The Sunk

A past outlay can remain painful long after it has ceased to be a reason for choosing one future over another. The earlier Quantities Cost made the marginal principle plain: what is gone is gone in equal measure whether one presses on or turns back. The earlier Collected Possession supplied the difficulty of obeying it, the account we recoil from closing in the red. Together they expose a peculiar burden laid upon the next decision: it is asked to vindicate an earlier one, even when the alternatives still available cannot do so. The sunk-cost principle removes that burden by comparing what each choice can now change. Its force depends on the comparison being complete. Past spending may have produced knowledge, equipment or obligations that matter today; the irrecoverable payment itself must be distinguished from those surviving consequences. Only then can we see when persistence serves a future and when it merely postpones admitting a past.

The arithmetic of a sunk cost can recommend continuing an enterprise that will never repay its total expense. Take an invented project on which £90 has been irrecoverably spent. Finishing now costs another £20 and yields a certain £30; stopping yields nothing, and there are no other effects or opportunities. Continuing improves the position by £10, although the project’s lifetime balance is a loss of £80. Stopping would leave a loss of £90. Now change only the remaining expense to £40. Finishing would produce a loss of £100, so stopping is better by £10. The same past £90 accompanies opposite decisions because the future comparison has changed. Demanding that the whole undertaking break even would reject a worthwhile completion in the first case; insisting that the first £90 be honoured would authorize a further loss in the second. The relevant threshold is what completion adds from here.¹

Experimental choices show how an unrecoverable price can acquire authority over the enjoyment still to come. Arkes and Blumer asked students to imagine having paid $100 and $50 for two ski trips that accidentally fell on the same weekend. Neither ticket could be sold or refunded, and the less expensive trip was explicitly expected to be more enjoyable. Thirty-three of sixty-one respondents nevertheless chose the more expensive trip. Their separate field experiment assigned discounts to people already buying theatre subscriptions: full-price purchasers used more tickets in the first half of the season in the reported planned comparisons, but the difference was not significant in the second half. These are bounded findings. The ski choice was hypothetical, and a difference in attendance alone does not measure lost welfare. What the studies make visible is the possibility that the payment’s history continues to influence use after its amount has become unrecoverable.²

Closing an account at a loss supplies one explanation for the payment’s lingering force. In Thaler’s mental-accounting treatment, consuming what was prepaid can complete an ordinary transaction, while abandoning it makes the failed expenditure conspicuous. The earlier Possession followed that logic from the tedious film to the failing venture: keeping the undertaking alive can postpone the moment at which its loss is acknowledged. There is a moral vocabulary ready to assist. Finishing looks faithful, stopping wasteful, and another payment can resemble an act of care for the money already spent. Yet the two decisions in the invented project reveal how readily that care misfires. Finishing at £20 salvages £10 of value; finishing at £40 destroys another £10. Respect for the resources that remain requires telling those cases apart. A plausible psychology of persistence must explain why the earlier expense clouds that distinction, without assuming that every person or every continued project is governed by the same mechanism.³

The part of an earlier purchase that can still be recovered belongs in today’s comparison. Suppose, as a further invented variation, that the unfinished project could be sold for £15, while completing it still costs £20 and produces £30. Completion now offers £10 from this point; sale offers £15. Selling is better by £5, even though the comparison without salvage favoured completion. The sale option changes what can be recovered, so the earlier assumption of a wholly unrecoverable outlay no longer holds. The decision must count the asset’s present £15 value, which continuing would forgo. Time has the same distinction. Hours already spent cannot be returned, while the next hour can be used elsewhere. The familiar injunction to ignore the past becomes careless if it erases present assets or alternatives along with it. Marginal reasoning asks what survives and what can still be changed, and charges the decision with every opportunity it uses up.

Past expenditure can also carry information that changes the forecast of what another expenditure will achieve. McAfee, Mialon and Mialon make the point with a project whose total completion cost is uncertain: failure after a given investment changes the conditional probability of finishing with the next increment. In their model that information can favour either persistence or withdrawal, depending on the distribution of costs. An invented illustration makes the distinction concrete. Suppose completion requires either £60 or £100, initially equally likely; success returns £90, and only one further £10 investment is permitted. Spending £10 after £50 has been spent would have a one-half chance of completion. If £60 has already been spent without success, the remaining requirement is known to be £40, so another £10 cannot finish it. The expenditures are still unrecoverable; what differs is the knowledge available for choosing. Using that knowledge follows the prospective rule. Counting the past payment again as a benefit of continuing would violate it.⁴

The prospective rule also leaves room for commitments whose consequences extend beyond a single project. A reputation for keeping promises may encourage others to invest with you; abandoning a shared undertaking may impose costs on them. Conversely, a manager may preserve a career by concealing a failed decision while spending somebody else’s resources. Those cases require identifying whose future is being served, and including the consequences that the isolated project account leaves out. They cannot be settled by the word persistence. The question posed by this plate therefore has two answers that must be kept in view together. We can honour an unrecoverable outlay because admitting the loss is painful, allowing an account of yesterday to command tomorrow. We can also attend to past investment because it has changed today’s information, opportunities and responsibilities. The discipline is to carry those consequences forward without granting the bare expenditure an additional vote. The past remains available for learning, judgment and accountability; the next choice must answer for the difference it can still make.⁵