Nobody Has To Be Wrong

The eightieth article was about a finding whose exceptions mattered more than the finding. This one is about something rarer in this series: a result that cannot fail to be true, because it follows from the definition of winning.

The Verdict, Stated First

Five claims, in descending order of confidence.

One. The mechanism is arithmetic and cannot be disputed. Winning selects for the highest estimate, so the winning estimate is the maximum of the estimates, and the expected maximum exceeds the true value.

Two. No bias is required and that is the point. Every bidder can be perfectly calibrated and the winner still overpays, which makes this the only finding in this series that survives everybody being rational.

Three. More competition makes it worse, not better. On our own arithmetic the required shading rises from 8.5 percent against two bidders to 28.0 percent against twenty, on a 15 percent dispersion.

Four. It is survivable and the arithmetic says exactly when. Your margin must exceed the expected maximum times your dispersion, which at a 20 percent margin against ten bidders permits a dispersion of 13.0 percent and no more.

Five. At least one industry has learned to correct. A 1996 paper's title asserts that commercial construction does, which is the most encouraging sentence in this article and one we could not read.

Our Grades For These Claims

Applying the scheme from the first article in this series.

Grade A for our own arithmetic, which is exact: the expected maximum of two standard normal draws is one over the square root of pi, and we checked our numerical values against that and against the closed form for three.

Grade A for the citations, confirmed across five independent reference lists carried on peer-reviewed articles.

Grade C for the mechanism's clearest statement, which reaches us from an encyclopedia rather than from a paper, flagged at every use.

Grade D for everything about what the studies found, because we obtained no abstract and no result from any of them, only titles and citations.

Our position: this is the article in the series where our own arithmetic is stronger evidence than our sources, which is an unusual and slightly uncomfortable position to be in.

A Note On Method

Everything here is verified to August 2026, and the sourcing is unusual in a way a reader should know about.

We obtained no abstract from any of the underlying papers. Every empirical claim below reaches us as a title, a citation, or an encyclopedia's summary, and we grade them accordingly.

The mechanism's clearest statement comes from an encyclopedia entry[1], flagged at every use.

The citations are confirmed across five independent reference lists carried on peer-reviewed journal articles[2][3], which establishes that the papers exist and where, and nothing about what they say.

One testable formulation and its result reach us from a journal abstract[4], which is the single piece of empirical content in this article that comes from a publisher.

All arithmetic is ours. The expected-maximum figures are exact; the dispersions, margins and contract values are invented.

This article discusses research on bidding. It is not bidding, valuation or transaction advice.

The Mechanism

The setup, in the clearest statement we found.

An encyclopedia entry describes the original setting: "OCS auctions are common value auctions where the value of the oil in the ground is essentially the same to all bidders. Bidders have their own estimate of the (unknown) value at the time that they bid."[1]

Then the consequence: "If bidders ignore this adverse selection effect inherent in winning the auction, it will result in below normal or even negative profits. The systematic failure to account for this adverse selection effect is referred to as the winner's curse: you win, you lose money, and you curse."[1]

This is an encyclopedia, not an academic source, flagged here and at every use.

Four observations, ours.

The key phrase is common value. The thing being bid for is worth the same to everyone, and only the estimates differ. A painting somebody loves is a private value and does not behave this way.

The term adverse selection is the technical name and it is precise. Winning is not a neutral event; it is a filter, and what it filters for is optimism about the value.

The definition given is of the failure to account for the effect, not of the effect itself. The selection happens regardless; the curse is what happens when you have not adjusted for it.

And that distinction is the whole article. The arithmetic is unavoidable and the loss is not, which puts this among the more actionable things in the series.

Two further points about the framing, because the word curse invites a misreading. Nothing has gone wrong with the auction. A competitive process that transfers value from bidders to the seller is working exactly as intended, and the seller is not doing anything improper by running one.

And the party who suffers is the one who did not do the arithmetic, not the one who bid least carefully. A meticulous estimator who does not adjust for selection loses to a sloppy one who does, which is an uncomfortable ranking and follows directly.

Even If Unbiased

The clause that makes this different from everything else in this series.

The same entry: "Even if these estimates are unbiased, bidders must account for the informational content inherent in winning the auction: the winner's estimate of the common value is (one of) the highest estimates."[1]

Four observations, ours.

Even if these estimates are unbiased. Nobody is overconfident, nobody anchors, nobody is subject to any of the eighty findings this series has covered, and the loss still arrives.

