When Can Cognitive Agents Be Modeled Analytically versus Computationally?

Abstract

When agents are modelled with psychologically realistic decision making rules, the assumptions made about their decisions and behaviour are usually more complex than those made in traditional economics. This has made it harder for behavioural and cognitive economists to derive mathematical results analogous to those of the Arrow-Debreu theorem and similar economic findings. It has also made it difficult to generalise behavioural economic results – for example Fudenberg [Fud06] criticises the approach of modifying individual assumptions instead of considering them as a group.

As a result of this additional complexity, many modellers in behavioural economics are turning to agent-based computational methods instead of attempting to find analytic, closed-form solutions to economic problems (e.g. [Boq11], [Tes02]).

Computational methods have many advantages over traditional analytic methods, but also some disadvantages (for example, they make it harder to make very general findings, or to place an economic interpretation on some results).

This paper proposes a new method of modelling agent decision making and behaviour, based on information processing rather than utility and preferences. Models can be built whose agents follow such rules; these produce different micro and macroeconomic predictions to conventional economic models. By shifting the basis of the model to information and how it is transformed by agents, it becomes possible to develop new kinds of economic models which can be understood analytically, not just by computer simulation.

The agents in the model have goals which stochastically become salient at different times ([DTM08]). They learn and use strategies to achieve those goals, such as adaptive heuristics ([HMC03]) and fast-and-frugal rules ([GG02]). They process information using the approach of Payne, Bettman and Johnson 1993, and make choices based on heuristics such as attribute substitution (for example [KF02]).

This approach can complement computational methods and provides some of the generality and elegance which is often thought to be missing from behavioural approaches. It can also replicate some of the standard economic results as well as being compatible with a number of empirically discovered ”anomalies” from the judgement and decision-making (JDM) and behavioural economics literatures. It may also offer a way to consider aspects of certain phenomena at the cognitive level which are outside of the scope of traditional choice-based economics, but which are clearly important to real individuals and have real-world consequences: motivation, happiness, deliberate ignorance, and the learning of new preferences.

The paper is currently a work in progress and represents one step towards a possible closed-form macroeconomic model based on cognitive and behavioural microfoundations.

Thanks to David Laibson, Liam Delaney and others for encouragement and discussions on this topic; to Elina Halonen for debate and development of the ideas, and suggestions of useful sources. Especial thanks to four anonymous reviewers for useful suggestions and comments on an earlier version of this manuscript, which helped to move it forward to the current version.