Sunk Costs: Why Past Spending Is Irrelevant to the Decision

A sunk cost is money already gone, unrecoverable, and identical in every option still under consideration, so it cannot rank those options and never enters an appraisal. Sthira Capital Partners paid Rs 52,00,00,000 of deal fees; from the next morning that figure is the same whatever it does. The Rs 45,00,00,000 for a plant not yet built is not sunk, and counts in full.

Underneath that sits a fact about what a decision is. A decision is a choice between futures, and the only thing that can separate one future from another is something that will still change. An amount that has already left the bank is present in every future on the table: it is there if the project is approved, there if it is refused, there if the whole thing is shut down. A quantity that is the same in every option cannot rank the options. The argument is entirely arithmetic rather than philosophy, and it holds even in the cases where obeying it feels wasteful.

What exactly makes a cost sunk?

Three things have to be true at once, and most readers can name two of them.

The first is that the money has already been incurred. Not planned, not budgeted, not approved in principle. Gone. The second is that it cannot be got back: there is no refund, no clawback, no resale that returns it. The third carries the actual weight, and it is also the one readers let go of first: no option now on the table changes it.

Because the third condition attaches the word to a specific list of choices rather than to the calendar, it turns a vague feeling about the past into a usable test. A cost is not sunk in the abstract. A cost is sunk relative to the options actually being weighed this morning. Change the options and the same rupees can wake up again.

Take a wedding. A household has paid a deposit on a hall for a date in November. If the hall keeps the deposit no matter what, and the only remaining choices are which caterer to hire and how many people to invite, that deposit is sunk against those choices: it is the same figure under every caterer. But if the choice is still between that hall and a different one that would refund the deposit on transfer, one of the options on the table now brings the deposit back, and the deposit is live again. Nothing about the money changed. The list of options changed.

What about a cost that is agreed but not yet paid?

Here is where the word committed does more damage than any other word in capital budgeting. A committed cost has been agreed and not yet paid. Signing feels like spending, and a signature has a date on it and a name at the bottom, so people treat a committed cost as settled. But a signature is not a payment, and the question is never how binding something feels.

If cancelling costs nothing, a committed amount is fully live and belongs in the appraisal at its face value. Run it through the third condition. In the future where the project goes ahead, the order costs what the order costs. In the future where it is dropped, the order was cancelled free and costs nothing. Two different amounts in two different futures, so the line stays in.

The interesting cases sit in between. A cancellation charge of a fifth of the order value splits the amount cleanly. The cancellation fifth will be paid in both futures, so it is sunk the moment the contract is signed, and the remaining four fifths are live. Nobody has to argue about whether the order was a mistake. The only question is what each future costs.

What about a cost that is sunk without looking it?

The mirror case is a spend on something still physically held. A machine was bought last year. The machine sits on the floor. The accounts carry it at its written-down valueWhatever an asset still shows in the books once the depreciation charged so far has been taken off what was originally paid for it.. Everybody in the room can see the thing, and seeing it makes the money feel less gone than it is.

The amount splits. The sum the machine would fetch if sold today is a live amount: it arrives in the future where the machine is sold and is forgone in the future where it is kept. The price paid for the machine, and the value the books now carry, are neither of them things any current option can change. Only the unrecoverable part of a past spend is sunk, and the resale amount is a live number that belongs in the decision at whatever it is today.

A scooter makes this concrete. The scooter cost Rs 90,000 two years ago and will fetch Rs 40,000 now. In deciding whether to keep it or switch to a monthly bus pass, the Rs 90,000 has nothing to say: it is spent under both plans. The Rs 40,000 arrives only under one of them, so it has a great deal to say. The value the accounts put on the scooter is a third number again, and it is in the decision by accident rather than by right.

So what is the test, in one question?

Everything above collapses into a single question that can be asked of every line on a sheet, one line at a time, without knowing anything about the project:

Does this amount differ between taking the project and not taking it?

