Money already spent, and the decisions it keeps making

The rule, and why it is not controversial

A sunk cost is money already spent that no choice available to you now can recover. Because every option leaves it equally gone, it is the same number in every column of the comparison — which is precisely what makes it useless for choosing between them.

Take a concrete version. Six years ago a traditional insurance policy was started at ₹60,000 a year, so ₹3.6 lakh has gone in, and the seventh premium is now due. The decision in front of you is whether the next ₹60,000, and the ones after it, do more inside this policy than outside it. The ₹3.6 lakh does not appear in that comparison at all, because it is gone whichever answer you give.

That is the whole rule. Only future costs and future benefits can differ between the options, so only future costs and future benefits can decide between them. A cost that no available choice can change is sunk, and its defining property is being identical across every option.

Nobody argues with this stated that way. It is a claim about what a comparison is rather than a finding about how people decide — arithmetic, not psychology. And the gap between accepting it in the abstract and applying it to a policy with ₹3.6 lakh of your own money in it is the actual subject of this article.

One clarification before going further, because it prevents the commonest over-correction. Sunk does not mean large, and it does not mean wasted. It means unrecoverable by any decision you can still make. The ₹3.6 lakh bought six years of life cover, which was delivered. What it cannot do is buy back the choice about the seventh premium.

What the research shows, and where it stops

The effect is well documented, its boundaries are not, and the popular retellings are unusually loose — so it is worth separating what has been demonstrated from what circulates.

Richard Thaler named the anomaly in economics in Toward a Positive Theory of Consumer Choice (Journal of Economic Behavior and Organization, 1980), where it sits alongside several other cases in which choices depend on things a purely forward-looking account says are irrelevant, and connects to the reference-point treatment of gains and losses.

The best-known demonstration is a field experiment rather than a questionnaire. Hal Arkes and Catherine Blumer reported it in The Psychology of Sunk Cost (Organizational Behavior and Human Decision Processes, 1985): buyers of a theatre season ticket were randomly assigned to pay the full price or a discount, and those who had paid more went to more of the plays. The plays were the same plays. The random assignment is what makes the study worth citing — the groups did not differ in how much they liked theatre, only in a randomly assigned sunk amount. The paper also reports the limit, which the retellings drop: the difference showed up in the first half of the season and had faded by the second.

Barry Staw had already produced the organisational version in Knee-Deep in the Big Muddy (Organizational Behavior and Human Performance, 1976). It is a laboratory study rather than a field one — business students allocated research funding inside a simulated company, were shown how that allocation had gone, and then allocated again — and more money went to the course of action that was going badly. That literature calls it escalation of commitment, and its most useful finding is in the next section.

Now the part usually left out. Arkes and Peter Ayton reviewed the evidence in The Sunk Cost and Concorde Effects: Are Humans Less Rational Than Lower Animals? (Psychological Bulletin, 1999) and reached an awkward conclusion: lower animals and young children generally do not honour sunk costs, while adult humans do. Their proposed explanation inverts the usual story. The effect looks less like a primitive impulse and more like an over-applied rule about waste — a perfectly sound principle, learned in adulthood, extended to a case where it does not hold.

That boundary is itself contested, and saying so is more honest than tidying it away. Brian Sweis and colleagues ran a foraging task in parallel across mice, rats and people and reported in Science in 2018 that all three were sensitive to time already spent waiting — but only once they had committed to the wait, rather than while still deciding whether to. A clean split between species does not survive that, and nor does any account treating the effect as purely a matter of adult reasoning.

Two things follow for how much weight to put on this. The direction shows up across laboratory studies, field experiments and, on the more recent evidence, other species. The size is another matter: it moves with how a scenario is described, who is deciding and what is at stake, and no single number describes it. Treat the direction as well supported and any headline magnitude as contested.

Finally, a naming footnote worth having, because the phrase is nearly always misattributed. The Concorde fallacy comes from a note by Richard Dawkins and Tamsin Carlisle in Nature in 1976 — and it was coined to criticise an argument in the animal-behaviour literature, not to report a finding about investors. The aircraft was the illustration, and it has been travelling ever since as though it were the evidence.

The cost the arithmetic leaves out

If the rule is uncontroversial, the interesting question is not why it is right. It is why knowing it changes so little — and the honest answer is that abandoning something carries a real cost that no rupee figure contains.

