Bounded Rationality

Bounded rationality is a model of human decision-making and economic agency formulated by polymath and Nobel laureate Herbert A. Simon (1947, 1955, 1956) that posits that rational choice in real-world economic environments is constrained by three inescapable boundaries: incomplete and costly information about alternatives and consequences, the finite computational and working-memory capacity of the human mind, and the limited time available to reach a decision. Within business and economics theory, bounded rationality challenged the neoclassical postulate of Homo economicus—which endowed agents with unlimited substantive rationality capable of solving high-dimensional dynamic programming problems and calculating Bayesian inference posteriors instantaneously—and laid the intellectual foundation for behavioral economics (prospect theory), evolutionary economics, transaction cost economics, and artificial intelligence heuristic search.

Simon emphasized that bounded rationality is not "irrationality" or random error; rather, it is procedural rationality adapted to both the internal architecture of human cognition and the external statistical structure of the task environment. In his famous "scissors" metaphor (1990), Simon wrote that human rational behavior is shaped by a pair of scissors whose two blades are the structure of task environments and the computational capabilities of the actor: attempting to understand economic behavior by studying only cognitive biases or only environmental general equilibrium theory is like trying to cut paper with a single blade.

Substantive Optimization versus Simon's Satisficing Stopping Rule

In standard neoclassical choice theory, an agent confronts a known choice set X, possesses a complete, transitive preference ordering represented by a utility function U(x), and selects the global optimum:

When the choice set X is vast (for example, the ~10¹²⁰ possible game trees in chess, or millions of feasible engineering designs and financial portfolios) and discovering each option x_t incurs a positive search cost c > 0, global optimization becomes computationally intractable (NP-hard) or self-defeating once deliberative search costs are subtracted. Simon (1955) replaced global maximization with satisficing—a portmanteau of satisfy and suffice. Under satisficing, the agent establishes an endogenous aspiration level (reservation threshold) A_t across one or more payoff dimensions and evaluates options sequentially (x₁, x₂, ..., x_t) using an indicator payoff function V(x_t) ∈ {0, 1}:

The stopping time τ = min{ t ≥ 1 : U(x_t) ≥ A_t } terminates search at the first encountered alternative that meets or exceeds the aspiration threshold A_t. Moreover, the aspiration level is not static; it adapts dynamically over time according to the difficulty of search:

If satisfactory alternatives prove easy to find, the agent raises A_t+1; if prolonged search over t periods yields no option satisfying U(x) ≥ A_t, the agent gradually lowers the aspiration threshold A_t+1 (at decay rate δ > 0) to ensure that a choice is eventually reached within finite time. As George Stigler (1961) and Reinhard Selten (1998) subsequently analyzed, whereas optimal sequential search theory requires knowing the prior probability distribution F(u) of unseen payoffs to equate the marginal cost of search c to the expected marginal gain ∫_A^∞ (u − A) dF(u), Simon's adaptive satisficing rule operates effectively even under Knightian uncertainty when the distribution F(u) is unknown.

Two Divergent Research Traditions: Heuristics-and-Biases vs. Fast-and-Frugal Heuristics

Simon's concept of bounded rationality inspired two major, partially contrasting research programs in psychology and economics:

1. The Kahneman–Tversky Heuristics-and-Biases Program (1974): Daniel Kahneman and Amos Tversky investigated how judgmental heuristics—specifically representativeness (judging probability by similarity to a stereotype while neglecting base rates), availability (judging frequency by the ease with which instances come to mind), and anchoring and adjustment (insufficiently adjusting estimates away from an arbitrary initial number)—lead to systematic, predictable departures from the normative axioms of logic, Bayesian inference, and expected utility maximization. This program culminated in prospect theory, loss aversion, and Richard Thaler's nudge choice-architecture framework.
2. Gerd Gigerenzer's Fast-and-Frugal Heuristics and Ecological Rationality (1999): Gigerenzer and the ABC Research Group argued that heuristics are not second-best cognitive flaws, but ecologically rational tools that exploit the statistical structure of natural environments to outperform complex multi-variable regression models out of sample via the bias-variance trade-off:

When sample sizes are small or environments are non-stationary, a parameter-heavy optimization model suffers from high estimation variance (overfitting noise), whereas a zero-parameter fast-and-frugal heuristic—such as the Take-the-Best lexicographic cue heuristic (which inspects cues in order of validity and stops at the first cue that discriminates between two options) or the 1/N equal-weight portfolio heuristic (DeMiguel, Garlappi, and Uppal, 2009, which often outperforms sample-estimated Markowitz modern portfolio theory out of sample)—accepts a small structural bias to eliminate parameter estimation variance.

Incomplete Contracts, Transaction Cost Economics, and the Behavioral Theory of the Firm

In organizational and institutional economics, bounded rationality is the indispensable microfoundation for explaining why firms exist and why contracts are incomplete. As Oliver Williamson demonstrated in transaction cost economics (building on Ronald Coase and Herbert Simon), if economic agents possessed unbounded substantive rationality, partners to any complex supply relationship could write a complete contingent Arrow-Debreu contract at t = 0 specifying exact obligations, prices, and remedies for every conceivable future state of nature, eliminating ex-post holdup without needing vertical integration. Because bounded rationality makes it impossible to foresee, describe, and calculate all future contingencies—while opportunism creates moral hazard when uncontracted states arise—long-term contracts are necessarily incomplete, compelling firms to bring high-asset-specificity transactions inside corporate hierarchies.