That makes it a structural problem rather than a psychological one, and structural problems have the useful property that they can be solved by arithmetic rather than by self-improvement.

It also means the usual remedies do not apply. Better estimating does not help; the curse operates on the dispersion of estimates, and a room full of excellent estimators who disagree slightly still produces it.

And it explains why this survived economists' scrutiny when so much behavioural work did not. You do not have to believe anything about psychology to accept it, only that winning correlates with having guessed high.

One consequence for how this article should be read, and it cuts against the usual shape of this series. The finding needs no defending and the behaviour does. Nothing about the arithmetic is contested; the only open question is whether real bidders adjust for it.

Which is why the two most valuable papers named below are the ones asserting that they do. The interesting empirical question is not whether the curse exists but whether anyone still falls for it, and that is a question about practice rather than about mathematics.

The 1971 Origin

Where it comes from.

Capen, E. C., Clapp, R. V., and Campbell, W. M. (1971), Competitive Bidding in High-Risk Situations, Journal of Petroleum Technology, 23(6), 641–653, June[2].

The encyclopedia records the claim: "three petroleum engineers, wrote an article in which they claimed that oil companies suffered unexpectedly low returns 'year after year' in early Outer Continental Shelf (OCS) oil lease auctions."[1]

We did not obtain the paper.

Four observations, ours.

The authors were petroleum engineers, not economists or psychologists, which is worth registering. This came from practitioners noticing their industry losing money.

The journal is a trade and engineering journal, not an economics one, and the finding entered economics afterwards.

The phrase "year after year" is the observation that made it a problem rather than bad luck. A single disappointing auction is noise; a pattern across years is a mechanism.

And the setting is close to ideal for the effect. Oil under the seabed is worth the same to whoever gets it, and nobody knows how much is there, which is the common-value case in its purest commercial form.

One point about why it took until 1971 to notice, which is worth a sentence because the arithmetic is not difficult. The effect is invisible in any single transaction. A disappointing lease looks like a dry hole, and a dry hole is what everyone in the industry expected some of the time.

It only appears in the aggregate, which requires both a long record and someone willing to look at it. The observation needed a pattern across years, and the pattern needed somebody to be counting, which is the seventy-fourth article's conditions arriving in an oil field.

The 1983 Experiment

The laboratory version, which we can name and not describe.

Bazerman, M. H., and Samuelson, W. F. (1983), I Won the Auction But Don't Want the Prize, Journal of Conflict Resolution, 27(4), 618–634, December[2][3].

We did not obtain it and report the title and citation, confirmed across five independent reference lists.

Three observations, ours.

The title states the finding as an experience rather than a statistic, which is why it is quoted so often, and a title is not a result.

The journal is one on conflict resolution rather than economics or marketing, which is a slightly odd home and reflects the period's interest in bargaining.

And its presence in five separate reference lists across finance, economics and operations research establishes that the paper is foundational across several fields, which is the most we can say without reading it.

What Winning Tells You

The arithmetic, which is the heart of this article. Ours, and exact.

Take N bidders, each with an unbiased estimate of the same true value, each off by a random amount with standard deviation S. The winner is whoever estimated highest, so the winning estimate is the maximum of N draws.

The expected maximum, in units of S, and what it means if S is 15 percent of the value:

2 bidders: 0.564 times S, or 8.5 percent.

3: 0.846, or 12.7. 4: 1.030, or 15.4. 5: 1.163, or 17.4. 6: 1.267, or 19.0. 8: 1.424, or 21.4.

10: 1.539, or 23.1. 15: 1.736, or 26.0. 20: 1.868, or 28.0. 30: 2.043, or 30.6 percent.

Four observations.

These figures are exact rather than estimated. The expected maximum of two standard normal draws is one over the square root of pi, which is 0.5642, and of three is three over twice the square root of pi, which is 0.8463. Our numerical values match both.

The normality assumption is ours and is the one soft spot. Estimate errors need not be normally distributed, and a different distribution moves these numbers while leaving the direction untouched.

The growth is sublinear: going from two bidders to twenty roughly triples the overestimate rather than multiplying it tenfold.

And the practical reading is a single instruction. Whatever your estimate says, the fact that you are about to win says it was high, and the table gives the size of the correction.

Two properties of that table are worth drawing out, because they change what a firm should work on.

The correction is proportional to your dispersion. A firm whose estimates vary by 5 percent faces a third of the problem a firm at 15 percent faces, at every field size, which makes consistency the highest-value thing in an estimating department.

