 years, new frameworks for thinking about and quantifying risk have emerged in the insurance and risk management space. One was Dynamic Financial Analysis (DFA), a perspective from which actuaries could acknowledge and quantify the stochastic nature and interrelationships of financial, economic, and risk processes. Later, Enterprise Risk Management (ERM) took hold, looking at a system and an organization in a “portfolio” framework as an integrated whole.[1]
Looked at from an evolutionary perspective, DFA and ERM shared an underlying actuarial compulsion: to better understand the contextual socioeconomic environment in which risk processes operate and to provide a framework for quantification and holistic consideration of the relationships and interactions among elements of the entire system.[2] Over the last two decades, the development of both frameworks has led to their adoption and consistent use by actuaries and others interested in the quantitative modeling of financial and risk-related processes. Because of this, ERM (as was true, although to a lesser degree, with DFA earlier) has a significant presence on the CAS’s Education Syllabus.[3]
Given the additional insights and level of maturity reached by actuaries using the DFA and ERM perspectives, it is time for actuaries to take the next logical step in this evolutionary analytical path: to recognize insurers, risk-related organizations, and even entire insurance markets as complex systems of interrelated risk and socioeconomic processes, which can benefit from an agent-based modeling (ABM) approach and provide a framework for strategy and operational decision-making. Since a complex system involves not just numerous interactions between and among its components and the environment but also behaviors, emergent phenomena, and feedback responses, a modeling framework such as ABM is necessary to measure the full impact of interrelationships and unpredictable outcomes and to quantify the impact of changes in the wide variety of parameters.
Importantly, because ABM is a technique that (as will be discussed in the next section) models environmental and organizational conditions and processes at a “micro-level” and then simulates their collective aggregate impact which can be observed at the “macro-level,” ABM can be used to evaluate any aspect of the broader complex system. Like DFA and ERM, the typical application of ABM may involve the quantification and relative impact of an organization’s alternative operational and strategic decisions. But because it involves modeling of the contextual subprocesses as well as how they interact and aggregate, ABM can also provide a framework for modeling and analysis of these subprocesses themselves. Or it can provide an analytical tool for a system or market in its entirety.[4]
As a practical matter, the number of potential real-world situations to which an ABM approach can be usefully applied is considerable, and it is perhaps only limited by the imagination of the actuarial user. To appreciate the potential benefits of ABM, it might be beneficial to consider one or more of the following potential areas of application while reading this paper:
-
Risks that involve some type of spreading, e.g., a disease or virus through a population, a wildfire through a geographical area, or a catastrophe percolating through a collection of properties. In each case, various assumptions regarding the micro-level dynamics of the spread can be tested for their differential macro-level impact.
-
Insurance or risk management activities that depend on the behavior of individuals, e.g., responsiveness of policyholders to changes in premiums or policy provisions or to external events (including economic, financial, regulatory, societal events, and more).
-
The market behavior of individual consumers and insurers, and their impact on the macro-level dynamics of the insurance market. For example, the competitive market impact with respect to interacting rate changes and market shares.
-
The potential effects of strategic or operational decisions by insurers.
-
Considerations of public policy proposals that impact, or are implemented at, the individual or organizational level.
-
New or emerging risks, and speculation about how such risks might percolate through society.
The remainder of this paper is organized as follows. Section 2 provides background with respect to the nature and characteristics of complex systems, as well as the fundamentals underlying the ABM framework. Section 3 touches on relevant literature associated with complex systems, particularly on some applications of ABM to issues related to risk management, insurance, and actuarial science. Section 4 provides several illustrative risk-related ABM examples, both for pedagogic purposes and to demonstrate potential applications of a complex system and ABM framework to issues of actuarial interest. Section 5 identifies some possible further actuarial applications of ABM—a list that has the potential to not just expand but explode! Section 6 concludes and summarizes.
