You now know where the curse lives — common value. This lesson opens the engine and shows how it works. The whole thing turns on one deceptively simple idea: winning is information. The event “I won this auction” is not neutral. It’s a filter that tells you something specific and unwelcome about your own estimate. Learn to read that signal and the curse stops being a mystery and becomes a number you can subtract.
Before you read — take a guess
Before we start — take a guess. Your estimate of an oil field's value is unbiased: on average, across all fields, it's neither too high nor too low. You bid your estimate and WIN. Conditional on winning, is your estimate still unbiased?
Winning is a filter, not a coin flip
Start with a single bidder in isolation. Your geologists estimate an oil field at $100M. Their method is honest — over many fields, their estimates are unbiased, landing above the truth exactly as often as below it, scattered symmetrically around the real number. If someone picked a field at random and asked “is your estimate too high or too low here?” the answer would genuinely be a coin flip.
Now put you in an auction with nine rivals, all using similarly honest methods, all estimating the same field. Everyone bids their estimate. You win — meaning your $100M was the highest of the ten estimates. Ask the question again: given that you won, is your estimate too high or too low?
It’s no longer a coin flip. You won because your estimate topped everyone else’s, and the way to have the highest estimate of ten is to have drawn a big positive error. Losing bidders were disproportionately the ones who estimated low (closer to, or below, the truth); winning bidders are disproportionately the ones who estimated high. The auction hands the prize to the optimist. So conditional on winning, your estimate is biased upward, even though it was unbiased before the auction started. The winning bid isn’t a readout of the field’s value — it’s a readout of the largest error in the room.
E[value | you won] < E[value]
Here’s the model in one line of notation. Your estimate is unbiased, so
E[value] = your estimate. But the estimate conditional on winning is inflated:
E[value | you won] < your estimate. The gap between them is the winner’s curse. Bid
your raw estimate and you’re paying E[value] for something whose value, given that
you won, is really E[value | you won] — systematically less. You lose the
difference on average, every time you win.
A worked example with real numbers
Let’s put numbers on it so the effect isn’t just a slogan. A field is truly worth $100M. Each bidder’s estimate is the truth plus a random error, and to keep the arithmetic clean say each error is equally likely to be anywhere from −$30M to +$30M (uniform). So every individual estimate is unbiased: its average is exactly $100M.
Now vary how many bidders show up, and track the expected highest estimate — the one that wins.
| Bidders (N) | Expected highest estimate | Overshoot above the $100M truth |
|---|---|---|
| 1 | $100M | $0M |
| 2 | $110M | +$10M |
| 4 | $118M | +$18M |
| 6 | $121M | +$26M… |
| 10 | $125M | +$25M |
| 20 | $129M | +$29M |
(For a uniform −a to +a error and N bidders, the expected maximum sits about
a·(N−1)/(N+1) above the truth — with a = $30M that’s $10M at N=2, $18M at N=4,
$25M at N=10, nearing the full $30M ceiling as N grows.)
Read down that column and the model jumps out. With one bidder, no selection, zero overshoot — your unbiased estimate is just unbiased. With two, the winner already overshoots by $10M on average. By ten bidders, the winning estimate is $125M for a $100M field — a 25% overpayment baked in purely by the selection, before anyone has done anything foolish. Every bidder was honest; every estimate was unbiased; and the auction still manufactured a systematic $25M overpayment out of thin air, just by crowning the maximum.
And notice the direction of the effect: more competition makes it worse. In almost every part of life, more options and more scrutiny help the buyer. Here, each extra rival stretches the maximum further into the optimistic tail, so the winner’s overshoot grows. Twenty geologists guessing means the winner beat nineteen others — their guess had to be an even wilder overestimate to come out on top.
In the worked example, going from 2 bidders to 10 bidders raised the winner's expected overshoot from $10M to $25M. What does this imply about competition in common-value auctions?
Drive it yourself
Now feel the mechanism instead of just reading it. Run ×50 at a few bidders, note the average overpayment, then drag the bidder slider up and run ×50 again. Do the same with the noise slider. Both dials push the overpayment the same way — up.
