Loss psychology: the reference point, loss aversion and the disposition effect
The account shows USD 11,000: a thousand above starting capital, but USD 1,400 below its earlier peak. The same situation can therefore look like a gain or a loss, depending on the reference point. That comparison may influence whether a position is closed, held or increased, although it does not change market evidence. Article 10.5 examined process violations after an adverse outcome. We now consider how an outcome is evaluated, what research actually says about loss aversion, and how to distinguish behaviour consistent with a plan from a justification added afterwards.

A smaller red number is not necessarily a better account result
A position showing a loss disappears when it is closed. Reopening the same exposure creates a fresh entry price and a much smaller open loss. Before treating that change of display as an improvement, reconcile equity. A screen label and the resources remaining in the account are different objects.
Our hypothetical USD account has a settled balance of 5,000 USD and a long position of 30,000 EUR bought at 1.1180. The balance already includes all previously booked charges, including the original entry commission. Current EUR/USD is Bid 1.1150 and Ask 1.1152. A pip is 0.0001 USD per EUR. Assume complete immediate closing and reopening at these unchanged quotes, a commission of 0.90 USD on each new deal, and no other fees, financing, taxes or currency conversion. This is an accounting example, not a proposed way to manage a losing position.
Carry the realised loss into balance before opening the new row
| Checkpoint | Balance | Open price result | Equity |
|---|---|---|---|
| Before closing | 5,000.00 USD | −90.00 USD | 4,910.00 USD |
| After closing, before reopening | 4,909.10 USD | 0.00 USD | 4,909.10 USD |
| After reopening | 4,908.20 USD | −6.00 USD | 4,902.20 USD |
The original open price result is (1.1150−1.1180)×30,000=−90 USD. Equity before the operation is 5,000−90=4,910 USD. Closing realises that loss and charges the new exit commission: balance becomes 5,000−90−0.90=4,909.10 USD. The original entry commission is not deducted again.
Reopening buys the same 30,000 EUR at Ask 1.1152 and charges another 0.90 USD, leaving balance of 4,908.20 USD. At the unchanged Bid, the new position immediately shows (1.1150−1.1152)×30,000=−6 USD. Its displayed open loss is smaller, but the earlier loss is now in balance. Equity is 4,908.20−6=4,902.20 USD.
Identify what the change of reference actually cost
Compared with leaving the position untouched at the initial checkpoint, the operation lowers immediate equity by 4,910−4,902.20=7.80 USD. The new entry crosses a 2-pip spread, worth 6 USD at this volume, and the two new commissions total 1.80 USD. Their sum is the same 7.80 USD. Do not add the old 90 USD again to that difference; both alternatives already contain the earlier marked loss.
MetaQuotes describes buying at Ask and selling at Bid. The arithmetic here follows those stipulated deal sides. An actual operation may also involve changed quotes, a temporary flat interval, different contract conditions or tax effects. Those omitted features would require additional records. The calculation establishes only the stated immediate comparison, not that closing or holding is generally preferable.
Separate a change in the display from an explanation of behaviour
After reopening, the position’s entry reference is different, while the account still carries the previous loss and the extra costs. The same currency quantity is again exposed to future price moves. Changing the label does not supply new market evidence or an expected return. An exit rule, revised hypothesis or capital constraint still needs its own justification.
Imas’s research distinguishes realised and paper losses in experimental risk-taking. It does not show that paying to reset an entry improves a particular trader’s performance. Nor can this ledger diagnose loss aversion or identify a motive. Record what was said beforehand, the actual action and the resulting statement separately. The detailed lesson then examines reference points, research findings and counterexamples; this example first keeps the economic result visible when an old red row is replaced by a new one.
Three exercises with worked answers
The exercise data are hypothetical. Choose an answer, then check the calculation.
Continue to reference points and the research
The full lesson distinguishes loss aversion, risk aversion, the disposition effect and mental accounting. It explains the populations and tasks studied, why estimated coefficients are not universal constants, and what a pre-decision record can establish.
1. One number, five reference points
Kahneman and Tversky's prospect theory (1979) rests on an observation that sounds trivial until you see its consequences: people evaluate outcomes as changes relative to some point, not as states of wealth[1]. To nobody is an account holding 11,000 USD simply "an account holding 11,000 USD". It is a thousand better or four thousand worse – depending on what happens to be sitting in the background.