If it differs, it counts, at its full amount, whatever that amount is. If it does not differ, it leaves the sheet. There is no third verdict, and there is nothing else to check. The test does not ask whether the cost was wise, whether it was large, whether somebody is embarrassed about it, or whether it appears in the accounts. Wisdom, size, embarrassment and bookkeeping are all real facts about the money, and none of them is the question.

The single question buys something worth noting. Costs do not have to be sorted into categories, with a memorised list of which categories are allowed in. One question suffices, together with a willingness to run it on two futures at a time. An internal rate of returnOne percentage that summarises a stream of cash: the discount rate at which the stream would be worth exactly what it costs to start it. or a discounted value is only as good as the lines fed into it, and this is the sorting step that happens before any of that arithmetic begins.

Why does a sunk cost still appear in the accounts?

Readers who have followed the argument this far often hit a wall here, and it is a fair wall. If the money is irrelevant, why is it recorded? Why does the auditor insist on it? Why does it show up in the return the sponsor reports to its investors?

Because a recorded cost and a relevant cost are two different objects doing two different jobs, and only one of them is being asked to help with a choice. The accounts answer what happened. An appraisal answers what to do next, and no rule says the same figures serve both.

An accounting record is meant to be complete. Its job is to give a faithful account of every rupee that moved, and a record that quietly left out inconvenient items would be worthless. An appraisal is meant to be incomplete. Isolating what one choice does means throwing away everything the choice does not touch. Completeness would ruin it, in the same way that a photograph of everything in a room shows nothing in particular.

The confusion produces the error later on, so it is worth naming plainly. People assume that a number appearing in a proper, audited, signed document has earned the right to appear in every document. It has not. The number has earned the right to appear in that one document.

What did the fees on the buyout actually buy?

Take the worked case. Sthira Capital Partners, invented, is a financial buyerOwnership is temporary in a buyer of this kind. Debt does most of the funding, and the plan from the first day is a later sale rather than a permanent home for the business. that buys the whole of Sankalp Industrial Systems Limited, also invented, at 8.50 times the base year figure for earnings before interest, tax, depreciation and amortisation of Rs 2,88,00,00,000. The multiple applied to that base gives an enterprise value of Rs 24,48,00,00,000.

Its uses of fundsA single column listing where every rupee raised for a transaction ends up. The column has to total exactly what was raised for it. come to Rs 27,20,00,00,000, and five lines make up that total. Four of them buy something.

The last two lines add to Rs 52,00,00,000, and there is a clean way to see that they bought nothing at all. The sponsor wrote a cheque for Rs 12,00,00,00,000 of equity. On the day the transaction completed, the business it had just paid for was worth Rs 24,48,00,00,000 and carried Rs 13,00,00,00,000 of debt, so the stake the sponsor held that same evening was worth Rs 11,48,00,00,000. The gap between what went in and what came out the other side is Rs 52,00,00,000, and it is exactly the fees.

Now apply the test. From the morning after completion, is there any decision the sponsor can take that alters that Rs 52,00,00,000? Sell the business. Hold it. Refinance the senior loan. Defer the third valve line. Sack the management team, or roll them a bigger stake. In every one of those futures the figure is Rs 52,00,00,000. The fee total is sunk, and it falls out of every decision the sponsor makes from that day until the day it leaves.

Does it enter the choice to sell?

Five years later the sponsor faces the cleanest possible test of the rule. At the end of Year 5 the business produces Rs 4,32,00,00,000 of earnings before interest, tax, depreciation and amortisation. Priced at the same 8.50 times, that is an enterprise value of Rs 36,72,00,00,000. Net debtBorrowings less whatever cash is sitting in the bank, so it measures what would still be owed to lenders if every available rupee were handed over today. stands at Rs 7,91,85,00,000. Taking the net debt off leaves Rs 28,80,15,00,000 for the equity.