2. BACKGROUND
2.1. Complex Systems
Every type of risk, or line of business, of interest to actuaries has potential analytical challenges, in part because the system embodying that risk is multidisciplinary and interconnected. Determining the potential future cost of a risk involves a broad context and a large set of factors; one categorization of those factors might be:
-
Environmental factors such as socioeconomic and financial conditions. For example, future risk and insurance costs are at least partially functions of current and future inflation and interest rates.
-
Policy factors such as rules and regulations associated with the risk and the various risk management techniques that might apply. Insurance-related examples might include the prevailing judicial climate, regulations, or guidelines requiring or suggesting the use of certain risk control mechanisms.
-
Personal factors such as a person’s or organization’s skillset, risk-averseness, and other attributes. For example, a person’s background, training, and experience may influence their susceptibility to risk.
When one considers the depth and breadth of each of these three categories of factors, it becomes evident how multidisciplinary and interconnected—how “complicated”—any issue associated with risk is. However, the word “complicated” does not fully do explanatory justice to the nature of such an interconnected system.
For purposes of a taxonomy of systems, a “complicated” system is one for which the totality of the system is basically equal to the sum of its parts. The process commonly exhibits a “linear” relationship between cause (the components of the system) and effect (the outcome or result of the system). In a sense, such a system is basically deterministic or Newtonian. If one knows the starting conditions and an understands the “rules” and components of the system, then in principle one could largely predict the future of the system and quantify it (assuming, of course, that there are no exogenous shocks or changes in the contextual conditions, which in reality is extremely unlikely…).
These elements are present to a certain degree in many risk-related processes, but there is much more to it than that. More generally, a risk-related process is not so simple. Instead, the system (as well as even some of the system’s subprocesses) is more accurately labeled a “complex” system.
A complex system has a number of high-level attributes that distinguish it from a “merely” complicated system:
-
Agents: the “actors” in the system, i.e., people and/or organizations that are interacting with each other and with their environment. They are each endowed with rules regarding their behavior and original endowments of skills and resources.
-
Emergence: the whole is generally more than the sum of its parts, and the totality of the system’s properties stems from the self-organizing behavior of the components. In particular, the system organizes itself from the bottom-up (as opposed to top-down). This is frequently referred to as “spontaneous order” in economics or sociology.
-
Nonlinearities: the relationship between cause and effect is not typically linear. More specifically, the behavior of the system at the macro-level is not directly evident or inferable from knowing and understanding the components and rules that constitute the system at the micro-level.
-
Challenging to predict: while not necessarily chaotic, the future of the system (like any stochastic process) is difficult to predict because of the indirect nature and the nonlinearities of the relationships between the micro- and macro-levels of the system. One of the advantages of a complex systems perspective is to recognize that a better understanding of the system is a critical prerequisite to responsible and accurate modeling and prediction of the possible future states of the system. Viewing a system as “complex” helps provide the necessary context and vocabulary—emergence, nonlinearities, etc.—to better understand its processes and ultimately model the system.
-
Adaptive: most complex systems can be characterized as adaptive. This means that there are feedback loops. The agents in the system adjust or adapt their behavior based on the rules and policies associated with the system’s environment, as well as the behaviors of other actors and agents. They do this to improve, maximize, or optimize their situation. Although each type of risk will have different contexts in which feedback occurs, simple examples of adaptation to environmental factors in personal auto insurance would be the tendency of policyholders to drive less (with consequent lower exposure) in response to higher fuel costs, greater perceived dangers associated with driving, or stay-at-home orders or recommendations during a pandemic. Systems with this type of potential feedback are often called Complex Adaptive Systems (CAS). The ability of the system to self-organize and adapt means that it can exhibit flexibility and resilience in response to changes in the environment.