The 1979 paper is descriptive, not normative. Kahneman and Tversky showed how people decide in a laboratory, with gambles of known probability, most of them hypothetical. They did not write how one ought to decide, and they did not study traders. That distinction will matter for every number in this article.
The reference point moves, and rarely is there just one. There are at least five candidates: the capital deposited at the start, the account value at the beginning of the day or week, the highest value the account has ever shown, the entry price of an open position, and the figure the trader has set for themselves as a target. Each produces a different description of the same situation, and the description governs the decision.
Example: the same position, two reference points
Illustrative example; nominal values, with no reference to any specific account. A trader has an open position showing a profit of roughly half of what the plan written before entry assumed. The exit condition in that plan read: I close when price touches the planned level or when the level invalidating the scenario is breached.
- Reference point = entry price. The result is positive, so the question "am I about to lose what I already have" appears. Decision: close now, before the market gives it back.
- Reference point = the condition from the plan. Neither exit condition has occurred, although the financial result remains positive. Decision: do nothing and wait for a condition.
The same market data, the same setup, opposite actions. The difference does not lie in the chart analysis, but in what the result is being measured against. And one more thing: you cannot settle which decision was better by looking at what price did afterwards. That is an assessment of process, not of outcome – developed in the article on trade outcome versus process quality.
2. Loss aversion: a range, not a constant
Loss aversion means that a loss carries greater subjective weight than a comparable gain relative to a given reference point. It is a model concept and a pattern found in some decision tasks, rather than a universal law of experienced pain. The estimated effect’s scope and size depend on task design, method and population, so the result must be read alongside those conditions.
The popular coefficient of 2.25 does not come from the 1979 paper, though it is often attributed to it. It appeared in the later, cumulative version of prospect theory (Tversky and Kahneman, 1992), as a parameter fitted to the results of a small group of subjects[2]. The meta-analysis by Brown, Imai, Vieider and Camerer (2024), covering several hundred estimates across different studies, methods and populations, gives a value of 1.955 with a 95% Bayesian interval of [1.820, 2.102][3]. That is a pooled estimate, not a measurement of a particular trader’s parameter. Individual preferences can be estimated in separate tasks, but this average must not automatically be assigned to them.
There is also criticism of the phenomenon’s scope. Gal and Rucker (2018) argued that some classic findings can be explained by choice inertia, switching costs or question design[4]. Yechiam and Zeif’s re-meta-analysis (2025) found a coefficient close to one when tasks were restricted to symmetric gains and losses with unordered presentation[12]. This does not settle every context, but it rules out treating one pooled average as a law or a personal measure of every trader.
In the model shown, an additional step of the same size carries less subjective weight further from the reference point. "I am already deep in it; one more position changes nothing" may fit that interpretation, but does not establish it. Neither reduced care nor adding a position reconstructs the trader’s value function. The record can test the earlier reference point, condition and change in action; new information, costs and beliefs about the next move must also be considered. This tests a hypothesis rather than measuring a curve from one decision.
3. Loss aversion versus risk aversion
These two concepts are used interchangeably in trading texts, and that is a mistake which changes the conclusions.
Risk aversion means preferring a certain outcome to a lottery with the same expected value: in this example, the person chooses a certain 100 rather than equal chances of 200 or zero. The concept does not require classifying gains and losses relative to a reference point. It does not imply that one person will be equally cautious at every stake and in every setting.
Loss aversion describes an asymmetry around a reference point: an outcome labelled as a loss weighs more than a comparable outcome labelled as a gain. Without a reference point the concept loses meaning, because there is no telling what counts as a loss.
Prospect theory can describe a person choosing a more certain outcome in the gain domain and a riskier possibility in the loss domain. A position history may resemble this through earlier realisation of gains and longer retention of losses, but that pattern does not identify the model or mechanism. Probabilities, stakes, exit conditions and new information also matter. "Risk-averse" or "risk-loving" is too broad a label if it leaves out those conditions and the reference point.
4. The disposition effect: what Odean actually counted
The concept was introduced by Shefrin and Statman (1985): the tendency to sell winning positions too readily and hold losing ones too long[5]. The original paper contained both theory and evidence. Terrance Odean later examined the pattern in a large account sample.
