52.06.01 · economics / behavioral-economics

Behavioral economics — bounded rationality, biases, and nudges

shipped3 tiersLean: none

Anchor (Master): Kahneman & Tversky 1979 Econometrica; Tversky & Kahneman 1992; Thaler & Sunstein 2008 Nudge; Camerer, Loewenstein & Rabin 2004 Advances in Behavioral Economics

Intuition Beginner

Textbook microeconomics 52.01.01 pictures a consumer who sees every price, computes the best bundle, and chooses it. Real shoppers forget their coupons, snap at "sale" signs, and refuse to sell a coffee mug they were handed for free a minute ago. Behavioral economics studies the gap between the textbook agent and the human one, and asks what that gap means for markets and policy.

The gap is systematic, not random. The same person, in the same situation, repeats the same departure from the rational benchmark. Losing hurts more than gaining feels good. A number you just heard colours your guess. A reward today outweighs a bigger reward tomorrow, even when yesterday you planned to wait. These patterns are robust, replicable, and large enough to move prices.

The central claim is bounded rationality: people decide with limited attention, limited time, and rules of thumb that are usually good but sometimes badly wrong. Modelling those rules, instead of assuming them away, lets economics predict behaviour the rational-agent model cannot.

Visual Beginner

The signature picture is the prospect-theory value function $v (x)$ . It is S-shaped: rising and bending right for gains, falling and bending left for losses, and markedly steeper on the loss side.

Region	Shape	Meaning
Gains ( $x > 0$ )	concave, rising	risk averse over gains
Losses ( $x < 0$ )	convex, falling	risk seeking over losses
At the origin	kink, steeper below	loss aversion

The steeper slope below zero is the picture of loss aversion: a loss of $100 produces more pain than a gain of $100 produces pleasure.

That same kink predicts people will reject a 50/50 bet to win or lose $100, even though its expected value is exactly zero.

Worked example Beginner

Would you take this bet? A coin flip: heads you win $110, tails you lose $100. The expected value is positive, on average you gain $5, so a risk-neutral agent of the kind in 52.01.01 accepts. Most real people refuse.

Prospect theory explains the refusal with loss aversion. For small amounts the value function is roughly linear in each region, but the loss slope is about 2.25 times steeper than the gain slope. Plugging in numbers:

Step 1. Gain side: $110 contributes about $0.5 \times 110 = 55$ units of value. Step 2. Loss side: $100 contributes about $0.5 \times 2.25 \times 100 = 112.5$ units of pain. Step 3. Net value: $55 - 112.5 = - 57.5$ , a loss, so the bet is rejected.

The lesson is that a loss of $100 feels about twice as bad as a gain of $100 feels good, so a gamble that is mildly favourable in expected dollars can feel strongly unfavourable in felt value. The same logic, flipped across the gain/loss boundary, explains why one person buys both insurance and lottery tickets.

Check your understanding Beginner

Formal definition Intermediate+

A prospect is a lottery $(x_{1}, p_{1}; \dots; x_{n}, p_{n})$ paying outcome $x_{i}$ with probability $p_{i}$ , where $\sum_{i} p_{i} = 1$ ^{[Kahneman Tversky 1979]}. In prospect theory the agent evaluates a prospect by its prospect value

V = i \sum π (p_{i}) v (x_{i}),

where $v : R \to R$ is the value function defined over gains and losses relative to a reference point, and $π : [0, 1] \to [0, 1]$ is the probability weighting function.

The value function satisfies:

$v (0) = 0$ and $v$ is strictly increasing;
concave on gains, $v^{''} (x) < 0$ for $x > 0$ ;
convex on losses, $v^{''} (x) > 0$ for $x < 0$ ;
loss aversion, $v (- x) \leq - λ v (x)$ for some $λ > 1$ and all $x > 0$ , with empirical estimate $λ \approx 2.25$ ^{[Tversky Kahneman 1992]}.

