The Sunk Cost Fallacy: Throwing Good Money After Bad

A sunk cost is money already spent that no later action can recover. The fallacy is letting it change a decision it has no bearing on. Only what happens from here can differ between the choices on offer, and what was spent appears in none of them. The pull is strongest when the spending was visible, effortful, or the spender's own idea.

The fallacy rests on a distinction so plain that saying it out loud feels unnecessary. No decision taken today reaches backwards and changes what was already spent, so what was already spent cannot be a reason for deciding one way rather than another. Almost everybody agrees with that sentence stated in the abstract, and almost nobody acts on it when the money was theirs. The gap between agreeing and acting is what makes this worth a full treatment rather than a footnote. The logic was never what was missing, so learning it confers no protection from the pull. Worth having instead is a test that can be run on a decision from the outside, and a clear sight of the three things that make the pull stronger in one situation than in another.

What makes a cost sunk rather than merely large?

A cost is sunk when no decision now available can bring it back. The whole definition is that one sentence, and a great deal is missing from it. Nothing about size. Nothing about whether the spending was wise, recent, painful or public. A small cost can be entirely sunk and an enormous one can be almost fully recoverableAble to be got back if the course is abandoned, and therefore not sunk at all.. The only question is whether abandoning the course returns the money.

The clearest version sits outside finance. A wedding hall is booked with a deposit of Rs 40,000/-, of which Rs 25,000/- is returned if the booking is cancelled more than a month ahead. At the moment of payment, Rs 15,000/- became a sunk costMoney already spent that no future decision can recover. and Rs 25,000/- did not. If the date now clashes with something that matters more, the Rs 15,000/- has no vote. The Rs 25,000/- does. Cancelling actually produces it. The line between sunk and recoverable is not drawn by the size of the payment but by whether stopping gets it back.

Now the holding. Meera Sundaram, an invented investor whose decisions run through this guide, put Rs 3,00,000/- into Kesari Logistics Limited on 4 January. By 30 September the position is worth Rs 1,95,000/-, a fall of 35.0 per cent and a loss on paper of Rs 1,05,000/-. Of the original Rs 3,00,000/-, the part she can still get her hands on by acting today is Rs 1,95,000/-. The other Rs 1,05,000/- is gone in the only sense that matters here: no choice on her list produces it.

Why can money already spent not change what happens next?

The argument in full is short enough to hold in the head. A decision is a choice between options. Choosing means comparing, and comparing means looking for something that differs between them. Whichever option is taken, a quantity that is the same in every option is what it is, so it cannot separate them. The Rs 3,00,000/- Meera paid is spent if she holds and spent if she sells. The paid figure is an identical branchA quantity that appears the same whichever option is taken. quantity, and it is therefore silent on the choice.

Strike out everything that is identical across the options and whatever survives is the actual decision. The striking out is the forward-looking testAsking only what happens from here, ignoring everything already spent., and it is not a matter of moral discipline. It is arithmetic. If a number is subtracted from both sides of a comparison it changes neither side, and there is no version of the reasoning in which it comes back to matter later.

The street version makes it obvious. Somebody has paid Rs 900/- for a cinema ticket and, forty minutes in, is bored and slightly unwell. Staying costs the rest of the evening; leaving costs the rest of the evening too, minus the film. The Rs 900/- is spent in both cases, so the only thing that differs is what happens to the next ninety minutes. People still stay, and they explain it by saying they paid for it. The explanation is a description of the past presented as a reason about the future.

What is left once the identical part is struck out?

Something changes when the strike out is run. The question being asked quietly turns into a different question, and the two have different answers. With the Rs 3,00,000/- still in view, the question is how to get back to it. With the Rs 3,00,000/- struck out, the question is where Rs 1,95,000/- is best held from today. A recovery question and a placement question are not the same question wearing different clothes, and only one of them can actually be answered.

Look at what the recovery framing quietly demands. Getting from Rs 1,95,000/- back to Rs 3,00,000/- takes a rise of Rs 1,05,000/-. On what is in hand that is 53.8 per cent. Nobody chose that requirement and nothing about the holding produced it. The requirement arrived from the purchase date. The placement question asks something the holding can actually be examined for; the recovery question asks the holding to perform a task set by a receipt.