Study card. Odean (1998), Journal of Finance. Data: roughly 10 thousand randomly selected accounts at a US discount brokerage, the years 1987–1993, individual investors in the equity market[6]. Method: for every day on which an investor sold anything, count how many of the available winning positions were realised, and do the same for the losing ones. From that come two proportions: the proportion of gains realised and the proportion of losses realised.
The difference in frequency alone says only that something in the choices was systematic. The stronger part of the result concerns what happened afterwards: over the following twelve months the winning positions that were sold delivered an excess return of +2.35%, and the losing ones that were kept −1.06%. The difference is 3.4 percentage points against what the investors were holding on to.
Extrapolation to the currency market – flagged as such. These were equity accounts, without leverage in today's sense, in an environment where a position can be held for years and where US rules on realising tax losses applied. Retail FX trading has a different structure: leverage, a cost of carrying a position, protective orders as standard, and a horizon counted in hours or days. The direction of the effect is credible as a hypothesis to test against your own records. The magnitude does not carry over.
It is also worth knowing that the picture of "investors reluctant to realise losses" was later complicated. Ben-David and Hirshleifer (2012) analysed the propensity to sell as a function of a position's return so far and found a V shape: it rises both at large gains and at large losses[7]. That does not fit a simple reluctance to close losing positions, and fits rather a change in beliefs about the next move. The disposition effect is therefore the name of a pattern in data, not proof that loss aversion was at work in every single case.
Counterexample: holding a loss that is not a mistake
Illustrative example. Before entry the trader wrote: the scenario is invalidated only once a level is breached that sits clearly further away than a typical swing on this instrument; until then the position stays open regardless of the current result. The position is now under water, but that level has not been breached. The trader does nothing.
From the outside, there is an open losing position and no response. The pre-decision record distinguishes compliance. If the condition existed beforehand and remains unbreached, inaction executes the plan. If the record is missing, the decision’s basis is unknown; if a condition was breached, a departure from the plan is identified. None of these facts alone establishes an emotional cause or the disposition effect in a particular person.
The criterion works in the other direction too, and this is the part that is easy to miss: closing a losing position before the condition from the plan is also a departure, even though it looks like discipline. A trader who cuts a loss "to get some peace", ahead of the invalidation level, changed the rules exactly as much as one who failed to honour them the other way.
5. Paper losses and closed losses: mental accounting
Thaler (1985) described mental accounting: people assign outcomes to separate accounts in their heads and settle them separately, instead of treating wealth as one pool[8]. A mental account closes at the moment of realisation – and that is where the belief comes from that as long as a position is open, "nothing has happened yet".
Account equity includes open positions at their current marked-to-market value. An unrealised loss does not disappear because the position remains open. Closing it realises the result and removes further exposure; it may also involve execution costs and tax consequences. It therefore does more than change mental bookkeeping. The psychological distinction concerns excluding an existing result from the account assessment, rather than economic identity between an open and a closed position.
What this looks like in behaviour rather than in declarations:
- a weekly summary calculated from balance rather than equity, with open positions under water;
- the sentence "this week was flat" spoken with a losing position still open;
- a separate "experimental account" whose results do not enter any summary;
- treating transaction costs as something outside the result, because "that is not a loss on the market".
What this should not be confused with. Valuing a position by the condition from the plan rather than by the current deviation is not mental accounting – it is simply assessment against a previously adopted criterion. Likewise, splitting capital across two accounts at a broker can be a deliberate limitation on freedom of decision rather than self-deception; the difference lies in whether the split was designed beforehand and whether both accounts enter a common summary. Designing such constraints is the subject of the course article on checklists over willpower (10.7).
6. Breaking even and "gambling with the house money"
Thaler and Johnson (1990) studied how a prior outcome changes willingness to accept the next bet[9]. The study was laboratory-based, the participants were students, some bets were hypothetical and some were played for real, small stakes. Two effects came out of it which trading texts tend to confuse.
The house money effect: after a prior gain, willingness to accept a risky bet rises, because the result is booked as "playing with winnings" rather than with one's own wealth. The break-even effect: after a prior loss, acceptance rises for a bet that offers a chance of returning to the starting point – even if under normal conditions it would be rejected.
Carrying this across to the currency market is an extrapolation and that has to be said plainly: students in a laboratory made single decisions about small amounts, with no weeks of history behind them, no costs and no chance to learn from their own records. What carries over is the direction of the mechanism and its link to the reference point. What does not carry over is any number whatsoever.