Complex systems have entered the public consciousness over the last few decades, in part due to the research and outreach of the Santa Fe Institute (SFI), a high-profile thinktank whose mission is:
…to understand and unify the underlying, shared patterns in complex physical, biological, social, cultural, technological, and even possible astrobiological worlds…. As we reveal the unseen mechanisms and processes that shape these evolving worlds, we seek to use this understanding to promote the well-being of humankind and of life on earth.[5]
Through the efforts of researchers at SFI and other organizations, the Complex Systems perspective has become an important tool and approach to better understanding of processes in many scholarly disciplines that we see play out as our day-to-day reality.[6]
2.2. Agent-Based Modeling
ABM is a method of micro-simulation (small-scale simulation) that allows for the observation and analysis of emergent phenomena such as those associated with a complex system. ABM has become an important and valued approach to modeling true complex systems. This modeling framework should be considered by actuaries for inclusion in their toolkits to better understand the essential nature of a risk process and to help quantify the potential distribution of future costs of risk.
ABM allows for the observation of potential macro-behavior emerging from the underlying agent-level micro-activity and characteristics, and thus it can provide significant insight into the identification of potentially beneficial strategic and operational decisions.
Key characteristics of ABM include:
-
Agents: these are the actors—individual people and/or organizations—in the simulated complex system drama. Their interactions, with one another and with the modeled environment, are the basis for the macro-phenomena and properties that emerge, often unexpectedly, from the system. Agents act autonomously (there is no central planning mechanism) according to prescribed rules and behaviors and consistently with their programmed endowments. In this way, the agents collectively produce, from the bottom-up, the system’s macro-level emergent phenomena, and they do so “spontaneously.”
-
Adaptation: agents are, within their prescribed behavioral rules and endowment properties, flexible and capable of responding to changes in the system’s interconnected network and environment. Agents can learn from their experiences and adjust their behaviors (within prescribed rules) to optimize their positions in pursuit of their programmed goals and objectives.
-
Heterogeneity: agents are heterogeneous with respect to both the characteristics of their endowed state (e.g., their age, education, family situation, abilities) and their behaviors. Information regarding the diversity of real-world characteristics across people and organizations provides important aspects for parameterizing the model. This is an area of intense advantage for the ABM approach to modeling complex systems, as it is difficult to reflect agent heterogeneity in traditional/mainstream economic equation-based modeling.
Once an ABM is designed and running, it can serve as a basis for understanding the underlying dynamics of the system and its multifactor interactions, including elucidating how macro-properties emerge from micro-behaviors. Through observation and scenario testing, critical aspects and parameters of the system can be identified, and those points can be targeted for additional research and data collection. For example, by treating each factor or parameter in the system to be equivalent to a knob which, when turned, gradually changes the value of the parameter; then each knob can be slowly turned while all others are held constant to identify the manner and degree to which the emergent characteristics of the system change. Proposed changes to the system can be tested to help evaluate their potential impacts, at both qualitative and quantitative levels. Better understanding how adaptation occurs within a system can provide insight into how best to amend or influence the system, for example, for proposed policy changes.
The value of an ABM framework is that it allows for the simulation and observation of macro-structures and behaviors produced by the components and micro-behaviors within the system, even though the former are not typically directly anticipated from knowledge of the latter. This is an important capability, because an inability to clearly infer aggregate properties from the components of a system means that equation-based modeling—the primary basis for modeling and insight into, for example, economic processes—is of very limited value.
This identifies the primary advantage and payoff from adopting a complex systems perspective, quantified through an ABM approach. The macro-properties of a system emerge from the detailed characteristics of the many diverse individual components, the behavioral rules by which they interrelate, and the manner in which they interact with the environment. Indeed, often a complex system’s ultimate characteristics are unexpected based on a superficial observation of the starting conditions; that’s why “emergence” and “nonlinearity” are critical traits of such a system. The complex systems perspective provides new insights into the underlying causes of the behavior of stochastic systems, behavior that is not a simple sum of its parts. ABM provides a mechanism by which the modeler can go beyond simple scenario testing or simulation resulting from changed high-level assumptions. ABM more thoroughly considers possible future states and values of the system by taking account of micro-level interactions and feedback loops throughout the system.