Stopping the policy is not only a financial act. It is the moment the earlier decision gets classified as a mistake, in front of whoever advised on it, whoever in the family agreed to it, and yourself. Continuing keeps the question open. That is a second and separate cost, and it is paid in self-assessment rather than in money, which is exactly why it is invisible to a spreadsheet that only knows about rupees.

Leon Festinger's A Theory of Cognitive Dissonance (1957) describes the general form: holding an action alongside the belief that the action was wrong is itself uncomfortable, and the discomfort can be reduced either by revising the belief or by adding reasons for the action. Adding reasons is cheaper in the moment. It is also what a persuasive case for continuing feels like from the inside.

Staw's escalation work locates the effect precisely. Commitment was stronger where the person deciding whether to continue was the same person who had made the original choice. That detail is the one to carry away, because it is the tell: the difficulty attaches to authorship of the earlier decision, not to the numbers, which are the same numbers a stranger would see.

There is a second mechanism stacked on the first, and it belongs to a neighbouring article. Abandoning a position closes an account that was still open, and closing it converts a decline that could still have recovered into a settled fact — which is the machinery described in loss aversion. The two compound rather than duplicate. Loss aversion makes the closing painful; the sunk total supplies the argument for not closing. One provides the reluctance and the other provides the justification.

None of which makes the reluctance foolish. The second cost is genuinely there, and a decision that leaves it out is not thereby the more rational one — it has simply hidden one of its own inputs. The specific error is narrower and worth stating exactly: billing it to the portfolio. Protecting a past judgement is a thing a person may reasonably want. Paying for it with the next fourteen premiums is a price that was never quoted.

Which rupees actually belong in the decision

Here is where the popular version of this rule does damage, because "ignore what you have already paid" gets read as "ignore the cost of leaving" — and the cost of leaving is a future cost, which means it belongs in the comparison.

An exit charge, a surrender penalty, an exit load, the tax triggered by a switch: none of these has been incurred yet. Each of them happens because of the decision under consideration, and each differs between the options. They are the opposite of sunk. The test is not whether a payment feels like a penalty for the past; it is whether it is still avoidable by choosing differently.

Read the middle column and the rule reduces to one question asked of each line: does this number differ depending on what I choose? If it does not, it is doing no work. If it does, it is the decision.

The tax line is worth a sentence of its own because it is routinely put in the wrong column. Switching an equity-oriented holding realises whatever gain has accumulated, and gains above ₹1.25 lakh a year on units held beyond 12 months are taxed at 12.5% — treated in full in capital gains tax. That is not a punishment for having held the thing. It is the price of the switch, incurred in the future, and it belongs in the column for costs of leaving alongside the exit load.

The policy that gets harder to leave every year

Return to the ₹60,000 policy, because an annual commitment has a structural feature that a one-off purchase does not: it presents the same decision again every twelve months.

At the second premium, ₹60,000 sits behind the decision. At the seventh, ₹3.6 lakh does. Nothing about the forward comparison has necessarily changed — the same premium buys the same benefits inside the same contract — but the apparent case for continuing has grown every year, because the number being cited in its favour is the number that grows. Call it the ratchet. Each year the argument for staying strengthens on a quantity that carries no information about whether staying is worth it.

The ratchet also fixes something about when the question is cheapest to ask. The quantity available to argue for continuing is smallest at the start and larger every year after it, by construction, whatever happens to the policy's merits in the meantime. The one moment where nothing is yet sunk is the free-look window of 30 days after receiving the policy document, the only point in the contract's life at which the question can be asked with an empty column behind it — and even there a free-look return is a refund net of the deductions the insurer is entitled to make, not a reversal of the transaction.

Now the more useful correction. Framed as sunk cost, the decision looks binary: continue paying, or surrender and take whatever comes back. That framing hides a column. A traditional policy can usually be made paid-up once a minimum number of premiums has been paid — future premiums stop, the contract continues with reduced benefits, and no surrender charge is triggered. The comparison is three columns, not two: continue, paid-up, surrender.

That matters because the two costs point in opposite directions. Continuing costs every future premium. Surrendering costs the difference between the policy's value inside the contract and what it pays on exit, and in the early years of a traditional policy that difference is typically material. Paid-up removes the first cost without incurring the second. Whether it wins depends on numbers specific to the contract, but a decision framed as continue-or-quit never reaches the question.