Breaking even can be a particularly salient reference point: it allows a mental account to close without a loss. The study does not establish that this level always has more influence than every other one. The market does not know a personal entry price or recovery target. When returning to that value begins to change successive orders outside the plan, the sequence becomes the subject of the article on revenge trading and tilt. Wanting to break even does not alone establish a violation.
7. Sunk costs and escalation of commitment in cold blood
Arkes and Blumer (1985) showed in a series of experiments that costs already incurred and irrecoverable influence decisions about continuing, although from the point of view of future outcomes they should not[10]. Their most convincing piece was a field experiment: buyers of theatre season tickets were randomly granted discounts, and those who had paid the full price subsequently attended more performances. The ticket price did not change the value of any single show – it changed the sense of how much had already been invested.
Staw (1976) described a related phenomenon, and a more important one for trading: escalation of commitment. In his study, participants making a simulated financial decision allocated more funds to a course of action they were themselves responsible for, and did so particularly after feedback about failure[11]. The participants were business students and the decision was a simulation – but the mechanism concerns responsibility for an earlier choice, not arousal.
The distinction matters. Adding to a losing position does not require a visible outburst of frustration. It may accompany reported calm and an elaborate rationale, but that establishes neither an absence of emotion nor greater difficulty recognising it. The checkable fact is further commitment without a previously permitted condition or recorded new information. "I was upset" does not settle compliance; neither does "I was calm".
What this should not be confused with. Information that came into being after the position was opened can genuinely matter for the future outcome – that is not a sunk cost. Scaling a position written into the plan before entry is not escalation either, even if the next tranche goes in at a worse place. A sunk cost concerns only what has already been spent and cannot be recovered: money, learning time, months spent on one approach.

8. Behaviour, mechanism and the simpler explanation
Naming a bias after the fact proves nothing. Before pinning a label from this article onto a behaviour, you have to rule out explanations that require no psychology at all: market conditions, the instrument, the time of day, volatility, the way the platform behaves. The table below organises that distinction – the fourth column shows what settles the dispute between the second column and the third.
| Observable behaviour | Mechanism from this article | Simpler explanation to be ruled out | What settles it |
|---|---|---|---|
| Closing a winning position before the condition from the plan | Reference point = entry price; disposition effect | A previously written condition for partial realisation; a news item changing the scenario's assumptions | The record of the exit condition made before entry |
| No reaction to a position under water | Reluctance to book the loss | The invalidation level sat further away than you remember; the order was rejected or never arrived | The recorded invalidation level and the order log |
| A further commitment in the same direction after a loss | Escalation of commitment, sunk costs | Scaling written into the plan before the first entry | A scaling plan dated earlier than the first order |
| Rejecting a signal that met the conditions right after a loss | Asymmetry around the reference point | The signal failed one of the conditions; the volatility filter worked correctly | A checklist completed before the decision, not after |
| A weekly summary without open positions under water | Mental accounting | The report shows balance by default, not equity | The report setting and the field definition in the journal |
| Increasing commitment as the account approaches its starting value | The break-even effect | Market conditions genuinely improved against the criteria in the plan | Entry criteria written down before the session |
No row in this table is settled by the outcome of a trade. A position closed earlier than the plan foresaw may have ended better than the plan – and it was still a departure. That distinction matters enough to have its own instalment in the course: outcome versus the decision process.
9. How to record the reference point so it can be checked
Everything above comes down to one practical problem: the mechanism operates before the decision and is recognised afterwards. The only way out is to record the part of the situation that is checkable – namely, what the result is being judged against.
Below is the full form of the procedure. It is not a journal template nor a set of risk rules; each of those has its own place in the course (the trading journal and the risk management plan).
| Field | Content |
|---|---|
| Trigger condition | Every new order and every decision to change an existing position. |
| Action | One sentence before the order is placed: what you are judging this position against (the invalidation level and the exit condition) and which account value you are treating as the reference point this week. |
| Completion criterion | The entry closes at the moment the order is placed. It is not edited once the outcome is known – after-the-fact commentary goes into a separate field. |
| Audit trail | The timestamp of the entry earlier than the timestamp of the order in the broker's history. Without that, the entry is indistinguishable from a rationalisation. |
| Failure criterion | After thirty closed positions, count in how many cases the entry lets you settle whether the decision matched the condition. If that share is not rising, the field is badly designed – usually too general. That is a flaw in the tool, not in your character. |
Two notes on what this procedure does not give you. It does not remove the mechanism: recording the reference point does not stop it operating, because awareness of a bias is no protection against it. What it does give you is data on which, after a few dozen decisions, you can check whether departures form a regularity or are noise. And it does not replace an assessment of the strategy itself – if the method has no foundation, the best-recorded reference point will fix nothing.