And the correction is independent of your accuracy. Being right on average is exactly what the table already assumes, so a firm that improves its average and not its spread has not moved a single number in it.

No Bias Required

Stating the implication as directly as we can. Ours.

Four observations.

Every number in that table assumes perfectly unbiased estimators. The average estimate across all bidders is exactly right, in every row.

What produces the loss is not error but selection. The auction hands the contract to the person whose error happened to point in one direction, and that is what an auction is for.

Which means the effect gets worse as the field gets better informed only insofar as they agree. Reducing dispersion helps; adding bidders hurts; and improving average accuracy does nothing at all.

And that last point is the one to carry into an estimating department. Effort spent making estimates less variable is worth more than effort spent making them more accurate, on this arithmetic, which is a genuinely counterintuitive management conclusion.

One qualification on that, because stated baldly it is too strong. Accuracy still matters enormously for everything other than winning: pricing, scheduling, cash planning and knowing whether a job is worth doing at all.

The claim is narrower and still useful. For the specific problem of not overpaying in a competitive process, spread is the term that multiplies and average accuracy does not appear, and most estimating improvement programmes are aimed at the wrong one of the two.

The Part That Runs Backwards

The consequence most people get wrong. Ours, from the same table.

Intuition says a crowded tender means bidding harder. The arithmetic says the opposite, because winning against more people is worse news:

Against 2 bidders, shade down 8.5 percent. Against 3: 12.7, up 4.2 points. Against 5: 17.4, up 4.8. Against 10: 23.1, up 5.6. Against 20: 28.0, up 4.9.

Four observations.

More competitors should make you bid less aggressively, not more. That is exactly backwards from how competitive pressure feels.

The reason is not that you are less likely to win. It is that winning has become stronger evidence that you were the optimist, and the correction has to grow with the field.

The behavioural failure mode is therefore predictable and specific. Competitive pressure pushes bids up while correct reasoning pushes them down, and the two forces point opposite ways at exactly the moment the stakes feel highest.

And this is the single sentence worth taking from the article if a reader takes only one. A crowded tender is a reason to bid lower or not at all, which is not what a crowded tender feels like.

What It Costs To Ignore

Putting money on it. Our own arithmetic, with invented figures: a $500,000 contract, estimates dispersed by 15 percent, and you bid your own honest estimate with a 12 percent margin built in.

On a low-bid tender you win by having estimated the cost lowest, so the winner's cost estimate is too low by the same amount the table gives:

Against 2 bidders, your estimate is 8.5 percent under the truth, leaving a realised margin of plus 3.5 percent.

Against 3: 12.7 under, realised minus 0.7. Against 5: 17.4 under, realised minus 5.4. Against 10: 23.1 under, realised minus 11.1. Against 20: 28.0 under, realised minus 16.0 percent.

Four observations.

A 12 percent margin survives two competitors and fails against three, on these invented figures, which is a much thinner cushion than it sounds.

At five bidders the job loses 5.4 percent, which on a $500,000 contract is about $27,000 against an expected profit of $60,000.

The pattern in a business is the one that hides it. You win the crowded jobs and lose money on them, and win the quiet ones and make money, and the average across a year looks like ordinary variability.

And this is the mechanism behind a complaint every trade makes. Winning too much work at bad prices is not a pricing failure, it is a selection effect, and it is arithmetic rather than a market being irrational.

One diagnostic follows immediately and costs nothing to run. If your win rate is high and your margins are thin, those two facts are the same fact, and the arithmetic above says which way the causation runs.

The inverse is the reassuring version and worth stating too. A low win rate with good margins is what correct bidding looks like, and a firm being told to bid more aggressively should ask what the realised margin on the extra work would be.

How Much Dispersion You Can Afford

The survivability condition. Ours, and it is the useful form.

Your margin must exceed the expected maximum times your dispersion. The largest dispersion a given margin survives:

At a 10 percent margin: 17.7 percent dispersion against 2 bidders, 8.6 against 5, 6.5 against 10, 5.4 against 20.

At 15: 26.6, 12.9, 9.7, 8.0. At 20: 35.4, 17.2, 13.0, 10.7. At 30: 53.2, 25.8, 19.5, 16.1. At 40: 70.9, 34.4, 26.0, 21.4.

Four observations.

A 20 percent margin against ten bidders survives a dispersion of 13.0 percent and no more. If two competent estimators in your own firm routinely differ by more than that, open tendering at ten bidders is a losing proposition.