2.3. Criticisms of Agent-Based Modeling
ABM (and simulation in general) is not without its critics. The criticisms seem to revolve around:
-
The non-generalizability of ABM/simulation results. In particular, findings from ABM/simulation models do not generalize because they do not provide the theoretical framework for results and analytical conclusions that a system of equations does.
-
The difficulty of interpretation, i.e., a simulation structure does not have the clear interpretational foundation that accompanies a system of equations.
-
Difficulty of estimation and parameterization of the model.
While an extensive discussion of these criticisms, and replies to them, is beyond the scope of this paper, we briefly mention two papers that propose solutions and counterarguments to these issues. Leombruni and Richiardi (2005)
“critically assess the pros and cons of agent-based simulations, vis-à-vis analytical models. We then argue that much of the criticism… of ABMs fails. Moreover, ABMs have some important advantages, like the richer specification they can support, a feature which allows the description of complex phenomena.”
Avegliano and Sichman (2003) suggest embracing both equation-based models and ABM and
“propose 2LevelCalibration, a multi-stage approach for the calibration of unknown parameters of agent-based models.”
One other aspect of these criticisms is context. It is important to remember that, in addition to quantifying possible future values of the cost of risk, ABM can help us understand the nature of the system and its complex network of interrelationships. Within this context, certain criticisms of ABM may take on somewhat less significance.
2.4. Agent-Based Modeling Software
With respect to ABM logistics used in this paper, NetLogo is the software used for several reasons. It is a very popular and commonly utilized framework, it is user-friendly, and it is freely and publicly available.[7] NetLogo is a multiagent programmable modeling environment and has formed the modeling basis for numerous practical applications as well as research projects.
In summary, ABM is an attractive framework for modeling and understanding complex systems. Based on their penchant for modeling and simulation, actuaries should find it a familiar and comfortable concept.
3. LITERATURE REVIEW
ABM research occurs in several economic and social science fields of study. There has been some application of it to the insurance industry but very few applications specifically to actuarial science. This section will identify and briefly summarize some of the literature that is potentially relevant for casualty actuarial purposes—whether because of a paper’s usefulness with respect to ABM in general, or because of its exploration of a particular subject matter.
Several recent publications involving complex systems, and advocating for the complex systems viewpoint, include (all references are listed in the References section):
-
Farmer (2024): A very readable and approachable book describing the complex systems perspective and the evolution of complex systems as a discipline, especially in its application to economics.
-
Krakauer (2025): This opening essay, at nearly 100 pages, is from a large, four-book series, each presenting dozens of original papers in chronological order and with commentary, representing the historical foundations of current complex systems science.
-
Jensen (2023): A (somewhat more advanced) textbook on complexity as a “science in its own right.”
Several papers discussing ABM, both in general and to demonstrate the broad range of (nonactuarial and noninsurance) applications of the ABM framework, include:
-
Bankes (2002) describes the potential for an ABM revolution in social science and discusses what must happen to achieve this potential.
-
Chliaoutakis and Chalkiadakis (2016), Axtell et al. (2002), and Olševičová, Cimler, and Machálek (2013) describe applications of ABM to model ancient societies’ population growth and organizational structure.
-
Grimm et al. (2010, 2006) describe the overview, design concepts, and details (ODD) protocol that helps standardize the published descriptions of ABM, in the hopes of making such models less susceptible to criticism (particularly regarding irreproducibility[8]).
Finally, and perhaps most interestingly for the purpose of this paper, there are several insurance- and actuarial-related papers worth identifying and summarizing:
-
England, Owadally, and Wright (2022) considers an ABM model with two types of agents—auto insurers and customers—and explores the impact of word-of-mouth networks that allow for the spread of information (positive or negative) among customers regarding the performance of insurers.
