Modeling Radical Uncertainty and Anticipating Uncertain Change with Models

Abstract

In this paper, I argue that agent-based modeling and simulation in a complexity framework can be useful scientific tools to think about the uncertain future. I will argue that the usefulness of this approach depends on its capability to generate surprise and discuss what conditions are needed for surprises to happen. The main value of agent-based models and simulations is to produce possible futures. This is the anticipation function of a model. It is important to emphasize that anticipating possible futures is different from predicting the most likely future evolution of some variables and requires a different understanding of what the science of economics can achieve. Therefore, I will discuss some epistemological thoughts related to the economic study of uncertain futures. Building on these foundations, the paper presents three methodological approaches of how agent-based models and simulations could be used to anticipate potential futures.

Keywords:

• radical uncertainty
• agent-based modeling
• surprise
• anticipation
• methodology

Notes

¹ I use the term “real world” as the opposite to the “artificial world” of models (see Sugden, 2000). Another term would be “external world” in contrast to the “internal world” of the model and the researcher’s imagination.

² Uncertainty can also be conceptualized in a more refined way. Kyläheiko (1998) distinguishes risk, parametric uncertainty, structural uncertainty, mutual dependency uncertainty, procedural uncertainty, and radical uncertainty. Radical uncertainty corresponds to Knightian uncertainty and is the strongest departure (or “de-isolation” in Kyläheiko’s terminology) from certainty.

³ Uncertainty induced by unknown strategic actions and reactions of other players and/or endogenized events corresponds to mutual dependency uncertainty in Kyläheiko (1998).

⁴ See http://www.technovelgy.com/.

⁵ I would like to thank an anonymous reviewer for pointing to Shackle’s potential surprise theory. An excellent summary of this theory and its relation to scenario planning can be found in Derbyshire (in press).

⁶ General introductions to agent-based modeling in economics are e.g. Axelrod (1997), Borrill and Tesfatsion (2011), Epstein (2006), Farmer and Foley (2009), Hamill and Gilbert (2016), Miller and Page (2007). A discussion of agent-based models in macroeconomics and a focus on radical uncertainty can be found in Roos (2015).

⁷ There is no written critique of economic agent-based modeling in the literature, because mainstream economists do not bother writing methodological papers. But skepticism is often expressed in conversations and it is not easy to publish this kind of work in leading journals. Axelrod (2006) reports difficulties in publishing papers with agent-based models in the field of biology as well.

⁸ Admittedly, there is a considerable epistemological difference between the ABM approach and the mainstream approach of analytical modeling regarding conception of what it means to “explain” something (see Epstein, 1999, 2006). Neoclassical analytical modeling is deductive, which means that an observation is explained if a proposition about that observation can be deduced from more general propositions. In contrast, agent-based modeling is generative, in the sense that agent-based models explain by providing computational demonstrations that given assumptions at the micro level are sufficient to generate an observed macro pattern of interest. I wish to thank the Editor for his remark on this point.

⁹ For a discussion in philosophy of science see Frigg and Reiss (2009).