The test is measurable inside your own business, which is the reason we like this one. Have two people estimate the same job independently, several times, and look at the spread.

Low-margin businesses have almost no room at all. A 10 percent margin against twenty bidders permits a 5.4 percent dispersion, which is tighter than most estimating processes achieve.

And the condition explains why some trades tender and others refuse to. Work with predictable costs can be tendered and work with uncertain costs cannot, and that is a statement about dispersion rather than about competitiveness.

Two ways to shift the condition rather than accept it, ours. Reduce dispersion by standardising the estimate: a checklist, historical unit costs, and a second reviewer all narrow the spread, and the spread is what multiplies.

Or reduce the field, which means competing where fewer firms can. Specialisation is usually justified on margin grounds and it has this second justification, which is arithmetically larger than the first at high bidder counts.

But An Industry Corrects For It

The most encouraging thing in this article, and we can only report its title.

A reference list records: Dyer, D., and Kagel, J. (1996), Bidding in common value auctions: how the commercial construction industry corrects for the winner's curse, Management Science, 42, 1463–1475[2].

We did not obtain it. The title asserts that the correction happens and we cannot tell you how.

Four observations, ours.

The title's verb is "corrects", in the present tense and without hedging, which is an unusually confident thing to put in a title.

If it holds, it is a direct counter to the pessimistic reading of the laboratory work. Experienced professionals in a real industry with real money at stake do not simply suffer this, they have developed something.

It also fits the seventy-fourth article's conditions. Construction bidding gives frequent, unambiguous feedback, being whether the job made money, which is exactly the environment where skill can develop.

And this is the paper we would most want to read, because it presumably describes the correction, and a business could adopt it. That we could not obtain it is the largest gap in this article.

What That Title Implies

Reasoning from the title alone, flagged as such. Ours.

Four observations.

A correction that an industry develops informally is unlikely to be an expected-maximum calculation. It is more likely a rule of thumb, and rules of thumb for this problem are easy to imagine.

The obvious candidates are all observable in construction practice: refusing to bid on crowded jobs, adding a contingency that scales with the number of bidders, walking away when you win by an unusual margin, and pricing off historical cost rather than a fresh estimate.

Every one of those has the right shape. They reduce either the dispersion or the exposure to the field size, which are the two terms in the condition above.

And we would flag hard that this is us reasoning from a title. The paper may describe something entirely different, and a reader who needs the answer should obtain it rather than take ours.

A Sceptical Title Too

Because the literature is not unanimous and the same reference list shows it.

It records: Cox, J., and Isaac, M. (1984), In search of the winner's curse, Economic Inquiry, 22, 579–592[2].

We did not obtain it, and the title's phrasing suggests scepticism about finding it.

Four observations, ours.

"In search of" is not neutral. Papers titled that way usually report difficulty locating the thing.

That is entirely compatible with everything above, and worth understanding why. The arithmetic is not in doubt; what is in doubt is whether people fail to correct for it, and those are different claims.

So a sceptical paper here would be arguing that bidders do adjust, which is the same direction as the construction finding rather than a contradiction of it.

And we would not overstate our reading of a title. We are reporting that a paper with a sceptical-sounding title exists in this literature, and that a reader should know the question was contested.

The Takeover Version

The application to buying companies, and the one place we have an actual abstract.

A journal abstract states the hypothesis: "The winner's curse hypothesis states that, in any bidding situation, a party which unknowingly overestimates the value of a given object tends to bid higher than its competitors and is, therefore, more likely to win it."[4]

And defines the measurement: "In a takeover the magnitude of the winner's curse is defined as the difference between the bid premium of the winning bidder and the maximum offerable premium conditional on the capital market's estimate of expected takeover gains."[4]

Three observations, ours.

That definition is operational rather than conceptual, which is what makes the study testable: it turns the curse into a number you can compute from market data.

It also imports an assumption worth flagging. The benchmark is the capital market's estimate, so the test measures overpayment relative to the market's view rather than to the truth.

And the abstract's own phrasing contains an error we quote as found. It reports support for the "winner's course hypothesis"[4], which is a typo we return to below.

Three Predictions

The part of that abstract worth an owner's attention.

It records that the magnitude "is predicted to increase with (1) increase in the divergence of opinion amongst acquirers with respect to the size of takeover gains, (2) increase in the degree of competition for control of the target firm and (3) increase in the pre-acquisition profitability of the winning bidder." And: "The empirical results provide support."[4]

Four observations, ours.