-
Dubbelboer et al. (2017) develops an ABM framework that models the evolution of the dynamics involving flood risk and peoples’ vulnerability to that risk.
-
Sajjad et al. (2016) builds an ABM framework to model the dynamics of family formation. Gorvett (2024) uses this as an example of how to apply the ABM framework to questions involving retirement plan and pension evolution in response to changes in family structure.
4. ILLUSTRATIVE ABM EXAMPLES
In this section we refer to several models in the NetLogo library. In particular, three examples of publicly available ABM models are presented:
-
“Life”—a simple two-dimensional cellular automaton.
-
“Fire”—a model of fire risk in a wooded area.
-
“Virus”—a model of the spread of a virus.
In addition to being publicly available (allowing the reader to follow along, replicate, and further experiment with each model), these examples were chosen (i) for pedagogical purposes and (ii) to demonstrate the potential applications and capabilities of even basic ABM models. Each of these three models is relatively simple to understand, and the concept and documentation are clear and straightforward. Each represents a system that an actuary might be interested in modeling, whether directly or analogously. The examples are of varying relevance for actuaries. The fire model is of significant direct relevance (especially given recent wildfire activity) and can also be analogized to any risk that percolates its way through either physical geography or human society. Similarly, the virus model has direct application for health insurance, or any other risk affected by epidemiological considerations, and can also be analogized to the spread of social behavior and beliefs (perhaps this is a mechanism through which to better understand reputational risk, always a difficult exposure in an ERM context).
Admittedly, the life model probably does not have direct actuarial relevance.[9] It is a remarkably simple and transparent “game” that demonstrates how simple rules can lead to not-so-simple outcomes and patterns. That’s why it’s included here, for its pedagogical value, and because many readers may already be familiar with it at some level. Its usefulness as a starting point for better contextualizing and understanding more sophisticated complex systems is the reason it is the first example.
4.1. “Life” Example
The “Game of Life” was invented by mathematician John H. Conway and is described in Gardner (1970), among many other places. In NetLogo’s incarnation,[10] it is in the form of a two-dimensional cellular automaton (CA). A CA is a computational model involving a finite-dimensional grid of cells, where each cell is in one state or another.
For “Life,” each cell is in either a dead (black) or alive (green) state, and each cell’s state at the end of a time interval, or “tick,” depends on the states of the cell’s grid neighbors at the start of the interval. In particular, the rules are as follows:
-
The cell dies if, at the beginning of the tick, the cell has fewer than two neighbors (of the eight that surround the cell) that are alive (this might be akin to “loneliness”).
-
The cell dies if, at the beginning of the tick, the cell has more than three neighbors that are alive (might be thought of as “overcrowding”).
-
If the cell has exactly two neighbors that are alive, the cell stays in the same state during that tick.
-
If the cell has exactly three neighbors that are alive, the cell comes to life (if dead, it goes from dead to alive; if alive, it stays alive).
These rules apply to each cell during each tick.
The compelling thing about CAs is that interesting and sophisticated—and often surprising—patterns can result from rather simple rules.[11] For our purposes here, looking at a CA example is also useful pedagogically (the rules are generally simple enough that the future states of the model can be visually confirmed) and for setting the stage for more advanced ABM models that involve more sophisticated rules of behavior.
Figure 1 shows six consecutive screenshots from the NetLogo “Simple Life” model, showing the evolution of these rules over time. The first picture is the initial (time 0) randomly generated state of the system. The next five are pictures of the system after 1, 2, 3, 4, and 5 ticks, respectively.
What generally happens in this model is that, over time, some shapes emerge that persist and have a fixed position, while other shapes may change slightly and move around in a structured way. And sometimes the entire grid ends up completely dead. Given the fixed rules, the ultimate system configuration depends on the initial state(s) of the system.
While not necessarily directly applicable to an actuarial-related risk analysis, this type of model is useful for its transparency and simplicity.