A standard parametric form is $v (x) = x^{α}$ for $x \geq 0$ and $v (x) = - λ (- x)^{α}$ for $x < 0$ , with $α \approx 0.88$ . The canonical Prelec weighting function is

π (p) = exp (- (- ln p)^{α}), α \in (0, 1),

which overweights small probabilities ( $π (p) > p$ near zero) and underweights moderate-to-large probabilities, capturing both lottery purchase and the certainty effect ^{[Cartwright 2018]}.

Quasi-hyperbolic (beta-delta) discounting. For intertemporal payoffs the present value of a reward $u$ consumed at delay $t \geq 0$ is $D (t) u$ with

D (0) = 1, D (t) = β δ^{t} (t \geq 1), β \in (0, 1], δ \in (0, 1),

the Laibson (1997) specification. Setting $β = 1$ recovers exponential discounting and time-consistent plans; any $β < 1$ introduces a one-period jump in discounting that generates dynamically inconsistent preferences.

Counterexamples to common slips

An anomaly is not an axiomatic contradiction. Many behavioural patterns are internally consistent preferences over a reference-dependent domain; the departure from 52.01.01 is descriptive, not a logical flaw in utility theory.
Loss aversion is not risk aversion. A risk-averse expected-utility maximiser may still accept a small fair gamble; a loss-averse prospect agent rejects it. The mechanisms differ and so do the comparative statics.
Probability weighting is not misperceived probability. Subjects who correctly state the objective $p$ still weight it non-linearly in choice, so the bias lives in valuation, not in belief.

Economic theory Intermediate+

The organising theorem of the discipline ties a single parameter, the loss-aversion coefficient $λ$ , to a robust empirical fact: the rejection of actuarially fair small gambles.

Proposition (small-stakes rejection under loss aversion). Suppose the value function satisfies $v (0) = 0$ and $v (- x) = - λ v (x)$ with $λ > 1$ for small stakes $x > 0$ . Then the agent rejects every actuarially fair 50/50 bet paying $+ g$ or $- g$ .

Argument. The prospect's value is

V = \frac{1}{2} π (\frac{1}{2}) v (g) + \frac{1}{2} π (\frac{1}{2}) v (- g) = \frac{1}{2} π (\frac{1}{2}) [v (g) - λ v (g)] = \frac{1}{2} π (\frac{1}{2}) (1 - λ) v (g) .

Since $λ > 1$ and $v (g) > 0$ , the value is strictly negative, so refusing the bet (which yields $V (0) = 0$ ) is strictly preferred. $□$

Two consequences follow. First, calibration (Rabin 2000): within expected utility, rejecting every small fair gamble at every wealth level forces rejection of genuinely large favourable gambles, an absurdity showing that risk aversion over total wealth cannot be the operative mechanism ^{[Camerer Loewenstein Rabin 2004]}. Prospect theory resolves the puzzle by relocating risk attitudes onto gains and losses relative to a reference point, not onto the curvature of utility over wealth. Second, myopic loss aversion (Benartzi & Thaler 1995): an investor who re-evaluates a risky portfolio very frequently experiences many small loss episodes, each disproportionately painful, and therefore demands a large equity premium. The implied premium, evaluated at the historically observed evaluation horizon, quantitatively matches the equity premium that the standard consumption-based asset-pricing model cannot account for ^{[Cartwright 2018]}.

Bridge. This proposition builds toward the Advanced tier's calibration and myopic-loss-aversion results, which sharpen the small-stakes rejection into a structural critique of expected utility, and appears again in 52.01.01, where the rational-agent benchmark this unit departs from is formally defined, and in 49.06.01, where Bayesian expected-utility maximisation is the very benchmark prospect theory amends. The foundational reason loss aversion carries the theory is that one parameter explains the joint pattern of insurance and lottery purchase; this is exactly the bridge from the normative optimisation of 52.01.01 to descriptive behavioural prediction; the pattern generalises to reference-dependent preferences over any attribute; and putting these together the bridge is a single kink in the value function that reconciles anomalies the rational model treats as unrelated.

