Winning Can Be Bad News

The cruelest sentence in auction theory is:

congratulations, you won

In a private-value auction, that may simply mean you valued the object more than everyone else did. A painting fits this story well enough. If it will hang on your wall, your willingness to pay can be personal.

In a common-value auction, the object is worth roughly the same thing to everyone, but nobody sees that value perfectly. Oil leases, mortgage pools, takeover targets, ad inventory, distressed debt, cloud capacity, and crowded alpha signals all have this flavor. Each bidder sees a noisy signal about one shared value.

Now winning means something sharper. It means your bid beat the field. If bids are increasing in signals, that means your signal was one of the most optimistic signals in the room.

Winning is information about your error.

Do Not Export Vickrey Too Far

Vickrey’s second-price auction is one of the most beautiful mechanisms in economics.¹ With independent private values, bidding your true value is a dominant strategy. If you win, you pay the second-highest bid, not your own. Raising your bid cannot lower the price you pay, and lowering your bid can only make you lose objects you would have wanted.

That theorem is real. It is also easy to overexport.

The word “value” is doing a lot of work. In the private-value model, my value is my payoff from owning the object. If I win at a price below that value, I am happy. Your value is different from mine.

In a common-value model, my “value estimate” is not my payoff. It is a noisy measurement of a payoff that everyone shares. If I bid that measurement, I am not bidding my value. I am bidding an estimate that has already survived a selection process.

Wilson’s early common-value model made the strategic problem explicit: bidders must shade bids because the act of winning says their information was favorable.² Milgrom and Weber later gave the classic affiliated-values framework, where bidders’ signals are statistically linked and auction formats can reveal or conceal useful information.³

The small but dangerous mistake is to say:

second price means truthful bidding

The better sentence is:

second price makes truthful bidding dominant under independent private values

Those extra words are the guardrail.

A Small Auction House

Here is a synthetic auction house.

Each auction sells one object with true value

\[V \sim N(100, \tau^2).\]

Each bidder $i$ observes

\[S_i = V + \epsilon_i,\qquad \epsilon_i \sim N(0, \sigma^2).\]

The auction is sealed-bid second price with an optional reserve. The lab compares three deliberately simple bidding rules:

signal bids $S_i$ directly;
posterior bids $E[V \mid S_i]$, shrinking the signal toward 100;
winner-aware subtracts a crude order-statistic adjustment for the fact that the winner’s signal is selected from the high tail.

The third rule is not an equilibrium solver. It is a didactic correction. The point is to make the conditioning visible: your estimate should change after you learn that your estimate won.

Bidders: 7 Auctions: 2.4k Value volatility: $20.0 Signal noise: $25.0 Winner adjustment: 100.0% Reserve: $0.00

Signal bid Posterior bid Winner-aware bid Winner signal bias

Deterministic synthetic common-value auctions. In each counterfactual market, every bidder uses the same rule. The winner-aware rule is a transparent posterior-minus-selection adjustment, not a claim about equilibrium bidding.

At the default setting, the signal bidder wins often and loses money. That is not because the auctioneer is cheating. It is because the highest signal is usually too high.

Increase Bidders. The signal bidder’s winner overestimate rises. With more draws from the same noisy distribution, the maximum noise term becomes more extreme. The price is not just a market-clearing number; it is an order statistic of other people’s errors.

Lower Signal noise. The curse almost disappears. When signals are precise, winning tells you much less about being lucky. The winner still has the high signal, but the signal does not contain much noise to select on.

Set Winner adjustment to zero. The winner-aware rule collapses to the plain posterior rule. That rule knows the signal is noisy, but it does not know that the winning signal is special.

Winning Changes the Posterior

The whole idea fits in one inequality:

\[E[V \mid S_i=s,\ i\text{ wins}] < E[V \mid S_i=s].\]

The right side says: “I observed signal $s$.”

The left side says: “I observed signal $s$ and also learned that every rival’s bid was lower.”

In a common-value world, those are different events. If everyone’s signal is noisy evidence about the same hidden value, then rivals’ lower signals are bad news about the object. Winning reveals that bad news indirectly.

The Bayesian posterior from one Gaussian signal is simple. If the prior variance is $\tau^2$ and the signal noise variance is $\sigma^2$, then

\[E[V \mid S=s] = 100 + \frac{\tau^2}{\tau^2 + \sigma^2}(s - 100).\]

This is just shrinkage. A signal of 140 does not mean the object is worth 140 if signals are noisy. Some of that number is value, and some is measurement error.

The winner’s curse is the next layer. The winning bidder’s signal is not a typical signal. It is the top signal among bidders who were all sampling the same object. Even after shrinkage, the event “this signal won” carries additional information.