4.2. “Fire” Example
Now let’s ratchet up the sophistication level of the ABM framework a bit. And in doing so, we increase the potential applicability of the ABM approach to situations and risks of more relevance to actuaries.
Fire, under many insurance guises, is a hugely significant risk for casualty actuaries. An ABM model in NetLogo is designed to model the spread of a fire through a forest.[12] While directly relevant for actuarial concerns, it is even more applicable analogously, as a way to model percolation or contagion in many forms.
Figure 2 shows a screenshot of the NetLogo model’s user interface. As shown, the primary input factor is the density of the forest. Other factors are coded into the model. Although both of these can be changed by the user, the default coding assumes that there is no wind (thus, the fire cannot “jump” over a non-wooded cell) and that the fire can only spread from a cell to an adjoining, non-diagonal cell (i.e., the cell directly north, south, east, or west of a cell that is currently on fire).
One use of this model is to consider how the density of the wooded area impacts the likelihood of the fire spreading all the way across the grid (a fire starts on the left edge and tends to work its way toward the right). Figure 3 provides an example for a particular density level (it happens to be 58%) in a sequence of screenshots, starting with the initial configuration and then with pictures of the system after about 100 and 200 time ticks, respectively.
Based on the last screenshot in Figure 3, it should not be surprising that, for this particular simulation run, the fire petered out before the 300th time tick, and it did not come close to reaching the far right side of the grid. In fact, at a density of 58%, it is very unlikely, based on numerous simulations, that a fire would percolate all the way to the right side of the grid.
However, additional analysis can be done to determine whether there is a critical density point at which the likelihood of a burn-through to the right side is much more likely to happen. As an illustration, I did this with (admittedly far too few) simulated trials: 10 at each integer percentage density level. The result was a clear-cut zone of criticality (or nonlinearity) between 58% and 61% density (Figure 4).
While simple, this is an example of the kind of analytical insights available from ABM and a complex systems perspective.
4.3. “Virus” Example
Although somewhat on the periphery of some casualty actuaries’ responsibilities, health risks and disease transmission is a familiar and relevant issue for many of us. This is also a very active area of research at the SFI.[13] Another reason for using this model to illustrate ABM is that it is simple enough to also, if the reader wishes, do a parallel equation-based analysis. For those more familiar with a traditional SIR (Susceptible-Infectious-Recovered) differential equation system, it may be comforting to see how an ABM approach is related to comparable results from an SIR equation-based approach.
A screenshot of the NetLogo model’s user interface appears in Figure 5. As shown, the inputs (all of which can be user-specified and changed) include:
-
300 people in the system (which determines the population density).
-
people with the virus have an 80% infectiousness rating (i.e., infection is spread roughly four out of every five chance encounters between people).
-
infected people have a 90% chance of recovery.
-
a 10-week duration of infectiousness.
Other input factors that are not explicitly shown on the interface, but are programmed into the code, include:
-
population turnover, including lifespan and birthrates (people die of old age at 50, and the chance of healthy people reproducing in a time period is 1%, as long as the population is below the assumed carrying capacity of the population).
-
the initial number of infectious people (10 in this example, or 3.3% of the population).
-
degree and length of immunity after recovery (the default assumption in the model is that immunity ends after one year).
Running the model involves the interaction of agents—those who are healthy, sick, or immune at any given time—with one another and the environment, and the visual is updated at each tick or iteration (in this case, one tick = one simulated week). Basically, people (agents) move around randomly (at whatever rate is specified in the code). If an infected person ends up on the same cell as a healthy person, the healthy person is infected. That newly infected person now moves around, potentially infecting others, and begins their progression to either death or immunity.
Stopping the model every 10 ticks allows us to visualize the virus’s progression (Figure 6).
These screenshots (at 0, 10, 20, and 30 weeks, respectively) provide a sense of the evolution of the population over time, with different colors designating the sick (red), immune (gray), and healthy (green) people. One can also see the progression of the various subpopulations over time in a more traditional chart provided by NetLogo (Figure 7).
