Bounded rationality, learning and expectations: A critical assessment
Abstract
The rational expectations hypothesis, central to modern macroeconomics, assumes that agents make optimal use of available information and correct their expectation errors immediately and accurately. However, Vernon Smith’s experimental evidence shows that individuals do not always correctly identify the source of their errors and that convergence towards efficient outcomes depends on institutional design and accumulated experience rather than perfect cognitive abilities. Complementarily, Siegwart Lindenberg’s RREEMM model offers a theoretical framework that incorporates cognitive constraints, limited resources, and active motivational frameworks, explaining why agents may detect misalignments without being able to correct them or even decide not to do so due to cognitive costs or normative and social priorities. This approach combines ecological and social rationality, integrating laboratory evidence with theoretical foundations of behavioural microeconomics, and offers a more realistic perspective on expectation formation and learning in complex environments. The results suggest that effective rationality is situational, adaptive and mediated by incentives and institutions, offering a bridge between normative models of rational expectations and empirical observations of human behaviour.
Keywords
Rational expectations | Bounded rationality | Ecological rationality | RREEMM | Microfoundations of behaviour
1. Introduction
Lucas and Sargent explain:
“The hypothesis of rational expectations is imposed here to adhere to the tenets of equilibrium theory. This requires agents to make the best possible use of their limited available information and assumes they know the pertinent objective probability distributions.” (Lucas & Sargent, 1979).
The rational expectations (RE) hypothesis is one of the pillars of modern macroeconomics based on general equilibrium theory. Under this conception, agents not only use available information optimally, but also form expectations that, on average, coincide with the objective probabilities of the economic environment. This implies that errors are random and non-systematic, and that markets, as a whole, coordinate the allocation of resources efficiently.
However, this formulation presupposes an extremely robust level of rationality: individuals must correctly identify when their expectations are incorrect, understand the exact nature of their errors, and adjust their decisions accordingly. This is a normative ideal that ignores the cognitive complexity and motivational limitations of real agents.
2. Difficulty identifying and correcting errors in expectations
Firstly, this implies not only that agents use all available information, but also that they correctly detect when their expectations are wrong, what type of error they made, and how to correct it. This idea presupposes a transparent learning process, without confusion about which variable failed or which mental model needs to be adjusted. However, it is clear that not only identifying an error in expectations, but also where the error is and what caused it, are fundamental correction mechanisms that present greater difficulty and, in turn, can generate friction in this adaptation.
3. Ecological rationality and experimental evidence (Vernon Smith)
Vernon Smith argues that optimization in real and experimental markets does not arise from perfect conscious calculation, and that “common information is not sufficient to yield common knowledge or expectations” (1994). In his Nobel Lecture, he explicitly asks the question: “What is the subject’s perception of the problem they’re trying to solve?” (2003), suggesting that the difficulty lies in the fact that subjects do not interpret the environment in an identical or transparent manner. In other words, even if they observe an error, they do not necessarily know what to correct.
4. Structural limitations of rationality (Siegwart Lindenberg and the RREEMM model)
Additionally, Siegwart Lindenberg (2015) demonstrates that expectations are structurally limited by resources, constraints, and motivational frameworks. The model he develops, called RREEMM (Resourceful-Restricted-Evaluating-Expecting-Maximizing-Man), replaces the figure of the omniscient agent with someone with multiple motivations, limited cognitive resources and real constraints. Therefore, even if the individual detects an error, they may not have the ability, motivation or clarity to correct it.
Lindenberg (2015) asserts that overarching goals can capture the entire mind and frame how people perceive and react to the world, leading to different behaviors when social contexts activate different goals (as discussed in goal-framing theory by Lindenberg & Steg, 2007). He explains that expectations are formed through the interaction between Restrictions (limited information), Resources (capabilities), Motivations (active goals) and Mechanisms (available heuristics). An agent may notice a mismatch without being able to pinpoint the cause because their goal frame (e.g., a goal frame of gain or affiliation) filters attention towards certain elements and ignores others. Therefore, cognitive updating is not perfect, but mediated by motivations and structural limits.
5. Costs and motivations in error correction
Rational expectations theory assumes that individuals always correct errors when they occur. Lindenberg argues that agents are satisfiers, not heroic optimisers, and that correcting errors can sometimes entail such a high cost that it is not done.
When the active goal is normative or affiliative, subjects may ignore relevant economic information, as their attention is directed towards social signals. Furthermore, a more detailed analysis of the error may be too costly. Thus, even in the presence of clear evidence, individuals do not correct, not because they are irrational, but because they act within a motivational and economic structure that makes it reasonable not to invest in additional analysis.
6. Connection with the criticism of Sargent and Lucas
Precisely, the criticism of Sargent and Lucas focuses on the search for a more robust realism with greater explanatory power than Keynesian macroeconomics, and what is presented here, in a similar spirit, seeks only to bring that search for realism to the assumptions about rationality in decision-making and in the formulation of expectations.
7. Conclusion
In summary, criticism of the rational expectations hypothesis arises from a clear empirical observation: agents do not correct ‘the right mistakes,’ but only those they can detect from their heuristics and institutional constraints, as shown by Vernon Smith’s experimental evidence. Rationality is situated, ecological, and dependent on the design of the environment, not an internal and omniscient property of the individual. Lindenberg’s RREEMM model allows us to refine this intuition by conceiving of the actor as a ‘restricted maximiser’ whose cognitive resources, norms, interpretative frameworks and opportunities determine what information they can process and what kind of learning is possible. This perspective better integrates the real limitations observed in the laboratory and in the field, and explains why errors can be persistent and systematic without violating the rationality of the agent.
References1
Google. (2025). Gemini Flash 2.5 (Flash 2.5 version) [Large language model].
Google. (2024). NotebookLM [Large language model].
Lindenberg, S. (2015). From individual rationality to socially embedded self-regulation. In R. Scott & S. Kosslyn (Eds.), Emerging trends in the social and behavioral sciences (etrds0136). John Wiley and Sons. https://doi.org/10.1002/9781118900772.etrds0136
Lucas, R. E. (1976). Econometric Policy Evaluation: A Critique. Carnegie-Rochester Conference Series on Public Policy, 1, 19–46
Lucas, R. E., Jr., & Sargent, T. J. (1979). After Keynesian macroeconomics. Quarterly Review, 3(2), 1–16. Federal Reserve Bank of Minneapolis
Ravier, A. O. (2018). In search of full employment: Studies in Austrian macroeconomics and comparative economics. Unión Editorial, S.A. https://puntodevistaeconomico.com/wp-content/uploads/2018/06/ravier-en-busca-del-pleno-empleo-pdf.pdf
Smith, V. L. (2003). Constructivist and ecological rationality in economics. The American Economic Review, 93(3), 465–508. https://doi.org/10.1257/000282803322156954
Smith, V. L. (1994). Economics in the laboratory. Journal of Economic Perspectives, 8(1), 113–131.
Footnotes
AI Tools Used
The authors utilized large language models (LLMs) exclusively as research and synthesis assistants. Specifically, Gemini Flash 2.5 (Google, 2025) and NotebookLM (Google, 2024) were employed to aid in the structural synthesis of existing research and to ensure adherence to APA 7th edition citation guidelines. All interpretive analysis and final conclusions remain the sole work of the authors.↩︎