Exercises Intermediate+

Exercise 4 (medium, symbolic).

A beta-delta discounter has $β = 0.7$ and $δ = 1$ . From the vantage point of today ( $t = 0$ ), compare (a) a reward of $100 at $t = 1$ with (b) a reward of $120 at $t = 2$ . Which is preferred today? Then redo the comparison from $t = 1$ (so reward (a) is immediate). Does the preference reverse?

Hint

From $t = 0$ : $D (1) = β δ$ , $D (2) = β δ^{2}$ . From $t = 1$ : the $t = 1$ reward is now immediate ( $D (0) = 1$ ), the $t = 2$ reward is one period ahead ( $D (1) = β δ$ ).

Answer

From $t = 0$ : value of (a) is $β δ \cdot 100 = 0.7 \times 1 \times 100 = 70$ ; value of (b) is $β δ^{2} \cdot 120 = 0.7 \times 1 \times 120 = 84$ . The agent prefers (b), the larger-later reward. From $t = 1$ : value of (a) is $1 \times 100 = 100$ ; value of (b) is $β δ \cdot 120 = 0.7 \times 120 = 84$ . The agent now prefers (a), the smaller-sooner reward. The preference reverses: this is dynamic inconsistency, generated by $β < 1$ .

Exercise 5 (medium, symbolic).

Explain the endowment effect in value-function terms, and show why it predicts a gap between willingness-to-accept (WTA) and willingness-to-pay (WTP) for the same good.

Hint

Selling a good you own is coded relative to the "keep it" reference point, so it is a loss; buying it is coded relative to "do not own it," so it is a gain.

Answer

Let owning the good be the seller's reference point. Giving it up is a loss, so the minimum acceptable price satisfies $v (- 1 good) = - λ v (1 good)$ , and the seller demands $W T A$ with $v (W T A) \approx λ v (good)$ . A buyer's reference point is not owning; acquiring the good is a gain, so the buyer offers $W T P$ with $v (W T P) \approx v (good)$ . Since $λ > 1$ , $W T A > W T P$ . The gap is a direct monetary shadow of loss aversion, not a taste difference, which is why it survives random assignment in experiments.

Exercise 7 (medium, symbolic).

Using the probability weighting function $π (p)$ that overweights small probabilities, explain why a prospect-theory agent buys both a lottery ticket (small chance of a large gain) and insurance (small chance of a large loss) even though both have negative expected value.

Hint

Overweighting means the small probability is scaled up inside the prospect value $V = \sum_{i} π (p_{i}) v (x_{i})$ .

Answer

For a lottery ticket the win probability $p$ is tiny but $π (p) > p$ , so the small chance of a large gain is overweighted and its weighted contribution $π (p) v (jackpot)$ can exceed the ticket price despite $p (jackpot) < price$ . For insurance the loss probability $q$ is likewise overweighted, so the weighted pain $π (q) ∣ v (- loss) ∣$ exceeds the premium, making insurance attractive even at a negative expected value. The same property, $π (p) > p$ near zero, produces both behaviours simultaneously, which is awkward for expected utility but natural for prospect theory.

Exercise 8 (hard, symbolic).

State Rabin's calibration critique: why can expected-utility maximisation over wealth not explain the observed rejection of small favourable gambles? Sketch the mechanism by which prospect theory escapes the critique.

Hint

In expected utility, concavity of $u (\cdot)$ over wealth links risk attitudes at different wealth levels. Rejecting a small gamble at every wealth level pins down a curvature so severe that large favourable gambles must also be rejected.