This is why the curse gets worse with bidder count. The expected maximum of independent noise grows with the number of bidders. Competition does not merely raise prices. It selects more aggressively for optimism.

The Lab Results Hurt

The winner’s curse became famous partly because people kept falling into it.

Kagel and Levin’s laboratory common-value auctions found that bidders often bid too aggressively and earned negative profits, especially as the number of bidders increased.⁴ The result is uncomfortable because the setup is clean. Participants are not buying oil fields with geological ambiguity, political risk, and debt covenants. They are playing a controlled auction with a well-defined value process.

The lesson is not that humans are uniquely foolish. The lesson is that the conditioning is cognitively unnatural. We are good at thinking:

my signal is high, so the object is likely good

We are worse at thinking:

my signal is high enough to beat other signals, so part of its height is likely luck

Markets institutionalize this mistake whenever a process rewards the most optimistic estimate and then treats the winning estimate as if it were unbiased.

Finance Keeps Wearing Auction Costumes

Many finance problems are common-value auctions with different stationery.

An acquisition auction is the obvious case. The target has one uncertain value. Each bidder has a model. The winner is likely to be the bidder whose synergy estimate, revenue forecast, cost-cutting plan, or financing assumption was most optimistic. Paying the second-highest bid does not remove the problem if the second-highest bid is also conditioned on a noisy common value.

A distressed-debt process has the same smell. Everyone is underwriting recovery value, legal timing, liquidity, and macro outcomes. The buyer who wins may be the buyer whose model was least bothered by the ugly tail.

Quant research has a cousin of the same problem. A team tests many signals, and the one that survives is the one with the highest estimated Sharpe. The strategy did not “bid” in an auction, but selection chose the most optimistic estimate. The backtest winner is an order statistic too.

Even cloud spot markets and ad auctions can inherit common-value components. The private value of a unit may differ across buyers, but shared uncertainty about future demand, conversion quality, or resale value creates a common-value layer. A mechanism can be truthful for private values and still leave bidders with a forecasting problem about the common part.

The practical question is not “is this auction private value or common value?” Most real settings are mixed. The question is:

how much of my bid is my idiosyncratic value, and how much is a noisy estimate of a shared value?

The second part deserves a winner’s-curse discount.

Revenue Is Not Buyer Wisdom

From the seller’s point of view, the winner’s curse is not automatically bad. Competition and noisy information can raise expected payments. Auction design is partly about extracting surplus, partly about allocating the object efficiently, and partly about making participation robust enough that informed bidders show up.

Myerson’s optimal auction theory shows how reserve prices and allocation rules can maximize expected revenue under private values using virtual valuations.⁵ Riley and Samuelson developed related optimal-auction results around reserve pricing and entry.⁶ Klemperer’s surveys emphasize that real auction design also turns on entry, collusion, information, and the risk that a format deters precisely the bidders the seller needs.⁷

Common values add another design tension. If bidders fear severe adverse selection, they shade aggressively or stay home. Revealing more public information can reduce the winner’s curse, but it may also change revenue and participation. Milgrom and Weber’s linkage principle formalizes one version of this idea: when signals are affiliated, committing to reveal information can raise expected revenue in important auction environments.³

So the seller’s dream is not simply “make everyone bid naively.” That can work once. A market that repeatedly punishes winners may train bidders to disappear, collude, or demand wider discounts.

Questions Before Raising the Paddle

Before bidding on a noisy common-value object, I want a separate memo answering:

What is the shared-value component of the asset?
Which parts of our model are private synergies and which parts are common forecasts?
How many serious bidders are in the process?
What information will we learn if we win?
What information would we infer if we lose narrowly?
How correlated are bidders’ information sources?
Does the auction reveal rival signals, interim prices, reserve levels, or only the final allocation?
What bid would be optimal under our standalone posterior?
What discount is required after conditioning on winning?
How sensitive is the bid to signal noise, bidder count, and reserve price?

The most important line is the fourth one. A bid committee should explicitly ask what winning would teach it. If the answer is “nothing,” the committee is silently assuming private values.

Sometimes that assumption is fine. Sometimes it is expensive.

The Sentence to Keep

The winner’s curse is not a proverb about overpaying.

It is a selection-bias theorem with a price tag. When values are common and signals are noisy, the winner is selected partly for favorable error. The higher the noise and the more intense the competition, the more suspicious the winning signal should become.

This is why “we won the auction” should not end the analysis. It should start a post-mortem:

what did winning reveal about the estimate that made us win?

If the answer lowers the value, that is not pessimism. It is Bayes being polite.