The Paradigm Shift: From Rational Actor to Boundedly Rational Agent

The central premise of neoclassical economics—that individuals consistently maximize expected utility—faces mounting empirical contradictions from controlled experiments and field observations. In a seminal 1979 study, participants chose between certain death of 600 people versus a one-third probability of saving all 600, with a two-thirds probability of saving none. Despite logically equivalent outcomes when framed positively (saving lives), 72% preferred the certain outcome; when framed negatively (causing deaths), 78% preferred the probabilistic option. This single manipulation reversed majority preferences, violating the principle of invariance—a cornerstone of rational choice theory. Such anomalies are not isolated curiosities but systematic patterns revealing that human decision-making operates under constraints and biases fundamentally different from the idealized rational actor model. This chapter formalizes the conflict between normative theory and descriptive reality, traces the intellectual genealogy of behavioral economics from Herbert Simon’s bounded rationality to Daniel Kahneman and Amos Tversky’s heuristics-and-biases program, and establishes the conceptual architecture—normative benchmarks, descriptive models, and prescriptive choice architecture—that structures this book’s investigation of three cornerstone effects: loss aversion, present bias, and the default effect.

The Collapse of the Rational Actor Model

The Formal Architecture of Expected Utility Theory

The neoclassical framework rests on the von Neumann-Morgenstern (vNM) expected utility theory, which provides a axiomatic foundation for rational choice under risk. An agent facing lotteries over outcomes $x \in X$ with probabilities $p_i$ possesses a utility function $u: X \rightarrow \mathbb{R}$ such that preferences over lotteries are represented by the expected utility $U(L) = \sum_{i} p_i u(x_i)$. The vNM axioms—completeness, transitivity, continuity, and independence—are necessary and sufficient conditions for such a representation. The independence axiom, in particular, mandates that if a lottery $L_1$ is preferred to $L_2$, then for any third lottery $L_3$ and any $\alpha \in (0,1]$, the mixed lottery $\alpha L_1 + (1-\alpha)L_3$ must be preferred to $\alpha L_2 + (1-\alpha)L_3$. This axiom encodes the principle of consequence: preferences should depend only on the outcomes and their probabilities, not on how options are described or bundled.

Mathematically, consider two lotteries:

• $L_A$: $p$ chance of $x$, $(1-p)$ chance of $y$
• $L_B$: $p$ chance of $z$, $(1-p)$ chance of $w$

If $x \succ z$ and $y \succ w$, then independence requires $L_A \succ L_B$. Any violation—where the addition of a common outcome changes the preference order—signals a failure of expected utility maximization. The Asian disease problem presented earlier is a direct test of this axiom: the “certain death” option and the “one-third probability of saving all” option are identical in expected lives lost (400), yet framing alters preferences by making the certain outcome feel like a sure loss versus the probabilistic option a potential gain.

Empirical Fractures: Allais, Ellsberg, and Beyond

Maurice Allais’s 1953 experiment provided the first rigorous demonstration of expected utility violations. In the “Allais paradox,” subjects choose between:

• **Option 1A**: 2,500 Francs with certainty
• **Option 1B**: 10/11 chance of 2,500 Francs, 1/11 chance of nothing

and separately between:

• **Option 2A**: 11/12 chance of 2,500 Francs, 1/12 chance of nothing
• **Option 2B**: 10/12 chance of 2,500 Francs, 2/12 chance of nothing

Expected utility theory predicts that if a risk-averse individual prefers 1A (certainty) over 1B, they must also prefer 2B over 2A, because both choices involve the same common consequence (10/11 or 10/12 chance of 2,500 Francs) and the independence axiom requires consistency. Yet in Allais’s original sample, 82% preferred 1A, but among those, 72% preferred 2A—a direct violation. The certainty effect—the overweighting of outcomes that are certain relative to probabilistic ones—emerges clearly.

Ellsberg’s 1961 urn experiments challenged subjective expected utility under ambiguity. Given an urn with 90 red balls and 10 balls that are either yellow or blue, subjects prefer betting on red (known probability 0.9) over betting on yellow (unknown probability between 0 and 0.1) even when the payoffs are identical. This ambiguity aversion violates the assumption that decision-makers can assign precise subjective probabilities to all events, revealing a preference for known risks over unknown risks—a pattern robust across cultures and contexts.

These paradoxes are not laboratory artifacts. Field evidence abounds: stock market investors hold losing positions too long (disposition effect) due to loss aversion; consumers dramatically overvalue warranties and insurance due to probability weighting; and retirement savings participation jumps when enrollment becomes the default option. The rational actor model, with its assumptions of stable preferences, unlimited cognitive resources, and consistent probabilistic reasoning, fails to predict behavior even in simple, high-stakes decisions.

The Computational Intractability of Full Rationality

Beyond experimental evidence, theoretical arguments from computer science and cognitive psychology undermine the feasibility of full rationality. Herbert Simon’s concept of bounded rationality posits that decision-makers are constrained by: 1. Cognitive limitations: Working memory capacity (~7±2 chunks), processing speed, and attentional resources. 2. Information constraints: Incomplete, noisy, or costly-to-acquire data. 3. Time constraints: Decisions often require fast heuristics rather than exhaustive optimization.

Consider the computational complexity of even simple choices. A consumer selecting a grocery cart among 100 items, each with 3 price-quality options, faces $3^{100} \approx 5 \times 10^{47}$ possible bundles—far exceeding the number of atoms in the observable universe. Full utility maximization is not merely difficult; it is algorithmically intractable for any finite-time agent. Simon argued that humans instead engage in “satisficing”—searching for an option that meets an aspiration level rather than optimizing globally. This shifts the question from “What is optimal?” to “What is good enough given constraints?”

Neoclassical economics responds by treating such constraints as external to the model—as “frictions” to be abstracted away. Behavioral economics, in contrast, internalizes these constraints, making them central to the explanation of choice patterns. The result is a paradigm shift: from a prescriptive theory of how choices *should* be made (given idealized conditions) to a descriptive science of how choices *are* made (given human cognition).

Historical Lineage: From Bounded Rationality to Heuristics-and-Biases

Herbert Simon’s Satisficing and Administrative Man

Herbert Simon’s 1947 work *Administrative Behavior* launched the systematic critique of the rational actor model. Simon distinguished between:

• **Substantive rationality**: Choosing the best possible action given constraints.
• **Procedural rationality**: Achieving good outcomes given the decision procedure’s limitations.

He introduced “bounded rationality” to describe the adaptation of decision procedures to environmental structures. Key concepts include:

• **Satisficing**: Setting an aspiration level $\alpha$ and choosing the first alternative that meets or exceeds $\alpha$.
• **Search termination**: Stopping search when marginal cost of additional information exceeds expected marginal benefit.
• **Adaptive toolbox**: Using heuristics tailored to specific environmental regularities (e.g., “recognition heuristic” in inferring city sizes).

Simon’s framework was computational and process-oriented, emphasizing that rationality is shaped