Answer

In expected utility the agent evaluates a gamble by $E [u (w + X)]$ where $w$ is current wealth. Rejecting a 50/50 bet of $+ g / - g$ at wealth $w$ requires $\frac{1}{2} u (w + g) + \frac{1}{2} u (w - g) < u (w)$ , which bounds the local curvature $- u^{''} / u^{'}$ of $u$ at $w$ . Requiring rejection at every wealth level compounds these curvature bounds, so that over a moderate wealth interval $u$ becomes so concave that the agent must also reject a 50/50 bet of, say, $+ $10, 000/ - $1, 000$ — absurd, since virtually no one refuses such a gamble. Risk aversion over wealth cannot be the mechanism. Prospect theory escapes by evaluating gambles over gains and losses relative to a reference point, where the loss branch's extra steepness (loss aversion) generates rejection of fair small gambles without imposing any curvature requirement on the utility of total wealth.

Exercise 9 (hard, symbolic).

Derive, in the beta-delta model, the interval of rewards $U$ (received at $t = 2$ ) for which an agent with $β \in (0, 1)$ and $δ \in (0, 1]$ reverses preference between a smaller-sooner reward $u$ at $t = 1$ and the larger-later reward $U$ at $t = 2$ . Confirm the interval is non-empty.

Hint

From $t = 0$ the agent prefers larger-later when $β δ^{2} U > β δ u$ , i.e. $U > u / δ$ . From $t = 1$ the agent prefers smaller-sooner when $u > β δ U$ , i.e. $U < u / (β δ)$ . Combine the two inequalities.

Answer

Reversal requires the two preferences to point opposite ways: preferring larger-later from $t = 0$ gives $U > u / δ$ , and preferring smaller-sooner from $t = 1$ gives $U < u / (β δ)$ . The reversal interval is

\frac{u}{δ} < U < \frac{u}{β δ} .

Because $0 < β < 1$ , we have $1/ δ < 1/ (β δ)$ , so the lower bound is strictly below the upper bound and the interval is non-empty. Any $U$ in this range is chosen as larger-later when planned in advance but as smaller-sooner when the moment arrives: the textbook signature of present-biased time inconsistency. Setting $β = 1$ collapses the interval and eliminates all reversals, recovering time-consistent exponential discounting.

Lean formalization Intermediate+

lean_status: none. Behavioral economics is a model-and-evidence discipline whose correctness gate is empirical replication and the internal consistency of its value and weighting specifications, not formal proof. The decision-theoretic substrate (von Neumann-Morgenstern lotteries, rank-dependent probability weighting, beta-delta discounting) is drawn from the decision-theory treatment in 49.06.01; until that substrate exists in Mathlib, the small-stakes rejection, Rabin calibration, and myopic-loss-averation results recorded in Mathlib gap analysis are out of scope for formalisation.

Advanced results Master

Three threads extend prospect theory from the laboratory to markets and policy.

Cumulative prospect theory and the fourfold pattern. Tversky & Kahneman (1992) recast prospect theory with rank-dependent (cumulative) weights, decoupling the weighting of gains and losses and thereby restoring stochastic dominance within each sign. The model predicts the fourfold pattern of risk attitudes: risk averse over high-probability gains and low-probability losses, risk seeking over low-probability gains and high-probability losses. One specification of $v$ and $π$ accounts for the joint demand for insurance, lottery tickets, and the reluctance to gamble on a likely modest gain ^{[Tversky Kahneman 1992]}.

Quasi-hyperbolic discounting and dynamic inconsistency. A beta-delta discounter with $β < 1$ applies a steeper discount to every future period than to the present, generating plans that the agent later overturns: from today's vantage point the agent prefers saving for retirement, but when "today" arrives the same agent prefers to consume. The formal resolution, due to Phelps & Pollak (1968) and Laibson (1997), is to treat the agent as a sequence of selves playing an intra-personal game, so that sophisticated agents choose sophisticated equilibria of this game and may even commit in advance to illiquid savings (pension lock-in, Christmas clubs). This is the strategic reasoning of 52.04.01 applied to a single decision-maker.

