Loss aversion is the asymmetric evaluation of gains and losses relative to a reference point: an equal-sized loss has a larger negative impact on subjective value than the gain has a positive impact. For a quant studying investor choices, that reference point can change the model's prediction even when the financial outcomes stay fixed. Rejection of a fair gamble alone does not identify loss aversion; risk aversion can also produce rejection.
First specify what counts as a loss
A reference point is the benchmark against which an outcome is coded as a gain or loss. It might be current wealth, a purchase price or an expectation. These are modelling hypotheses to check, not interchangeable definitions. An account ending at USD 9,000 is a gain relative to USD 8,000 and a loss relative to USD 10,000. The ending money has not changed; its classification has. Kahneman and Tversky discuss both reference dependence and shifts of reference in their original prospect-theory paper, pp. 277–279 and 286–288.
Isolating the asymmetry
Take a fictional value function for a monetary change from a fixed reference point:
The loss-side slope is twice the gain-side slope. Here the multiplier is chosen for teaching: it is neither a universal empirical constant nor an expected-utility risk-aversion coefficient. Value is measured in arbitrary value units, not dollars.
Compare doing nothing, which has value , with an equal-probability gamble that gains or loses USD 100. Its expected monetary payoff is zero, but its average subjective value under this model is negative:
A decision maker maximising this score chooses zero for sure. The is a value score; it does not say the gamble loses USD 50 on average.
Keeping the possible loss at USD 100, a possible gain of USD 200 makes the score break even:
This second gamble has expected money of USD 50, despite having zero average subjective value. In this particular model, a gain twice the loss compensates for the asymmetry; it is not a general rule for acceptable trades.
Original fictional illustration. The reference point stays fixed, probabilities receive their ordinary weights, and each side is linear. The kink isolates loss aversion; the model omits other features of prospect theory.
Three mechanisms worth separating
| Mechanism | What changes? | What the toy model does |
|---|---|---|
| Loss aversion | The relative value of equal gains and losses around a reference point | Makes the loss side steeper |
| Diminishing sensitivity | The marginal response as an outcome moves further into gains or losses | Omits curvature: each side is linear |
| Probability weighting | The decision weight attached to an outcome's probability | Uses the stated probabilities directly |
Under expected utility over final wealth, concavity can make a person prefer a sure outcome to a gamble with the same mean. That is a different specification from the reference-dependent asymmetry here. Likewise, loss aversion alone does not imply risk seeking among losses: in this toy model, a sure loss and an equal-mean gamble lying entirely on the linear loss side have the same score. Prospect theory combines features of value and decision weights; this illustration is not a full implementation. See the original paper's discussions of the value and weighting functions, pp. 277–281.
Evidence, trading behaviour and limits
The size and even the appearance of loss-averse choice depend on the task. Ert and Erev's experiments varied framing, actual stakes and the alternatives presented; their findings caution against treating a single multiplier as a psychological constant. Their Study 2 and discussion, pp. 220–221 and 225–227, distinguish observed choice patterns from broad claims about underlying preferences. A fitted coefficient needs the elicitation method, stakes, reference-point assumption and population attached to it.
Loss aversion is often invoked alongside the disposition effect, the tendency to realise gains more readily than losses. But that trading pattern does not establish loss aversion as its sole cause. Barberis and Xiong show that prospect-theory predictions depend on whether preferences apply to annual or realised gains and losses; some specifications fail to generate the pattern. See their 2009 analysis, pp. 751–753.
The practical use is to formulate a testable account of decisions: state the reference point, identify the outcomes being evaluated and distinguish a value asymmetry from curvature or probability weighting. A behavioural label does not forecast a security's return or determine whether an existing losing position should be sold.
Practice: change the reference, keep the money
Use the same fictional value function and ordinary probability weights. Compare certain final wealth of USD 2,000 with a 50/50 gamble ending at USD 1,800 or USD 2,200. What does the model prefer when the reference is (a) USD 1,800 and (b) USD 2,000? Only the reference changes; ignore costs and other risks.
Answer. With reference USD 1,800, the gamble gives changes of 0 and +400, so average value is 200. The certain change is +200, also worth 200: the model is indifferent. With reference USD 2,000, the gamble gives changes of −200 and +200, worth −400 and +200. Its average value is −100, below the certain outcome's value of zero. The model now prefers certainty. Identical final-wealth prospects can produce different rankings because this model evaluates deviations from a reference point.
Sources
- Kahneman and Tversky (1979), Prospect Theory: An Analysis of Decision under Risk, especially pp. 263–264, 277–281 and 286–288.
- Ert and Erev (2013), On the descriptive value of loss aversion in decisions under risk: Six clarifications, especially Study 2 and §6.
- Barberis and Xiong (2009), What Drives the Disposition Effect? An Analysis of a Long-Standing Preference-Based Explanation, abstract and introduction, pp. 751–753.