Sunk Costs and the Escalation of Commitment

Why a cost that cannot be recovered keeps deciding

A sunk cost is money already spent, and standard decision theory says it should not affect a choice, because it is identical under every option ahead. People violate this reliably. The demonstrations are decades old and still uncomfortable: people who had paid more for a ticket were more likely to travel to a game through a blizzard; groups deciding whether to fund a project were swayed by how much had already been spent on it, even when the remaining money would have been better used elsewhere. The effect is not confusion about arithmetic — the participants can usually state the rule correctly — it is that closing the position finalises the loss, and the loss is what the mind is avoiding. Escalation of commitment is the version that compounds. When a course of action is failing, the person responsible for it tends to commit more rather than less, and the effect is stronger when the original decision was theirs, when it was public, and when they expect to be judged on the outcome. In an account, that is a position held by the trader who chose it, discussed with people who know they chose it, and measured by a result they expect to be asked about. The psychological conditions for escalation are the ordinary conditions of trading, which is why averaging down feels like a considered decision rather than an escalation while it is happening. The trading form of the fallacy has a name of its own, and it is worth seeing as a mechanism rather than as a mistake of nerve. Averaging down spends new capital at a lower price, which lowers the average cost and therefore lowers the break-even, and it does so on a thesis that the market has just contradicted. The number improves and the position worsens: more capital is exposed, the exit is further away in dollars, and the decision is now large enough that closing it is a bigger admission than it was. That is why the two questions that end the loop are asked before the position, not after it — what is the maximum size, and at what price is the thesis dead? The add, priced — Original: 500 shares at $40.00: Cost $20,000 · Price now $28.00: Loss on the original shares: $6,000, or 30% ← · Add: 300 shares at $28.00: $8,400 of new capital · New position: 800 shares, $28,400: Average $35.50 — break-even now 26.8% away instead of 42.9% ← · Exposure at $28.00: $22,400, up 60% from $14,000 The loss did not change by a dollar. What changed is how much capital is now committed to the thesis, and the number that improved is the average — a description of what was paid rather than a statement about what the position is worth.

Adding, averaging down, and the one test that separates them

Adding to a losing position is not automatically wrong, and the distinction matters because “never average down” is remembered as a prohibition rather than understood as a test. Adding is legitimate when new information improves the case — a contract won, a competitor’s product failing, a price that has fallen to a level where the valuation is genuinely different — and when the total position size was decided in advance. Averaging down is the same act with the past as its reason: the case is unchanged, the price is lower, and the number that improved is the average of what was paid. The test that separates them is the one from the first lesson, applied to the addition. If the trader had no position and no history, would they open this position now, at this price, and at this size? A yes means the add is a new decision that happens to be in the same name, and it should be sized as one. A no means the add exists to change the average, which is a decision about a number rather than about a company. The practical version of the discipline is to decide the total before the first share: the full size the thesis deserves, split into the number of adds permitted, and the price at which the thesis is wrong. That converts the moment of the drawdown from a judgment call into a pre-written rule and removes the two conditions that make escalation strongest — the decision is no longer made under the loss, and it is no longer a fresh choice whose result the trader will have to explain. • Decide the total position size and the number of permitted adds before the first purchase. • Write the price at which the thesis is wrong, and treat it as a rule rather than a level to reconsider. • After any add, the decision is the whole position: would I hold this size today? • Adding needs new information; averaging down needs only a lower price. • A bigger position makes the exit a bigger admission, so the size itself is part of the escalation. The buy-it-today test has a failure mode of its own: a trader who is committed can always produce a reason to answer yes. That is why the answer has to be written down next to a date and a falsifier, so a later review can see whether the reason existed before the price fell or was assembled after it.