The first two predictions are exactly the two terms in our own arithmetic: divergence of opinion is the dispersion, and degree of competition is the number of bidders. That correspondence is not a coincidence and it is a check on both.

The third is not in our arithmetic at all and is the interesting one. More profitable acquirers overpay more, which is a claim about capacity or discipline rather than about estimation.

The obvious reading of the third is that a firm with money is less constrained by having to be right, and the sixty-third article's findings on slack would be the place to look.

And we did not obtain the paper, only this abstract, so we report the predictions and the claim of support without any of the results behind them.

Where It Does Not Apply

Because a mechanism this general invites overuse, and the conditions are specific. Ours.

Four observations.

It requires common value. An auction where bidders genuinely want the thing for different reasons, and would pay different amounts having seen everything, does not produce the effect.

It requires uncertainty about that value. Bidding for something whose worth is known and agreed produces competition on price alone, which is a different and less dangerous problem.

And it requires the bid to determine the payment. A tender where price is one criterion among several, weighted against quality and references, weakens the selection because winning no longer means having bid most aggressively.

That last point is the most useful for a small firm, and it is a reason to prefer some work over other work. Qualification-based selection is not merely more pleasant to compete in; it is arithmetically safer, because winning carries less information about your estimate.

Two Other Places It Shows Up

Applications outside bidding that follow from the same arithmetic. Ours, and untested.

Four observations.

Hiring. Several employers assess the same candidate, and the one who offers most has estimated their value highest. The candidate who accepts your offer over others is the one you valued above the market.

That is not an argument against hiring, and it is a reason to be careful about a specific case. A candidate you want much more than anyone else did is worth a second look, and the gap is the size of the question.

Selecting suppliers on price. The supplier who quotes lowest for the same specification is the one whose cost estimate was most optimistic, which is a reason cheap quotes overrun.

And that reframes a familiar frustration. The lowest bidder overrunning is not usually dishonesty; it is the winner's curse operating on your supplier, and you selected for it.

One consequence for how a small firm should buy, ours and untested. Three quotes are safer than ten, on this arithmetic, because a wider field selects harder for the most optimistic estimate and the optimism is what you eventually pay for.

And a quote well below the others deserves the same suspicion as a bid you won by a wide margin. The outlier is not usually a bargain; it is the estimate most likely to be wrong, and asking what it includes costs nothing.

What Actually Survives

Our reading, stated directly.

Five statements.

The selection effect is arithmetic and cannot fail to operate. Winning selects for the highest estimate, and the expected maximum of N unbiased estimates exceeds the truth.

No bias in any bidder is required, which makes this the only finding in this series that survives everyone being perfectly rational.

More competition requires more shading, not less, which points opposite to how competitive pressure feels.

The survivability condition is exact: margin must exceed the expected maximum times the dispersion, and both terms are measurable inside a business.

And at least one industry reportedly corrects for it, which is the finding we would most want to read and the one we have least of.

Your Tenders

The first application. Ours, and not bidding advice.

Four points.

Ask how many are bidding before you price, because on our own arithmetic the required shading roughly triples between two bidders and twenty.

Measure your own dispersion, by having two estimators price the same job independently several times, which turns the survivability condition from a formula into a number about your firm.

Treat winning as information rather than as an outcome. Winning a crowded tender by a wide margin is evidence you got something wrong, and the margin of victory is the size of the evidence.

And keep the record the seventy-fourth article recommends. Realised margin against bid margin, by number of bidders, which would show this effect in your own data within a couple of years if it is there.

Two notes on that record, because it is the most valuable thing in this section and the easiest to get wrong. Record the number of bidders even on jobs you lose, since the lost bids are what tell you whether your shading is too heavy rather than too light.

And record your estimate at bid time, not the final cost report. The comparison that matters is between what you thought when you priced it and what happened, and a cost report written afterwards has absorbed everything you learned.

Buying A Business

The second application, where the amounts are larger and the frequency lower. Ours, and not transaction advice.

Four points.

An acquisition is the common-value case in its strongest form. The business generates the same cash whoever owns it, at least for the parts of it that are not synergies, and only the estimates of that cash differ.

The dispersion in a small-business valuation is large. Reasonable people disagree by well more than 15 percent on a private company, which puts the required shading in the range our table calls severe.