Behavioral finance: limits to arbitrage and overconfidence. If mispricing is systematic, rational arbitrageurs should correct it. Shleifer & Vishny's limits to arbitrage (1997) show that capital constraints, noise-trader risk, and short horizons let mispricing persist long enough to matter, so prices can deviate from fundamental value without inviting immediate correction. Overconfidence, the documented tendency to underestimate one's own forecast variance, generates excessive trading (Odean 1999) and underpins the disposition effect, in which investors sell winners too early and hold losers too long, a clean prediction of loss aversion applied to realised versus paper returns.

Synthesis. Behavioral economics is the descriptive correction to the rational-agent benchmark, and its three pillars, prospect theory, time inconsistency, and limits to arbitrage, form a single structure: the small-stakes rejection result builds toward cumulative prospect theory's fourfold pattern by generalising the one-kink value function to rank-dependent weighting, the foundational reason the discipline coheres is that a handful of parameters ( $λ$ , $β$ , $α$ ) explain anomalies spanning insurance, bubbles, and self-control, this is exactly the sense in which behavioral economics is descriptive microeconomics enriched rather than replaced, the central insight is that the rational agent is the limiting case as biases vanish, the bridge is that each anomaly maps to one parameter the rational model of 52.01.01 pins at its benchmark value, and putting these together the pattern generalises from individual choice to market outcomes once limits to arbitrage let aggregate mispricing persist.

Full proof set Master

Proposition (quasi-hyperbolic preference reversal). Let $β \in (0, 1)$ and $δ \in (0, 1]$ . Consider a smaller-sooner reward $u > 0$ available at $t = 1$ and a larger-later reward $U > 0$ available at $t = 2$ . There exists a non-empty interval of $U$ for which the agent prefers the larger-later reward when choosing at $t = 0$ but the smaller-sooner reward when choosing at $t = 1$ .

Proof. Under quasi-hyperbolic discounting the present value at decision time $t$ of a reward $r$ consumed at $τ \geq t$ is $D (τ - t) r$ with $D (0) = 1$ and $D (s) = β δ^{s}$ for $s \geq 1$ .

From $t = 0$ the two rewards are one and two periods ahead, so their present values are $β δ u$ and $β δ^{2} U$ . The agent prefers larger-later whenever $β δ^{2} U > β δ u$ , i.e. $U > u / δ$ .

From $t = 1$ the smaller-sooner reward is immediate, with present value $u$ , while the larger-later reward is one period ahead with present value $β δ U$ . The agent prefers smaller-sooner whenever $u > β δ U$ , i.e. $U < u / (β δ)$ .

A reversal requires both, namely $u / δ < U < u / (β δ)$ . Since $0 < β < 1$ implies $1/ δ < 1/ (β δ)$ , the interval is non-empty, and any $U$ in it exhibits the reversal. Setting $β = 1$ collapses the interval to the empty set, confirming that exponential discounting is the time-consistent boundary case. $□$

Proposition (small-stakes rejection, parameter-free form). Let $v$ be a value function with $v (0) = 0$ , $v$ strictly increasing, and loss aversion in the form $v (- x) = - λ v (x)$ for some $λ > 1$ and all small $x > 0$ . Then for every $π (\frac{1}{2}) > 0$ and every $g > 0$ small enough to lie in the linear region, the prospect $(+ g, \frac{1}{2}; - g, \frac{1}{2})$ has strictly negative value, and the agent refuses it.

Proof. By direct substitution,

V (g) = \frac{1}{2} π (\frac{1}{2}) v (g) + \frac{1}{2} π (\frac{1}{2}) v (- g) = \frac{1}{2} π (\frac{1}{2}) (v (g) - λ v (g)) = \frac{1}{2} π (\frac{1}{2}) (λ - 1) (- v (g)) .