The same mistake, run by committees

Escalation of commitment is not a retail problem. The research that named it studied how organisations continue funding projects that any neutral observer would cancel — the case that gave the phenomenon its other name was a supersonic aircraft program justified partly by the money already spent on it. The organisational version is worth understanding because it shows the mechanism stripped of any excuse about sentiment: the people deciding are professionals, the analysis is written down, and the escalation still happens. What drives it is **accountability for a visible past decision**. Cancelling writes off a loss and attributes it to whoever approved the project; continuing keeps the loss unrealised and the decision open. For an individual that is a matter of pride, and for a committee it is a matter of careers. That has two implications for how to run your own money. The first is about how a loss gets framed: an unrealised loss in a position you chose is not “not yet a decision”, it is a decision you are making every day to keep holding it. The second is about **who should decide**. A commitment is easiest to break when the decision is made against a written rule rather than against a memory, because a rule does not have a reputation. That is why the total-position cap belongs in the plan before the first purchase: the choice of how much of the account a single idea may become is a different decision from the choice to average down, and the first one is made in the state where judgement is still intact. The organisational record also supplies a cleaner test than the ones a person applies to themselves in the moment. Projects that were cancelled on neutral criteria and projects that were kept have been compared, and the failings of the continuers are consistent: they change the success criteria as they go, they add a new justification when the original one is falsified, and they judge the project in isolation from the alternatives. All three are visible from outside and invisible from inside, which is why the written rule — a maximum size, a maximum number of adds, and an invalidation price that ends the position — is the version of the analysis you can still see when the money is committed. • Escalation is driven by accountability for a visible past decision, not by sentiment. • An unrealised loss is a decision made daily to keep holding, not a decision deferred. • Written rules break commitments; memories defend them. • The three tells of escalation: moving success criteria, adding justifications, judging the project alone.

The tests that break a commitment, and the identity trap

The rule in the previous read — decide the total before the first share — removes the decision from the moment of the loss, and it still meets a mind that has had weeks to build a relationship with the position. Three tests break that relationship, and all three work by changing the reference point rather than by adding information. The first is the **new-money test**: if you were handed the cash value of the position today, with the same account and the same plan, would you buy this position at this price and this size? If the honest answer is no, the position exists because it was already held, which is the sunk cost in its purest form. The test works because it deletes the entry price from the question, and the entry price is the number doing the psychological work. The second is the **reset test**: rewrite the plan as though the position had been opened this morning — same instrument, current price, current stop — and see what the plan says. If the new plan says hold with the same stop, the position is fine and the discomfort is about the path; if the new plan would place the stop somewhere other than where it is, the old stop is being defended for the wrong reason. The third is the **sunk-cost question asked out loud**: what would I do if I had never owned it? The phrasing matters, because it removes the loss from the frame, and what remains is a decision that can be judged on its own terms. The trap these tests are aimed at is not the arithmetic of averaging down, which the previous read already handles. It is the **identity** version, where the position stops being a position and becomes a statement about the holder: I am a long-term investor, so I do not sell; I bought it for the story, so I am not a trader; the thesis is contrarian, so being early is the point. Every one of those sentences is a commitment to a self-image rather than to a market view, and each is unfalsifiable in the way that matters — it cannot be checked against a price or a report, only against the consistency of a person’s self-description. That is why identity commitments are the ones that survive the arithmetic: they are not defended with evidence, so no evidence removes them. The detection test is linguistic and it is reliable: if the reason for holding is a sentence whose subject is “I” rather than the business or the chart, the reason is about the holder. The remedy is the same as the one the policy lessons use, applied to language rather than to size. Write the plan in the **third person** — “the position is held because the segment margin was expected to recover by the second quarter, and it is wrong if the margin falls below 14%” — and the statement becomes checkable, because it names a fact and a level rather than a disposition. And give the identity somewhere else to live: the plan can contain a clause about the kind of trader you are, and that clause should sit in the section on process, where it governs how setups are chosen, rather than attached to a single position where it governs whether a failing trade is exited. The distinction is the whole of this lesson in one sentence: a person’s trading identity belongs in the rules that generate positions, never in the reasons for defending one. Three tests and the sentences they catch — The new-money test: Would I buy this, at this price and size, with cash today? · The reset test: Where would the stop be if the position had opened this morning? ← · Asked out loud: What would I do if I had never owned it? · The identity trap: “I am a long-term investor” — a sentence about the holder, not the position ← The connection to the journaling lessons is direct: every one of these tests is answered by a written reason rather than by introspection, which is why the plan is kept in the third person and why the invalidation is stated as a level rather than as a change of heart. A commitment breaks most easily against a page that does not have a reputation to defend.

Sources

• The psychology of sunk costArkes & Blumer (1985), Organizational Behavior and Human Decision Processes
• Escalation of commitment to a failing course of actionStaw (1976), “Knee-Deep in the Big Muddy”
• Knowing when to pull the plugStaw & Ross (1987), Harvard Business Review