The choice is to sell at that price or to hold on for longer. Two futures, both real, both defensible, and a genuine argument to be had about them. The Rs 52,00,00,000 of fees appears in neither branch. The fee total is the same figure whichever way the decision goes, and an amount that sits identically on both sides of a comparison has no power to tip it.

Where do those same rupees become decisive?

In exactly one place, and it is a different document. When the sponsor breaks down what it actually earned, the fees are a named line with a number and a share against it. Value created came to Rs 16,80,15,00,000, being the Rs 28,80,15,00,000 of exit equity less the Rs 12,00,00,00,000 of equity put in. The value created splits four ways.

Read that table and then read the sell-or-hold decision again. The same rupees are a named line in one document and are absent from the other, and knowing which document is being written is the whole of the skill. A reader who can hold both of those facts at once, without feeling that one of them must be a mistake, is finished here.

One small note on the last row. A column stops adding up here without anybody noticing. Divide Rs 52,00,00,000 by Rs 16,80,15,00,000 and the raw figure comes to 3.094962 per cent. Rounded once, straight from that, it is 3.09 per cent, and the four shares then total exactly 100.00. Take it in two hops instead, four decimals and then two, and out comes 3.10 with a column that no longer totals. Round once, from the full value, every time.

Is there a second amount on this transaction that behaves the same way?

There is another figure on this deal that behaves the same way, and it is worth seeing because it looks so much more like something a person could still do something about.

The record sets out the highest entry multiple at which this particular structure still delivers a 20.00 per cent return: 8.35 times, being an enterprise value of Rs 24,05,47,00,000. The transaction was struck at 8.50 times, being Rs 24,48,00,00,000. Subtract one from the other and the price paid was Rs 42,53,00,000 above that ceiling, and paying over the ceiling is the reason the return lands at 19.14 per cent instead of 20.00. The ceiling multiple is itself a rounded number, and turning it back into rupees a second way would give a slightly different answer, so the subtraction is the honest route to the figure and the one to use.

The moment the cheque clears, the Rs 42,53,00,000 of overpayment joins the fees in the same category. Every hold-or-sell decision afterwards is made on the business as it now stands, at the price a buyer would now pay, and never on the price this buyer once paid. The overpayment is a real fact, it is a fair thing to raise at an investment committee reviewing how the deal was underwritten, and it is not a variable in any choice that remains.

So which of the company's own figures is live rather than sunk?

Set the deal aside and look at a number from the company itself. Project 5 is the effluent treatment plant, and the record sets it out like this.

The 12.00 per cent is the rate this company applies to everything it appraises, and it is settled under cost of capital. The plant itself goes up regardless of the answer. The site's consent to operatePermission from an environmental regulator to keep a plant running, granted with conditions attached that the site has to keep meeting. requires it.

The Rs 45,00,00,000 has not been spent. Run the same question: it is Rs 45,00,00,000 in the future where the plant is built and it is nothing at all in the future where it is not. The outlay differs, so it counts, in full, at its full amount. Far from being irrelevant, it is the single most relevant number on that project, and it is relevant for precisely the reason the fees are not.

Put the two side by side and notice what they show together. Relevance has nothing to do with size: Rs 52,00,00,000 is larger than Rs 45,00,00,000 and drops out while the smaller figure stays. Relevance has nothing to do with how painful the amount is, or with whether anybody regrets it, or with how recently it was agreed. Relevance turns on one thing only, and that thing is whether the amount still moves.

What does putting a sunk cost back in do to the answer?

Suppose somebody charges the Rs 52,00,00,000 of fees against a project it has nothing to do with. Not out of malice, and not out of stupidity. Usually because a cost has to be recovered from somewhere, and a live project with a positive value looks like a reasonable somewhere.

Project 1, the third valve line, stands at Rs 34,31,04,532. Load the fees onto it and it shows minus Rs 17,68,95,468. Project 2, the automation cell, stands at Rs 22,09,55,240. Load the fees onto that and it shows minus Rs 29,90,44,760. Both are rejected, and the company loses the value each of them would have added.