Figure 7 shows some interesting points of inflection. For example, consider where the number of sick people in the population maxes out. The two points where the numbers of immune and healthy people intersect is also interesting. Both of these items could be explored further by testing their sensitivities to starting points (endowments such as number of initial infected peoples and their geographical proximity to each other) and to behavioral rules (length of disease, rate of recovery, rate of infectiousness, etc.).
Although some of these insights could also be done using a corresponding differential equation system, there are many underlying, explicit interactions in the ABM model that would not be clearly captured in the equation system. In addition, while certainly challenging, the parameterization of the ABM may be superior to the corresponding equation system, where there tend to be fewer parameters because multiple underlying interactions are often subsumed in the equation parameters. And, of course, there is great value to the visuals. It is an appealing feature to be able to visually show, by re-running the simulations, the characteristics of the virus-infected population.
One can also easily imagine adding additional factors into this ABM, which might not be so easily accommodated by the equation system. In ABM, it would just be a matter of editing the code to incorporate an isolation/lockdown requirement, the impact of the introduction of a medicine or vaccine, etc.
5. POSSIBLE FUTURE ABM RESEARCH
Once an ABM framework and computational structure is in place, there is virtually no limit to the number of questions and issues that can be addressed. Basically, any risk system that:
-
involves multiple agents that interact with one another and with their environment
-
exhibits emergence
-
involves adaptive feedback
can be better understood through an ABM model. In particular, critical factors and parameters can be examined, strategic and operational decisions can be better informed, and the potential impact of proposed changes to policies or laws can be elucidated.
A partial list of items that seem likely to be susceptible to an effective ABM investigation include:
-
Evaluating potential geographical aggregations of insureds under various scenarios.
-
Understanding the dynamics of customer-insurer relationships from a variety of perspectives, e.g., marketing, rate changes versus market share.
-
Understanding the dynamics of insurer-reinsurer relationships from a variety of perspectives, e.g., impact or hardening or softening of the reinsurance market, possibly in response to different types of exogenous events.
-
Understanding the dynamics of organizational and strategic decisions, e.g., speed and quality of claims handling and its potential effect on settlement value.
-
Modeling the progression of diseases or other health-related issues.
-
Modeling the economic and sociological impacts of various types of natural hazards.
-
Modeling the impact of legislative or regulatory changes and perhaps even helping to influence the nature or likelihood of such changes.
6. SUMMARY AND CONCLUSION
In this paper, I have advocated for a complex systems perspective on risk-, insurance-, and actuarial-related issues. In addition, I believe that ABM should be considered an important resource in the actuarial toolkit. ABM can help actuaries better understand the nature of complex systems and provide scenario testing and sensitivity analysis to systems that exhibit emergence and adaptive feedback. This paper has been intended as a tutorial and guidebook on ABM models and their potential. It has provided background material on complex systems and ABM, a brief summary of some of the particularly relevant literature, and illustrative case studies to demonstrate a simple ABM application. A list of possible future casualty actuarial research projects involving ABM models has also been included and is itself undoubtedly a work in progress.
Acknowledgments
The author gratefully acknowledges the Casualty Actuarial Society for issuing this call for papers (2025 Ratemaking Call Paper Program on Technology and the Ratemaking Actuary), and many thanks to three reviewers who provided numerous excellent suggestions, all of which significantly improved this paper.
Biography of the Author
Rick Gorvett is professor and former chair of the Mathematics and Economics Department at Bryant University. He is the director of Bryant’s MS in Actuarial Science program. Rick is an award-winning educator, has published numerous research articles, and has been an active volunteer with the Casualty Actuarial Society, including serving on the CAS Board of Directors. He has a PhD in finance from the University of Illinois and an MBA from the University of Chicago. His professional actuarial and risk management designations include FCAS, MAAA, CERA, ARM, and FRM.