Winner's-curse lab
Watch the overpayment grow with rivals and noise
Every bidder guesses the same hidden true value, then bids. The highest guess wins and pays its own bid. Run auctions and watch the average overpayment — winning means you were the most optimistic.
Last auction
Press “Run one auction” to hold an auction. Then run ×50 to see the averages settle.
Averages so far
- Auctions run
- 0
- Avg winning estimate over truth
- $0.0
- Avg overpayment (bid − value)
- $0.0
- Avg winner's profit
- $0.0
Bidding rule
Two dials, one direction. More bidders and more noise both stretch the gap between the winning estimate and the truth. That’s worth sitting with, because it inverts two instincts at once: that competition protects the buyer (it doesn’t, in common value) and that you should bid boldly when the value is very uncertain (you should bid more cautiously — high noise is exactly when the curse is deepest).
The tie to regression to the mean
If you’ve met regression to the mean, the winner’s curse should feel like an old friend in a new costume — because it is one. Regression to the mean says: an extreme measurement is extreme partly because of genuine signal and partly because of lucky noise, and the noise part won’t repeat, so the next measurement drifts back toward average. The tallest fathers have shorter sons; the best first-quarter fund cools off; the rookie on the magazine cover slumps.
The winning estimate is exactly such an extreme measurement. It’s the maximum of many noisy guesses, so it’s high partly because the field really might be good and partly because that particular estimate caught a lucky updraft of positive error. The signal part is real; the noise part is not going to be there when reality shows up. So ex post — once the wells are drilled, the cash flows arrive, the player takes the field — the outcome regresses down toward the truth, and the deal underperforms the rosy estimate that won it. The winner’s curse is regression to the mean with a price tag attached: you paid the extreme, you receive the regressed value, and you eat the difference.
Selection, then regression, then disappointment
This is the shape of nearly every disappointing deal. First selection: the auction crowns whoever’s estimate ran highest. Then regression: that lucky high estimate falls back toward reality as the actual results come in. Then disappointment: the winner, having paid for the extreme, gets the regressed value and calls it “bad luck.” It isn’t bad luck — it’s the utterly predictable consequence of paying the maximum of a noisy sample and then watching it regress. The only real surprise is that anyone is surprised.
Which of these correctly connect the winner's curse to regression to the mean? (Select all that apply.)
When to use it
Pull out this mechanism the moment you catch yourself about to win a competitive bid on uncertain common value. The mental move is concrete:
- Condition on winning before you win. Ask: “If I win this, it’s because my estimate beat everyone else’s — so my estimate was probably near the top of the pile. How far above the truth does that put me?”
- Count the room. More rivals means a more extreme winning estimate, so more bidders should make you more cautious per signal, not less.
- Respect the noise. The more uncertain the value, the wider the tail the maximum is drawn from, and the deeper the curse. High uncertainty is a reason to shade harder, not to swing bigger.
- Expect regression. Whatever estimate won will look, in hindsight, like it regressed. Price the deal against the regressed value, not the winning one.
The trap this sets for you
The seductive misread is to treat your winning estimate as your best estimate — “my number came out on top, so it must be the sharpest read in the room.” Exactly backwards. In a common-value auction, coming out on top is evidence your number was the most optimistic, not the most accurate. The winning estimate is the one data point you should trust least, precisely because it won.
Recap
The engine of the winner’s curse is conditioning on winning. Your estimate can be
perfectly unbiased on its own, but “I won” is a filter that selects your optimistic
draws and discards your pessimistic ones, so E[value | you won] sits below your
raw estimate — and the gap is the curse. Worked out with numbers, ten honest bidders
guessing a $100M field produce a winning estimate near $125M: a 25% overpayment
manufactured by selection alone, and it grows with more bidders and more noise. It’s
regression to the mean wearing a price tag: the winning estimate is an extreme
maximum inflated by luck, which regresses down after the fact, leaving the winner to
pay the peak and collect the trough.
So the diagnosis is complete — now the cure. If winning means you overestimated by a predictable amount, the fix is to subtract that amount in advance. That’s bid shading, and it’s where we go next.