Run as a procedure it takes four steps and no judgement at any of them. List the options. List every quantity that appears in the reasoning. Delete each quantity that reads the same on every option. Decide on whatever is left. The deletion step is mechanical, and being mechanical is exactly what makes it useful. It demands no bravery about a loss, only a comparison of two entries on a list.

What does the test do to the Rs 3,00,000/- in Meera's sentence?

On 12 October Meera takes two decisions in one sitting. She sells Suvarna Chemicals Limited whole at Rs 4,60,000/- against a cost of Rs 4,00,000/-, booking Rs 60,000/- and 15.0 per cent. She keeps Kesari Logistics, then down Rs 1,05,000/-, and records a reason in the decision log kept by Palash Advisory Services Private Limited: she will sell it when it gets back to Rs 3,00,000/-. Devika Rao, the adviser at the practice, reads that line eight weeks later and stops on the last five words.

Apply the test to the sentence. From today, is Rs 1,95,000/- better left in Kesari Logistics or better placed somewhere else? The placement question is answerable. Answering it needs a view on what Kesari does from here and what the alternatives do from here, and it needs nothing whatsoever about 4 January. Now look at what the Rs 3,00,000/- is doing in her sentence. The figure is spent if she holds. It is spent if she sells. It is spent if she does something in between. A quantity that is the same on every branch cannot distinguish between them, so the Rs 3,00,000/- has no legitimate work to do and is doing all of it anyway.

The conclusion needs stating carefully, and so does its limit. The test does not show that keeping Kesari Logistics was wrong. Holding might well be the better placement for that Rs 1,95,000/-. The conclusion is narrower and harder than that: the Rs 3,00,000/- cannot be part of the reasoning either way, so a decision resting on it has not been taken on the merits, whichever way it happened to land. One case is never evidence that a rule works.

Why does knowing all of this not protect against it?

Because the pull does not arrive as an argument about sunk costs. If it did, it could be refuted and the matter would end there. The pull arrives as a feeling that stopping now would waste what has already gone in, and that feeling is not stupid. Waste aversionThe dislike of seeing past spending produce nothing, which is what drives the pull. is a decent instinct in most of life: a person who throws away food, materials and half finished work at the first difficulty is not somebody to put in charge of anything. The waste has already happened and no future action un-wastes it, so a sound instinct is being applied to the one situation where it cannot help.

The effect was set out by Arkes and Blumer in The Psychology of Sunk Cost, published in Organizational Behavior and Human Decision Processes in 1985, and the finding that matters is not that people continue. The finding is that they continue more when more has been spent, with the forward-looking question held identical. The same future, described in the same words, gets a different answer purely because of a number that no decision can change.

Which kinds of spending pull hardest, and why is it not the biggest?

If the amount were what mattered, the strongest pull would always come from the largest number. It does not. Three features of how the spending happened change the pull at an unchanged rupee value, and knowing them shows where care is needed long before the decision is anywhere near. The pull tracks how the spending was experienced, not how large it was.

Why does visible spending pull harder than spending that is never seen?

Meera pays Rs 25,000/- every month by standing instruction. Over a year that is Rs 3,00,000/-, and she has never once watched it go. On the evening of 19 February a television segment names Suvarna Chemicals and she adds Rs 1,00,000/- to that holding within the hour. One third of the money, and it is the one she can describe in detail. Ask her six months later which spending she would find harder to walk away from and it is not the larger one.

Spending is felt as spent when it was watched, not when a statement records it. Visibility is why the pull is often strongest on the smaller decision and why an automatic arrangement is easier to stop than a deliberate one, even when the automatic arrangement has consumed far more.

Why does effort pull harder than money at the same value?

Two people have each put Rs 50,000/- into something that is now going badly. The first wrote a cheque. The second did the work: six weeks of evenings, drafts, phone calls, a thing built by hand. The rupees are equal by construction. The second person will find it much harder to stop, and the reason is that money is fungible and effort is not. Rs 50,000/- of somebody else's work is a number on an invoice. Six weeks of one's own is six weeks of a life, and abandoning it says something about that time that a cheque never says about itself.

Effort spent is remembered as effort, and effort is not a quantity that can be written off in the way a payment can. A street vendor who bought a cart yesterday can sell it to the next vendor. The same vendor who spent two years building the round of regular customers has nothing to sell and nothing to write off, and that is precisely why the second is harder to leave.

Why does a person's own idea pull harder than somebody else's?