Within the firm itself, Richard Cyert and James G. March extended Simon's framework in A Behavioral Theory of the Firm (1963), showing that corporations do not maximize a single unified profit function. Instead, a firm operates as a political coalition of stakeholders (managers, workers, shareholders, suppliers) with conflicting goals who resolve disputes through quasi-resolution of conflict, uncertainty avoidance (relying on short-run feedback-react standard operating procedures rather than long-run forecasts), problemistic search (initiating R&D or strategic change only when performance falls below an aspiration level A_t), and organizational slack.

Quantal Response Equilibrium, Level-k Thinking, and Sims's Rational Inattention

In modern game theory and macroeconomics, economists have developed tractable mathematical models that embed bounded rationality directly into equilibrium analysis rather than abandoning equilibrium altogether:

• Level-k and Cognitive Hierarchy Models (Nagel, 1995; Camerer, Ho, and Chong, 2004): Instead of assuming infinite common knowledge of rationality as in standard Nash equilibrium, Level-k models posit a hierarchy of strategic reasoning depths. Non-strategic Level-0 (L₀) players choose uniformly at random (or follow a salient anchor); Level-1 (L₁) players best-respond believing all opponents are L₀; and Level-k (L_k) players best-respond to L_k−1 (or to a Poisson distribution of lower levels m < k with mean τ ≈ 1.5 steps in Camerer et al.'s Cognitive Hierarchy model). In beauty-contest p-guessing games (where players guess 2/3 of the group average on [0, 100] and the unique Nash equilibrium is 0), Level-k thinking accurately predicts the empirical modal spikes at 33.3 (L₁) and 22.2 (L₂).
• Quantal Response Equilibrium (QRE, McKelvey and Palfrey, 1995): Replaces sharp best-response correspondence with stochastic logistic choice probabilities where better strategies are played more frequently, but noisy utility evaluation (scale parameter λ_QRE) prevents 100% deterministic optimization:

• Rational Inattention (Sims, 2003; Matějka and McKay, 2015): Nobel laureate Christopher Sims modeled a boundedly rational agent as a finite-capacity Shannon communication channel that can process at most κ bits of mutual information I(S; A) between the unknown economic state S and action A per unit time (or at marginal information cost μ > 0 per nat of Shannon entropy reduction). Remarkably, Matějka and McKay (2015) proved that an agent who optimally allocates scarce attention across alternatives under a Shannon entropy cost μ generates the exact multinomial logit choice rule, bridging information theory, neuroeconomics, and discrete choice econometrics.

How does Simon's bounded rationality differ from savage optimization under information search costs?

In George Stigler's (1961) neoclassical economics of information, an agent faces external search costs but retains infinite internal computational power, solving a second-order optimization problem to equate the marginal cost of acquiring information to its expected marginal benefit. As Sidney Winter (1971) and Reinhard Selten pointed out, this creates an infinite regress of optimization: deciding whether it is worth calculating the optimal search rule requires solving a still more complex meta-optimization problem. Simon's bounded rationality breaks this infinite regress by replacing optimization with non-optimizing procedural stopping rules (satisficing and fast heuristics).

Why do simple heuristics sometimes outperform complex econometric models in financial and business forecasting?

A complex model with p free parameters (such as a 50-asset Markowitz covariance matrix requiring 1,275 parameters) must estimate every parameter from noisy historical samples, introducing large estimation variance that degrades out-of-sample accuracy when environments shift. Simple heuristics—such as the 1/N equal-allocation rule or single-cue lexicographic rules—have zero estimated parameters (zero estimation variance), allowing them to beat complex models whenever sample size T is modest relative to parameter dimensionality p.

How does bounded rationality explain price stickiness in macroeconomics?

Under Christopher Sims's (2003) and Bartosz Maćkowiak and Mirko Wiederholt's (2009) rational inattention models, firm managers possess limited cognitive bandwidth κ and must divide their attention between tracking idiosyncratic firm-level demand/productivity shocks (which are large and costly to ignore) and tracking aggregate monetary policy and inflation shocks (which account for a smaller fraction of profit variance). Because managers rationally allocate most of their attention bits to idiosyncratic conditions, corporate prices respond quickly to firm-specific shocks but adjust sluggishly to aggregate monetary shocks, generating real output effects without assuming physical menu costs.

1. Simon, H. A. (1955). "A behavioral model of rational choice." Quarterly Journal of Economics, 69(1), 99–118. https://doi.org/10.2307/1884852
2. Simon, H. A. (1956). "Rational choice and the structure of the environment." Psychological Review, 63(2), 129–138. https://doi.org/10.1037/h0042769
3. Kahneman, D., & Tversky, A. (1974). "Judgment under uncertainty: Heuristics and biases." Science, 185(4157), 1124–1131. https://doi.org/10.1126/science.185.4157.1124
4. Gigerenzer, G., & Goldstein, D. G. (1996). "Reasoning the fast and frugal way: Models of bounded rationality." Psychological Review, 103(4), 650–669. https://doi.org/10.1037/0033-295X.103.4.650
5. Sims, C. A. (2003). "Implications of rational inattention." Journal of Monetary Economics, 50(3), 665–690. https://doi.org/10.1016/S0304-3932(03)00029-100029-1)