The nearest test is in fact sooner than it seems: at the first position that goes against you in a week in which the account is already below its peak. That is when it turns out whether the morning entry contains a condition or only a hope. The article on checklists over willpower (10.7) deals with how to build such conditions so that they survive the moment when their author is no longer the same calm person who wrote them.
FAQ – the reference point, loss aversion, the disposition effect
Does a loss really hurt twice as much as an equal gain pleases?
I am holding a losing position. Is that the disposition effect?
Is an unclosed loss already a loss?
Are loss aversion and caution the same thing?
After what size of loss should I stop trading?
Is topping up the account after a big loss always a mistake?
Sources and references
- Kahneman D., Tversky A., "Prospect Theory: An Analysis of Decision under Risk," Econometrica, 47(2), 1979, pp. 263–291. The reference point, diminishing sensitivity, the asymmetry of gains and losses. A descriptive paper based on mostly hypothetical gambles; it contains no loss aversion coefficient.
- Tversky A., Kahneman D., "Advances in Prospect Theory: Cumulative Representation of Uncertainty," Journal of Risk and Uncertainty, 5(4), 1992, pp. 297–323. Source of the 2.25 parameter, fitted to the results of a small group of subjects.
- Brown A.L., Imai T., Vieider F.M., Camerer C.F., "Meta-Analysis of Empirical Estimates of Loss Aversion," Journal of Economic Literature, 62(2), 2024. Pooled estimate of the coefficient: 1.955, 95% Bayesian interval [1.820, 2.102].
- Gal D., Rucker D.D., "The Loss of Loss Aversion: Will It Loom Larger Than Its Gain?," Journal of Consumer Psychology, 28(3), 2018, pp. 497–516. A critical review of the evidence for the generality of loss aversion.
- Shefrin H., Statman M., "The Disposition to Sell Winners Too Early and Ride Losers Too Long: Theory and Evidence," The Journal of Finance, 40(3), 1985, pp. 777–790. Introduction of the disposition effect.
- Odean T., "Are Investors Reluctant to Realize Their Losses?," The Journal of Finance, 53(5), 1998, pp. 1775–1798. Roughly 10 thousand accounts at a US discount brokerage, 1987–1993; proportions of gains and losses realised 0.148 against 0.098; excess returns over the following 12 months +2.35% against −1.06%.
- Ben-David I., Hirshleifer D., "Are Investors Really Reluctant to Realize Their Losses? Trading Responses to Past Returns and the Disposition Effect," The Review of Financial Studies, 25(8), 2012, pp. 2485–2532. A V-shaped propensity to sell as a function of a position's return so far.
- Thaler R., "Mental Accounting and Consumer Choice," Marketing Science, 4(3), 1985, pp. 199–214. Mental accounting and the moment a mental account is closed.
- Thaler R.H., Johnson E.J., "Gambling with the House Money and Trying to Break Even: The Effects of Prior Outcomes on Risky Choice," Management Science, 36(6), 1990, pp. 643–660. A laboratory study on students, stakes partly hypothetical; the house money effect and the break-even effect.
- Arkes H.R., Blumer C., "The Psychology of Sunk Cost," Organizational Behavior and Human Decision Processes, 35(1), 1985, pp. 124–140. A series of experiments together with a field experiment on theatre season-ticket buyers.
- Staw B.M., "Knee-Deep in the Big Muddy: A Study of Escalating Commitment to a Chosen Course of Action," Organizational Behavior and Human Performance, 16(1), 1976, pp. 27–44. A simulated financial decision, business students; escalation of commitment after feedback about failure.
- Yechiam E., Zeif D., “Loss Aversion Is Not Robust: A Re-Meta-Analysis”, Journal of Economic Psychology, 107, 2025, 102801. Re-analysis; the estimated effect depends on amount symmetry and presentation order. Authors’ full text.