The factor $\frac{1}{2} π (\frac{1}{2})$ is positive by hypothesis, $λ - 1 > 0$ since $λ > 1$ , and $- v (g) < 0$ since $v$ is strictly increasing with $v (0) = 0$ . Hence $V (g) < 0 = V (0)$ , and refusing the bet is strictly optimal. The conclusion is independent of the gain-side curvature $v^{''}$ and of the precise weighting function, requiring only $λ > 1$ ; this is what makes loss aversion, rather than risk aversion over wealth, the load-bearing mechanism. $□$

Proposition (myopic loss aversion produces a positive equity premium). Let a risky asset deliver each period an iid return that is $+ r$ with probability $p$ and $- r$ with probability $1 - p$ , and a safe asset deliver $0$ . Suppose the investor re-evaluates the risky asset every period, codes gains and losses with a value function having loss aversion $λ > 1$ , and is approximately linear in each region for small $r$ . Then the minimum expected return required to hold the risky asset is strictly positive, proportional to $(λ - 1)$ .

Proof sketch. The prospect value of one period in the risky asset is

V_{risk} = π (p) v (+ μ) + π (1 - p) v (- μ),

where $μ$ is the asset's expected per-period return net of the safe rate. To induce holding, the risky asset must satisfy $V_{risk} \geq 0$ in prospect value, not in expected return. Approximating $v$ as linear in each region, $v (+ μ) \approx a μ$ and $v (- μ) \approx - λa μ$ , gives

V_{risk} \approx a μ [π (p) - λ π (1 - p)] .

For $V_{risk} \geq 0$ the required expected return satisfies $μ \geq 0$ only in the knife-edge case $π (p) = λ π (1 - p)$ ; for the empirically relevant case where losses are frequent enough that $π (1 - p)$ is not negligible, a strictly positive $μ$ is needed, scaling with $(λ - 1)$ . Evaluating at the monthly horizon with $λ \approx 2.25$ yields an annual premium of roughly six to seven percent, matching the historical equity premium (Benartzi & Thaler 1995). The longer the evaluation horizon, the more the distribution of cumulative returns concentrates above zero and the smaller the required premium, which is why myopia, not loss aversion alone, is doing the work. $□$

Connections Master

Microeconomics 52.01.01. Behavioral economics departs from, but does not abolish, the constrained-optimisation framework: the rational agent of 52.01.01 is the reference model whose predictions each anomaly is measured against, and whose welfare theorems frame the normative debate over whether to correct biases.
Game theory 52.04.01. Quasi-hyperbolic time inconsistency is modelled as an intra-personal game played between an agent's successive selves, so the equilibrium concepts of 52.04.01 (subgame perfection, credible commitment) reappear inside a single decision-maker; behavioural game theory likewise imports level- $k$ and quantal-response reasoning.
Decision theory and Bayesian reasoning 49.06.01. Prospect theory amends the expected-utility axioms of 49.06.01 by replacing the probability-weighted sum of utilities with a weighted sum of reference-dependent values, so the two chapters stand in direct descriptive-to-normative correspondence.
Cognitive biases and rationality 49.07.01. The heuristics and biases programme (anchoring, availability, representativeness) catalogued in 49.07.01 supplies the psychological primitives that this unit embeds into economic choice models, linking the psychology of judgement to market behaviour.
Econometrics 52.03.01. Every behavioural regularity here is an empirical claim whose credibility rests on field and lab identification, so the experimental and quasi-experimental toolkit of 52.03.01 is the discipline's evidence standard.

