Microfoundations of Expectations: The Role of Emotions in the Anticipation of Aggregate Outcomes of Human Interaction

January 17 - 19, 2017
Max Planck Institute for Mathematics in the Sciences

Workshop idea

By aggregation we understand the process by which behaviour of many individuals yields outcomes at the level of groups, organisations, markets, institutions, or macro-economies. Outcomes may take many forms including performance, prices, allocations, or satisfaction of individuals. What are the determinants of aggregate outcomes? How, and under what circumstances, are aggregate outcomes result from, and induce, individual behavior? In other words, how are aggregate outcomes of their actions, taken individually, anticipated by and factored into individual actions?

These classic questions are of central importance in the fields of management and economics, and in social sciences more generally. In the workshop, we aim at exploring fresh approaches and answers to these old questions. As one example, we propose to investigate the functional role of emotions in individual behaviour, as well as in the processes of aggregation and its anticipation. Moreover, we propose to develop some new ideas and models of the process of aggregation.

Since the later half of the 20th century, economists have employed two basic devices--rationality and equilibrium--to deal with the question of how individual actions generate aggregate outcomes, and how aggregate outcomes are anticipated. Muth (1961) applied these ideas to a broad range of macroeconomic phenomena by combining them under the label of 'rational expectations'. In this framework, anticipations of the aggregate-level outcomes of individual actions are required to be consistent with the actual outcomes. In other words, the subjective probability distributions over the outcomes are consistent with the outcomes that are generated by rational actions based on those same subjective beliefs.

Viability and attractiveness of Muth's proposal arose from its addressing several problems simultaneously. First, it gave a specific form to individual normative rationality by requiring that agents anticipate distribution of outcomes without bias. Second, it specified how aggregate outcomes enter into individuals' anticipations and consequent decisions. As expectations are assumed to be consistent with outcomes that actions based on such expectations generate in a probabilistic sense, mere updating on basis of draws from a known distribution need to take place. And third, it addressed the problem how the learning of agents feeds into actual outcomes. Thus, Muth's proposal is elegant from a theoretical viewpoint. In classic game theory (using the equilibrium concepts of Nash and Rationalizability), basically the same logic applies. Agents' rationality extends to anticipating aggregate outcomes that are consistent with common knowledge of rationality.

The weaknesses of this approach are also well known. Agents are fallible, the space of outcomes can evolve, and it is not clear from making rationality assumptions if agents coordinate on 'bad' or 'good' equilibria, e.g. the cooperate outcome in the prisoners' dilemma. Moreover, when we model aggregation via the assumption of common knowledge of rationality, the role of psychology is narrowed to account for mistakes --deviations from rational expectations. The functional role of emotions and other psychological processes cannot be addressed as long as the tight assumptions of rational expectations are retained.

In this workshop, we propose to explore a new approach. Instead of asking how to make the assumptions of the model more realistic, we strive to understand (a) what mechanisms may help individuals to anticipate aggregates, without overwhelming the mind with implausible levels of complexity; and (b) if there are complexity reducing mental mechanisms that can be understood and incorporated into models to help determine aggregate outcomes.

In terms of (a), we wish to address the following understudied topics:

  1. Constraints on higher order beliefs.In reality, beliefs about beliefs of other agents (i.e., higher order beliefs) seem to be constrained to one or two, and never more than a few orders. That means that agents may reason about other agents' reasoning, but rarely beyond the second or third level. While Muth (1961) was right that higher order beliefs are highly constrained, he did not specify \emph{how} they get constrained. The common knowledge of rationality in RE hardly constrains one from forming and acting on higher order beliefs. Global games are one example that show that outcomes are improved once we have less tightness as in the 'common knowledge of rationality' case.
  2. Emotions as elements of rationality.Although it has often been proposed that 'animal spirits' govern processes such as bubble formation, no theoretical or measurable basis for understanding this concept has been laid down and verified. Perhaps they refer to the first few orders of beliefs. Beyond that, we argue that forming and using additional layers of beliefs may be simply impossible for human cognitive endowment. Within our cognitive upper bounds, emotions may help regulate the level of thinking applied in specific circumstances. Instead being a counterpoint to rationality, as frequently assumed in economics and psychology, emotions may be an essential regulating and enabling element of rationality.
  3. Emotions as starting and stopping points.Emotions may be helpful in controlling and managing the potentially exploding complexity of higher order expectations. Emotions may instruct us when to think about others' thinking, and when to ignore such higher orders of analysis and just act without it. Are there elements in the human subconscious that provide us with functional emotions that help to handle the anticipation of aggregate outcomes?
  4. Emotions as sanity checks.Emotions may serve as sanity checks on expectations. Sometimes we feel that something is out of bounds. Our feeling about a team process may give us a better prediction of the outcome than a more complex weighting of cues about the outcome. Why is this the case? When and why do emotions give us information about complex processes in the aggregate?

In terms of (b), there are two understudied aspects of aggregation that we wish to investigate:

  1. Robustness of institutions Sometimes the way individuals form expectations is irrelevant for aggregate outcomes. Some institutions are robust to misguided expectation formation. Shyam Sunder's zero intelligence trader results (Sunder and Gode, 1993) point out that in some institutions, aggregate level outcomes are insensitive to the accuracy of individual expectations; important properties are consequences of institutional constraints, not individual behaviour.
  2. Coordination. When the focus is on aggregate outcomes, individual rationality may be of little help. In the prisoner's dilemma game, rationality does not guarantee that the 'good' outcome is chosen. Whether a debt crisis can be overcome depends on whether all agents are willing to concede a little; e.g., creditors taking a haircut and debtors paying even if they cannot be made to do so. This leads us to assume that emotions may have a functional role in aggregation processes. Emotions may help to coordinate to achieve 'better' outcomes.

While ultimately, we want to understand these issues in larger institutionally mediated interactions (like macroeconomies or large corporations), perhaps it is prudent for the workshop to start by aiming at either smaller groups of, say, 2 to 5 people (which can in principle still be modeled as a game), or the largest settings of perfect competition in markets.

The key conceptual shift away from 'behavioral game theory' is that the interaction is no longer described by a pay-off matrix. Rather, we can endow agents with specific individual behavior documented from neuroscience and psychology (and not invented ad hoc). Thus, empirically grounded models of human behavior will become the starting point of modeling aggregate economic behavior, and not assumptions about pay-offs.

Date and Location

January 17 - 19, 2017
Max Planck Institute for Mathematics in the Sciences
Inselstraße 22
04103 Leipzig
see travel instructions

Scientific Organizers

  • Timo Ehrig, MPI for Mathematics in the Sciences, Leipzig
  • Jürgen Jost, MPI for Mathematics in the Sciences, Leipzig
  • Thorbjørn Knudsen, Syddansk Universitet, Copenhagen
  • Rosemarie Nagel, Universitat Pompeu Fabra, Barcelona
  • Shyam Sunder, Yale, New Haven

Administrative Contact

Antje Vandenberg
MPI for Mathematics in the Sciences
Contact by Email

03.04.2017, 12:08