And the frequency is the problem the seventy-fourth article identified. A buyer who acquires once cannot learn this from experience, which is precisely where the arithmetic has to substitute for the feedback.

The one genuine escape is real and worth stating. If the business is worth more to you than to others for a specific reason, it is not a pure common-value auction, and the curse is correspondingly weaker. That defence requires the reason to be specific.

The One Defence

The condition under which none of this applies. Ours.

Four observations.

The whole mechanism requires common value. If the thing is genuinely worth more to you than to the other bidders, you can win without having overestimated anything.

That is the honest justification for paying up, and it has a test. Name the specific reason, in dollars, before you bid, and check whether it would survive someone asking why the other bidders could not do the same thing.

The failure mode is that every bidder believes they have one, and the word usually used is synergies. A reason everybody has is not a reason.

And there is a version of the defence that genuinely works and is unglamorous. A neighbouring business, an adjacent contract, a crew already on site, are all specific, checkable, and not available to the field.

One more that qualifies and is easy to overlook. Better information about the same value is a genuine advantage, and it works differently from the others: it does not raise what the thing is worth to you, it narrows your dispersion.

Which is worth more than it sounds on this arithmetic. A bidder whose spread is half the field's faces half the correction, so knowing the site, the client or the ground conditions better than anyone else is a real edge and one that compounds with field size.

When To Not Bid At All

The conclusion the arithmetic points at and nobody likes. Ours.

Four observations.

If your margin cannot cover the expected maximum times your dispersion, the expected value of bidding is negative, and bidding anyway is buying revenue at a loss.

The condition is checkable before the tender. Number of bidders, your dispersion, your margin, and two of the three are known in advance.

The commercial objection is real and we would not dismiss it. A firm that stops bidding on crowded work has less work, and utilisation has its own arithmetic.

But the choice is being made either way. A firm that bids everything is choosing to lose money on the crowded jobs, and choosing it without the calculation rather than with it.

Two honest complications, because the walk-away rule is easier to state than to follow. Fixed overhead does not care whether a job is profitable, and a firm with idle crews may rationally take work at a loss rather than at nothing.

That is a real argument and it has a boundary. It justifies taking work above marginal cost when the alternative is genuinely idle capacity, and it does not justify a bidding policy, because a firm that runs this way permanently has priced its whole book at marginal cost.

And When You Are Running The Auction

The mirror image, which most treatments of this omit. Ours.

Four observations.

Everything above is a reason for a buyer to be cautious, which makes it a reason for a seller to invite more bidders.

If you are selling a business or awarding a contract, the arithmetic runs in your favour. More bidders means the winning bid is drawn from a higher order statistic, and that is money in your pocket.

There is a limit and it is worth knowing. Sophisticated bidders shade for exactly this, so adding bidders who understand the effect adds less than adding bidders who do not.

And the ethical line is where it usually is for this publication. Running a competitive process is legitimate; concealing information that would reduce the dispersion is a different act, and the encyclopedia's own framing of the mechanism as adverse selection names what is being exploited.

One practical version of that line, ours. Publishing the information you have narrows every bidder's estimate, which reduces your take on this arithmetic and produces bids you are more likely to actually receive at the price quoted.

Which is the same trade a buyer of construction faces from the other side. A tender that extracts the lowest possible number has selected for the supplier most likely to overrun, and the money saved at award is frequently spent later on variations.

Two Typos In One Literature

The bibliographic entry, since this series keeps a count.

One reference list carried on a peer-reviewed article gives the 1971 paper's pages as "64–653" where every other source gives 641–653[3], a dropped digit.

And a journal abstract reports support for the "winner's course hypothesis" where it plainly means curse[4].

Three observations, ours.

Neither changes anything, and the second is in an abstract on a publisher's site rather than in a third-party reference list, which is a slightly worse place for it to be.

The first would defeat a page-range lookup and is the more consequential of the two for anyone trying to find the paper.

That brings the running count of bibliographic variants across this series to twenty-seven.

Two of these arrived in one article, which is worth a note. This is the first time we have found a typo in a published abstract on a publisher's own site rather than in somebody's reference list, and an abstract is the part of a paper most people read.

It also illustrates why we keep the count at all. Neither error would mislead anyone about the finding, and both would defeat a literal search, which is the practical cost of an error that changes no meaning.

What To Do

Find out how many are bidding. On our own arithmetic the required shading rises from 8.5 percent against two bidders to 28.0 against twenty, at a 15 percent dispersion.