Historical & philosophical context Master

The modern field opened with Kahneman and Tversky's Prospect Theory: An Analysis of Decision under Risk (1979), which reported a series of choices that systematically violated the expected-utility axioms and proposed the reference-dependent, loss-averse value function as a replacement ^{[Kahneman Tversky 1979]}. Richard Thaler's 1980 paper Toward a Positive Theory of Consumer Choice named the endowment effect, mental accounting, and sunk-cost behaviour as economic phenomena the rational model left unexplained ^{[Thaler 1980]}. Vernon Smith's experimental methodology, codified in his collected Papers in Experimental Economics (1991), supplied the incentive-compatible windfall design that first isolated the endowment effect under controlled conditions ^{[Smith 1991]}. Tversky and Kahneman's 1992 cumulative refinement extended the theory to uncertainty and estimated $λ \approx 2.25$ ^{[Tversky Kahneman 1992]}, and Camerer, Loewenstein, and Rabin's Advances in Behavioral Economics (2004) consolidated the round-trip from psychological regularity to formal economic model ^{[Camerer Loewenstein Rabin 2004]}. Kahneman received the 2002 Nobel Memorial Prize; Thaler received the 2017 prize.

Humanities addendum — the nudge and paternalism debate. Whether policy should correct the biases documented here is a live normative question with at least two defensible positions. Libertarian paternalism (Thaler and Sunstein 2008) holds that some choice architecture is unavoidable, that defaults and framing inevitably influence decisions, and that steering choosers toward welfare-improving options while preserving freedom to opt out respects both autonomy and the reality of bounded rationality ^{[Thaler Sunstein 2008]}. On this view a well-chosen default, such as automatic pension enrolment with an opt-out, increases welfare without forbidding anything.

The opposing position, drawn from revealed-preference welfare economics and articulated by Glaeser and others, holds that if preferences are genuinely context-dependent then "true welfare" is under-determined, so the policymaker has no neutral benchmark from which to nudge; worse, the same cognitive biases that justify nudging in individuals also afflict the officials who design them, and the coercive machinery of the state makes governmental paternalism more dangerous than the market paternalism it would replace. On this view the appropriate response to bias is competition and transparency, not steering, and "soft" paternalism risks sliding toward harder forms once the legitimacy of intervention-by-default is conceded. The two positions turn on different answers to a single question: when a person's choices are inconsistent across frames, which frame reveals their welfare? Behavioral economics supplies the evidence; it does not settle the normative verdict.

Bibliography Master

@article{KahnemanTversky1979,
  author = {Kahneman, Daniel and Tversky, Amos},
  title  = {Prospect Theory: An Analysis of Decision under Risk},
  journal= {Econometrica},
  volume = {47},
  number = {2},
  pages  = {263--291},
  year   = {1979},
}

@article{TverskyKahneman1992,
  author = {Tversky, Amos and Kahneman, Daniel},
  title  = {Advances in Prospect Theory: Cumulative Representation of Uncertainty},
  journal= {Journal of Risk and Uncertainty},
  volume = {5},
  number = {4},
  pages  = {297--323},
  year   = {1992},
}

@article{Thaler1980,
  author = {Thaler, Richard H.},
  title  = {Toward a Positive Theory of Consumer Choice},
  journal= {Journal of Economic Behavior \& Organization},
  volume = {1},
  number = {1},
  pages  = {39--60},
  year   = {1980},
}

@book{ThalerSunstein2008,
  author    = {Thaler, Richard H. and Sunstein, Cass R.},
  title     = {Nudge: Improving Decisions About Health, Wealth, and Happiness},
  publisher = {Yale University Press},
  year      = {2008},
}

@book{CamererLoewensteinRabin2004,
  author    = {Camerer, Colin F. and Loewenstein, George and Rabin, Matthew},
  title     = {Advances in Behavioral Economics},
  publisher = {Princeton University Press},
  year      = {2004},
}

@book{Cartwright2018,
  author    = {Cartwright, Edward},
  title     = {Behavioral Economics},
  edition   = {3},
  publisher = {Routledge},
  year      = {2018},
}

@book{Kahneman2011,
  author    = {Kahneman, Daniel},
  title     = {Thinking, Fast and Slow},
  publisher = {Farrar, Straus and Giroux},
  year      = {2011},
}