The error always runs in one direction. A sunk cost enters an appraisal as a subtraction, and subtractions push one way only: the project reads poorer than the facts warrant. Nothing about that can flatter a weak project. So the mistake never produces an approval it should not have produced; it produces a rejection it should not have produced.

The asymmetry is why the error is so rarely caught. An approved project that underperforms gets scrutiny, a post-completion auditA look back, once a project has run for a while, comparing what it actually delivered against what was promised for it at the time of approval., a set of awkward meetings and sometimes a change of personnel. A project rejected for the wrong reason files no report, produces no variance, and appears in no review. The rejected project leaves the building quietly, and nobody ever learns what it would have earned.

Is there an error that runs the other way?

Every treatment of sunk costs warns against putting one back in. Very few warn about the opposite error. The opposite runs the other way and does more damage when it happens.

The opposite is leaving out a cost that is not sunk at all. The omission happens when somebody hears the rule, likes it, and starts treating anything with a history as ancient. The land the new line will stand on was bought years ago, so it feels historic and gets omitted, when in fact the land could be sold and the sale proceeds are given up by building on it. A machine already on the floor gets treated as free, when it could have been sold or redeployed. A team already on the payroll gets treated as free, when the alternative is that they work on something else.

Leaving out a live cost is a subtraction that never happens, so it pushes the answer up and produces approvals that should not have been given. And unlike the first error, this one has consequences that show up in the accounts eventually: the project gets built, the money gets spent, and the return comes in below what the paper promised. The second error is caught more often, but only after the cash has gone.

Who makes the escalation error, and why do they keep making it?

Why does the rule feel wrong, and what is the honest answer?

The rule feels wrong because it looks like permission to waste money. Somebody spent Rs 52,00,00,000 and the rule says to stop thinking about it. Everything in a person that treats money seriously objects, and the objection is decent rather than foolish.

The honest answer is that the loss has already happened and nothing available now can undo it, so applying the rule does not make the loss go away. The rule does not forgive the spending or approve of it. The rule does one narrower thing: it stops the loss from choosing the next move as well as the last one.

The distinction between undoing a loss and containing it is the whole comfort on offer, and it is worth more than it sounds. A household that has paid a non-refundable deposit on a hall it now dislikes has lost that deposit either way. Holding the wedding somewhere better does not lose it twice; it only loses it once and gets a better wedding. Holding the wedding in the disliked hall to justify the deposit loses it once and gets a worse wedding. The money is identical in both stories. Everything else is not.

How does anyone use this outside a textbook?

Four people meet this rule in four different rooms, and each of them uses it to throw something away.

A lender on a credit committee is asked to fund the next tranche of a project already half built. The borrower's case leans heavily on the amount already drawn and the concrete already poured. The lender's only real question is whether the remaining cost, funded now, is covered by what the finished thing will produce and what it could be sold for. The concrete already poured is worth exactly what somebody would pay for a half-built plant and not one rupee more, and that resale figure is the live number in the room.

An equity analyst reads a company announcing that it is abandoning a project after spending heavily on it. Two readings are available. One is that the company has wasted money, a true statement about the past. The other is that the company applied this rule, a fact about how the company makes decisions and therefore a statement about the future. A company that can stop is worth understanding differently from one that cannot, and the announcement is one of the few public places that difference is visible.

An investor in a fund like Sthira Capital Partners looks at the money multipleHow many times over an investor gets the original cheque back, counted in rupees rather than as a rate per year. and the decomposition table, sees the fee line at minus 3.09 per cent, and learns something real about the cost of doing transactions. Reading the fee line that way is the correct use of the backward-looking document. The same investor cannot use the same figure to argue that the sponsor should hold the business longer to earn the fees back.

And a household does this every month without a name for it. Money spent on a course no longer wanted, a deposit on a flat since decided against, a phone repaired last year that has broken again. The right question is never how much has gone in, but only what the next rupee buys compared with what else it could do.