Shade more when the field is larger, which is the opposite of what competitive pressure suggests, because winning against more people is stronger evidence you were the optimist.

Measure your own dispersion. Two estimators, the same job, several times, and the spread is the number that goes into the condition.

Check the condition before bidding. Your margin must exceed the expected maximum times your dispersion, which at a 20 percent margin against ten bidders permits 13.0 percent and no more.

Work on consistency rather than accuracy. On this arithmetic, reducing the spread between your estimators is worth more than improving their average, which is counterintuitive and follows directly.

Treat a wide margin of victory as bad news. Winning a crowded tender by a lot is evidence something in your estimate was wrong.

Name your specific advantage in dollars or accept that you have none. The defence requires a reason the other bidders could not have, and a reason everyone has is not one.

Keep realised margin against bid margin by number of bidders. That record would reveal the effect in your own business within a couple of years.

The Limits Of This Analysis

Several caveats matter, and the sourcing one is unusually large. This article discusses research on bidding and is not bidding, valuation or transaction advice; the applications are our own reasoning and untested. Everything is verified to August 2026. We obtained no abstract and no result from any of the papers this article names, with a single exception: every claim about the 1971 origin, the 1983 experiment, the 1996 construction paper and the 1984 sceptical paper reaches us as a title, a citation, or an encyclopedia's summary, and titles are not findings. The clearest statement of the mechanism comes from an encyclopedia entry, not an academic source, flagged at every use. Our reading of what the construction industry's correction consists of is reasoning from a title and nothing more, and the paper may describe something entirely different. The one exception is a journal abstract giving the takeover hypothesis, its three predictions and a claim of empirical support, and we did not obtain that paper's results either. All arithmetic is ours. The expected-maximum figures are exact and we verified them against the closed forms for two and three bidders, but they assume estimate errors are normally distributed, which no source states and which is the assumption most likely to be wrong; a different distribution changes the magnitudes while leaving the direction intact. The dispersions, margins and contract values are entirely invented. The arithmetic also assumes every bidder bids their own estimate without adjustment, which is the behaviour the whole article recommends against, so the figures describe the uncorrected case and overstate the loss for anyone already shading.

Related Reading

No. 82 What A Failure Actually Tells You A 2020 meta-analysis of 55 studies found the victim derogation effect confounded with year of publication. And on our own arithmetic, one in eight... Behavioural Finance The Bias That Won The other case where an optimiser concentrates on its own largest errors. Behavioural Finance What Free Costs The Person Taking It Another case where the number on the page is not the number that matters.

Frequently Asked Questions

What is the winner's curse?

When several bidders estimate the same unknown value, winning selects for whoever estimated highest. So the winner's estimate is the maximum of the estimates, which exceeds the truth. An encyclopedia entry puts it as: you win, you lose money, and you curse.

Does someone have to be irrational for this to happen?

No, and that is what makes it unusual. The mechanism operates even if every estimate is unbiased. Nobody is overconfident and the winner still overpays, because winning is a filter that selects for optimism about the value rather than a neutral event.

How big is the effect?

On our own arithmetic, the winner overestimates by 0.564 standard deviations against two bidders, 1.163 against five and 1.868 against twenty. At a 15 percent dispersion those are overestimates of 8.5, 17.4 and 28.0 percent. The figures are exact given a normality assumption that is ours.

Should I bid harder when more people are bidding?

The arithmetic says the opposite. More competitors means winning is stronger evidence that you were the optimist, so the correction has to grow with the field. Competitive pressure pushes bids up while correct reasoning pushes them down, and they point opposite ways at the same moment.

When is a tender survivable?

When your margin exceeds the expected maximum times your dispersion. At a 20 percent margin against ten bidders that permits a dispersion of 13.0 percent and no more. Both terms are measurable: have two estimators price the same job independently and look at the spread.

Can anything be done about it?

A 1996 paper's title asserts that the commercial construction industry corrects for it. We could not obtain the paper and cannot tell you how. Reasoning from the title alone, the plausible corrections all reduce either dispersion or exposure to field size, which are the two terms in the condition.

Is paying up ever justified?

Yes, where the thing is genuinely worth more to you than to others, because then it is not a pure common-value auction. The test is naming the specific reason in dollars before bidding, and checking whether it would survive someone asking why the other bidders could not do the same. A reason everyone has is not a reason.