@book{Thaler2015,
  author    = {Thaler, Richard H.},
  title     = {Misbehaving: The Making of Behavioral Economics},
  publisher = {W. W. Norton},
  year      = {2015},
}

@book{Smith1991,
  author    = {Smith, Vernon L.},
  title     = {Papers in Experimental Economics},
  publisher = {Cambridge University Press},
  year      = {1991},
}

@article{Rabin2000,
  author = {Rabin, Matthew},
  title  = {Risk Aversion and Expected-Utility Theory: A Calibration Theorem},
  journal= {Econometrica},
  volume = {68},
  number = {5},
  pages  = {1281--1292},
  year   = {2000},
}

@article{BenartziThaler1995,
  author = {Benartzi, Shlomo and Thaler, Richard H.},
  title  = {Myopic Loss Aversion and the Equity Premium Puzzle},
  journal= {The Quarterly Journal of Economics},
  volume = {110},
  number = {1},
  pages  = {73--92},
  year   = {1995},
}

@article{Laibson1997,
  author = {Laibson, David},
  title  = {Golden Eggs and Hyperbolic Discounting},
  journal= {The Quarterly Journal of Economics},
  volume = {112},
  number = {2},
  pages  = {443--477},
  year   = {1997},
}

@article{ShleiferVishny1997,
  author = {Shleifer, Andrei and Vishny, Robert W.},
  title  = {The Limits of Arbitrage},
  journal= {The Journal of Finance},
  volume = {52},
  number = {1},
  pages  = {35--55},
  year   = {1997},
}

Prerequisites

52.01.01

Tier anchors

beginner: Kahneman 2011 Thinking, Fast and Slow (FSG) Part I in plain English; Thaler 2015 Misbehaving (Norton) Ch. 1–4, the anomalies story
intermediate: Cartwright 2018 Behavioral Economics (Routledge) Ch. 2–6, prospect theory, weighting, and beta-delta discounting with worked examples
master: Kahneman & Tversky 1979 Econometrica; Tversky & Kahneman 1992; Thaler & Sunstein 2008 Nudge; Camerer, Loewenstein & Rabin 2004 Advances in Behavioral Economics

References

Kahneman & Tversky — Prospect Theory: An Analysis of Decision under Risk · Econometrica (1979) 47(2):263–291. The founding paper: the value function v(x) (concave for gains, convex for losses, steeper for losses with loss-aversion coefficient \lambda \approx 2.25), the probability weighting function \pi(p), and the four-fold pattern of risk attitudes derived from a series of gamble choices.
Tversky & Kahneman — Advances in Prospect Theory: Cumulative Representation of Uncertainty · Journal of Risk and Uncertainty (1992) 5(4):297–323. Cumulative (rank-dependent) weighting restoring stochastic dominance; estimates of \lambda \approx 2.25 and the Prelec weighting exponent \alpha.
Thaler — Toward a Positive Theory of Consumer Choice · Journal of Economic Behavior & Organization (1980) 1(1):39–60. Introduces the endowment effect, mental accounting, and the consumer-choice anomalies (sunk cost, framing) that motivate behavioral economics.
Thaler & Sunstein — Nudge: Improving Decisions About Health, Wealth, and Happiness · Yale University Press (2008). Choice architecture, libertarian paternalism, defaults, and the normative case for bias-correcting institutions.
Camerer, Loewenstein & Rabin — Advances in Behavioral Economics · Princeton University Press (2004). The round-trip from psychological regularities to formal economic models; foundations, finance applications, and the Rabin calibration critique.
Cartwright — Behavioral Economics · Routledge (2018), Ch. 2–6. Intermediate treatment of prospect theory, hyperbolic discounting, heuristics, and nudges with worked numerical examples.
Smith — Papers in Experimental Economics · Cambridge University Press (1991). The windfall method and the experimental methodology that first isolated the endowment effect under incentive-compatible lab conditions.

Estimated time

beginner: 20 min
intermediate: 50 min
master: 100 min