The third feature is about authorship. Abandoning a course that somebody else proposed means accepting that their suggestion did not work. Abandoning a course a person argued for themselves means accepting that their own judgement did not work, and those two admissions do not cost the same. Authorship is why the pull is worst on the holding researched over weeks and mildest on the one a relative talked them into, even when the amounts are identical.

The stronger the sense that the choice was the person's own, the more the decision to stop looks like a verdict on them rather than on the holding. There is a useful inversion here. The decisions a person feels least proud of are the easiest to reverse, so the ones they feel proudest of deserve the test first.

What single question detects a sunk cost at work?

The test asks what somebody who had not spent the money would do. Not what a wiser person would do, not what the textbook says: what a specific person, facing exactly this holding, with exactly this future in front of them, and none of the spender's history, would choose today. A person described that way is the outside viewHow the same decision looks to somebody who did not spend the money.. The past the spender is carrying is the one thing that person does not have, so their answer is the forward-looking one by construction.

The question works because it demands no discipline. It swaps the person. If a stranger handed over Rs 1,95,000/- in cash and asked whether to place it in Kesari Logistics today, the answer would turn on what Kesari does from here. The stranger's decision is the same one Meera faces, described without her purchase. Nothing else about the situation moved, so if the answer changes when the person changes, the difference is the sunk cost.

Does past spending ever legitimately matter?

Yes, and the distinction is worth getting right. A reader who concludes that the past is always irrelevant will start making a different mistake. Past spending matters whenever it left something behind that still exists. A course paid for left knowledge that is now held. A machine bought left a machine. Two years of work left a set of customers who still come back. In each case the thing doing the work in today's decision is not the spending: it is the knowledge, the machine, the customers, all of which are present facts that happen to have been produced by past spending.

The test is not when the money went out but whether anything survives that a decision today can use. Where nothing survives, the spending is sunk and silent. Where something survives, name the surviving thing and put that in the reasoning instead, and the sentence stops referring to the past at all. Meera's Rs 3,00,000/- left a position now worth Rs 1,95,000/-. The position is a present fact and belongs in the decision. The Rs 3,00,000/- does not.

How does anybody actually use this on a live decision?

Nobody has ever answered yes when asked whether they are committing the sunk cost fallacy, so Devika Rao does not put that question to her clients. She asks for the reason in writing, and then she reads the reason for one thing only: whether it can be stated without naming what was paid. The Palash decision log records a written reason on 84 of its 240 decisions, being 35.0 per cent. On the other 65.0 per cent there is nothing to read at all.

The written line is what makes it visible. Meera's reason, sell when it gets back to Rs 3,00,000/-, is transparent once it is on paper and almost invisible while it is a thought. On 4 November, 20 of the 60 investors in the log adopted a written checklist. Across the next four quarters they recorded a reason on 34 of 41 decisions, being 82.9 per cent, against 19 of 63, being 30.2 per cent, for the other 40, and their ratio of realised gains to realised losses fell from 3.2 to 1.6. Eight quarters and 60 people cannot support a finding about returns. The only change on record is a change in what was written down, not in what was earned.

For a person deciding alone, with no adviser and nobody to answer to, the same procedure works with one change: the outside view has to be supplied by the decider, and the cheapest way is the written sentence. The reason is written down before acting, read back the next morning, and every clause that names a number from the past is struck out. If nothing survives the striking, the decision has not been made yet. A professional deciding on behalf of somebody else has a further duty of care in how a recommendation is reasoned and recorded, and the conduct and record keeping requirements that apply are set by the Securities and Exchange Board of India.

What does the sunk cost fallacy not explain?

The fallacy does not explain why people value something more once it is theirs. Possession alone does that, with no spending at all, and possession alone is the endowment effect. The fallacy does not explain why an organisation keeps funding a project everybody privately knows is finished either. Inside an organisation that is escalation of commitmentContinuing to put resources into a failing course because of what is already in it., driven by reputation, careers and other people watching rather than by one person's arithmetic, and those drivers belong with group decision making. The account here is deliberately narrow: one person, one past outlay, one decision about what happens next. Where it is being used to explain a group's behaviour, something else is doing most of the work.

Meera Sundaram, Devika Rao, Palash Advisory Services Private Limited, the Palash decision log, Suvarna Chemicals Limited and Kesari Logistics Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.