References

1. Encyclopedia entry on the winner's curse, recording that the story begins in 1971 when Edward Capen, Robert Clapp and William Campbell, three petroleum engineers, wrote an article claiming that oil companies suffered unexpectedly low returns year after year in early Outer Continental Shelf oil lease auctions; that such auctions are common value auctions where the value of the oil in the ground is essentially the same to all bidders, with bidders holding their own estimate of the unknown value at the time they bid; that even if these estimates are unbiased, bidders must account for the informational content inherent in winning the auction, since the winner's estimate of the common value is one of the highest estimates; that if bidders ignore this adverse selection effect it will result in below normal or even negative profits; and that the systematic failure to account for it is referred to as the winner's curse, glossed as: you win, you lose money, and you curse. The entry carries a reference list confirming Capen, Clapp and Campbell (1971), Journal of Petroleum Technology, 23, 641–653; Bazerman and Samuelson (1983), Journal of Conflict Resolution, 27(4), 618–634; Kagel and Levin (1986), American Economic Review, 76(5), 894–920; and Kagel and Levin (2002), Common Value Auctions and the Winner's Curse, Princeton University Press. Note: an encyclopedia, not an academic source, flagged at every use. Our source for the clearest statement of the mechanism and the only narrative account we obtained; we did not obtain any of the works it cites. encyclopedia.com
2. Reference list carried on a paper about auctions in the corporate takeover market, confirming Capen, E., Clapp, R., and Campbell, W. (1971), Competitive bidding in high-risk situations, Journal of Petroleum Technology, 23, 641–653; Bazerman, M., and Samuelson, W. (1983), I won the auction but don't want the prize, Journal of Conflict Resolution, 27, 618–634; Cox, J., and Isaac, M. (1984), In search of the winner's curse, Economic Inquiry, 22, 579–592; and Dyer, D., and Kagel, J. (1996), Bidding in common value auctions: how the commercial construction industry corrects for the winner's curse, Management Science, 42, 1463–1475. Together with a further reference list carried on a Journal of Finance article confirming the same 1971 and 1983 citations with month detail, and adding Gilley, O., Karels, G., and Leone, R. (1986), Uncertainty, Experience and the 'Winner's Curse' in OCS Lease Bidding, Management Science, 32, 673–682. Note: reference lists carried on peer-reviewed articles; citations only. These establish that the papers exist and where, and nothing whatever about what they found. The 1996 construction paper is the one we would most want to read and did not obtain. onlinelibrary.wiley.com
3. Two further independent reference lists, one carried on a book chapter and one on an economics and management strategy article, both confirming Capen, E. C., Clapp, R. V., and Campbell, W. M. (1971), Journal of Petroleum Technology, 23, and Bazerman, M. H. and Samuelson, W. F. (1983), Journal of Conflict Resolution, 27, 618–634; together with a fifth reference list carried on an academic preprint confirming Capen, Clapp and Campbell (1971) as Journal of Petroleum Technology, 23(06), 641–653. Note: reference lists; citations only. Recorded also as the source of a bibliographic variant: one of these gives the 1971 paper's page range as 64–653, a dropped digit, where every other source gives 641–653. onlinelibrary.wiley.com
4. Publisher record for a journal article on the winner's curse hypothesis and corporate takeovers, reproducing its abstract: that the hypothesis states that in any bidding situation a party which unknowingly overestimates the value of a given object tends to bid higher than its competitors and is therefore more likely to win it; that in a takeover the magnitude of the winner's curse is defined as the difference between the bid premium of the winning bidder and the maximum offerable premium conditional on the capital market's estimate of expected takeover gains; that the magnitude is predicted to increase with an increase in the divergence of opinion amongst acquirers with respect to the size of takeover gains, with an increase in the degree of competition for control of the target firm, and with an increase in the pre-acquisition profitability of the winning bidder; and that the empirical results provide support for the hypothesis. Note: the publisher's record, and the only source in this article supplying any empirical claim from a paper rather than a title. We obtained the abstract and not the paper, so no results are reported here. Recorded also as a bibliographic variant: the abstract as published reports support for the "winner's course hypothesis" where it plainly means curse. onlinelibrary.wiley.com

This article discusses research on bidding and is not bidding, valuation or transaction advice. No abstract or result was obtained from any underlying paper except one, whose abstract is quoted; every other empirical claim reaches this article as a title, a citation or an encyclopedia's summary. All arithmetic is the authors' own; the expected-maximum figures are exact under a normality assumption that is the authors' own and that no source states, and every commercial parameter is invented.