And one genuine reason to continue that has nothing to do with what has been paid, which most content on this topic omits. If the policy carries life cover and your health has changed since it was bought, replacing that cover may be expensive or unavailable — and that is a forward cost of leaving, sitting properly in the last row of the table above. A reason can look like sunk-cost reasoning and be sound. The distinction is not how attached you are to the policy; it is whether the reason names something that happens after today. The bundling of cover and savings that makes this hard to untangle is the subject of endowment and money-back plans and ULIPs against mutual funds.

The holding, and the average price that keeps moving

The market version is the same mechanism with faster feedback, and it takes two forms worth separating.

The first is simply staying. A holding bought at ₹800 and quoted at ₹500 presents a decision about the next rupee, and the ₹300 difference is in the past under every answer. What makes this one hard to see is that the purchase price is doing double duty: it is the sunk total and the reference point against which the current price reads as a loss. That second job belongs to loss aversion and is treated there.

The second form is more specific and has a name in ordinary use: averaging down. Buy 100 at ₹800 for ₹80,000, then 100 more at ₹500 for ₹50,000, and the ₹1.3 lakh across 200 shares gives an average of ₹650. Those are chosen round figures, not any real scrip. Watch what changed. The average cost fell by ₹150, the price on the screen did not move, and nothing about the underlying business is different from what it was before the second purchase.

So the arithmetic of the average is real and the comfort it provides is misdirected. The second ₹50,000 is a new decision about ₹50,000, and it competes with every other use of ₹50,000 you have. Averaging down is a coherent thing to do when the answer to that question is this holding. It is a different thing entirely when the answer is that the average needed lowering.

The fund version is quieter and more common. "I have been in this scheme for eight years, switching now wastes those eight years" describes nothing that switching can waste. The returns those years produced are already in the corpus and travel with you; what does not travel is the exit load if you are inside the load period and the tax on the realised gain, which are forward costs and belong in the comparison exactly as the table above puts them. A rule that fires on drift from a target weight rather than on what a holding cost is the standard way of taking the entry price out of the loop — rebalancing does this by construction.

Recognising the ledger in a sentence

None of this is diagnosable from outside, and the tells are not proof of anything. What each one identifies is a sentence that has quietly changed the subject — from what the next rupee buys to what the previous rupees deserve.

The clearest is the appearance of a total in a forward argument. "I have already put ₹3.6 lakh into this" is a true statement that answers a question nobody asked; the question was about the seventh premium. A useful habit is to finish the sentence out loud, because the completion is usually where the reasoning breaks — "…so the next ₹60,000 will do better inside than outside" does not follow from it and is not being claimed.

The second is the reframing of a decision as a completion. Plans acquire a shape, and once something is described as half-finished, stopping reads as abandonment rather than as a choice. Notice that the halfway point was set by an earlier decision, which is also the thing under review. A commitment cannot be its own justification.

The third is the stranger test, and it is the one that removes authorship from the problem. Staw's finding was that the difficulty attaches to having made the original choice, so the question that isolates it is: if this policy had been bought by someone else and handed to you today, with its current value and its remaining premiums, would you keep paying? A different answer to that question and to the real one is a measurement of how much the authorship is worth, in rupees, which is a number worth seeing.

And the mirror error, because articles on this subject reliably produce it. Reading "ignore sunk costs" as a general licence to abandon things is the same failure with the sign reversed: it drops the exit charge, the tax and the replaceability of what you hold, all of which are future costs. The rule removes exactly one input from the decision. It does not simplify the rest of it, and a decision made quickly in its name is not obviously better than the one it replaced.

Putting the forward comparison next to the sunk one

Everything above reduces to running the comparison with the past total left out of it, which is easier to say than to do while looking at a statement that leads with what you have paid.

FNOTrader's Mutual Funds app runs lumpsum and instalment schedules against the full AMFI NAV history — around 34 million NAV rows — and reports invested amount against value, the compound annual growth rate on a lumpsum, the internal rate of return for cashflows landing on irregular dates (XIRR) for instalments, the worst peak-to-trough fall along the path, and the distribution of outcomes across every available start date.

That last column is the one this article points at. A rolling-return distribution contains no entry price and no total paid — it describes what a scheme did across every starting month rather than the one that happens to be yours, which is a question about the scheme instead of about your ledger. The comparison it supports is the forward one: what a sum invested from today, in this against that, has historically been worth.

Past performance is a record of what happened and not an indication of what will. FNOTrader is not a SEBI-registered investment adviser and does not give investment advice.

Common questions

Continue reading

More in Behavioural Finance · App: Mutual Funds · Definitions: glossary · Free tools: calculators · All: every article