Background: The Origin and Formal Nature of the Winner’s Curse

The winner’s curse is one of the most robust and counter-intuitive results in auction theory and decision science. It was first systematically documented in the early 1970s by petroleum engineers E.C. Capen, R.V. Clapp, and W.M. Campbell while studying competitive bidding for oil-drilling rights in the Gulf of Mexico. Companies that won leases frequently discovered that the tracts were less productive (or more expensive to develop) than their pre-bid estimates had suggested. The winning bid was not a random sample from the distribution of estimates; it was the most optimistic estimate in the room. 

On average, therefore, the winner had overestimated the true value of the asset.

The phenomenon is not limited to oil leases. It appears whenever three conditions hold simultaneously:

1. The object being bid on has a common value (the same true worth to every participant once uncertainty is resolved).

2. Bidders have noisy, private estimates of that common value.

3. The allocation rule awards the object to the highest bidder (or, in sequential settings, to the first person willing to claim it at a given price).

Under these conditions the act of winning is itself informative: it reveals that every other participant held a lower estimate. Conditional on winning, a bidder’s estimate is biased upward. The larger the number of bidders and the greater the variance of private signals, the more severe the bias. 

A clean laboratory demonstration was popularized by Yale economist Ben Polak. In one of his lectures he filled a pesto jar with coins (true value $2.07) and auctioned it to a room of students. Forty independent guesses centered near the true amount; the winning bid was $5.00. The student who “won” was not the best estimator—she was simply the most optimistic. The lecture is free and remains one of the clearest public explanations of the concept.Formally, let each bidder ( i ) form an estimate

Vi=V+εiV_i = V + \varepsilon_i

V_i = V + \varepsilon_i

where ( V ) is the unknown true common value and

εi\varepsilon_i\varepsilon_i

is a mean-zero error term. The winner is the bidder with the largest

εi\varepsilon_i\varepsilon_i

. Therefore

E[εi∣win]>0.E[\varepsilon_i \mid \text{win}] > 0.

E[\varepsilon_i \mid \text{win}] > 0.

Rational bidders anticipate this selection effect and shade their bids downward (“bid as if you already know you will win”). 

Failure to shade produces systematic overpayment.

The Winner’s Curse in Snake Redraft Drafts

Science of Fantasy Football In a common-value setting, every participant is estimating the same underlying quantity—the player’s true expected fantasy points this season. Estimates differ because of private information, differing risk preferences, narrative bias, and noise. The manager who selects the player does so by ranking him higher (or taking him earlier) than every other manager. That selection effect systematically biases the winning valuation upward. The average of all estimates may be unbiased; the winning estimate is not. This is the winner’s curse.

Snake drafts are sequential implicit auctions. ADP is the market’s revealed average estimate. A reach of even one or two spots is an overbid relative to the room. Because the true value is common, the manager who reaches is, on average, the one whose private signal was most optimistic—and therefore most wrong.

Where the effect is strongest

Empirical patterns from multi-year championship-roster and league-winner studies show a clear structure:

• Rounds 1–4: Lower relative dispersion. Elite players have more stable historical data and clearer roles. Projection R² is higher and ADP standard deviations are tighter. Busts occur, but the gap between the average estimate and the most optimistic one is smaller.

• Rounds 5–10 (roughly ADP 50–120 in a 12-team league, scaled for 20-round boards): Peak curse territory. Role ambiguity, committee situations, scheme changes, sophomore leaps, and injury recoveries create maximum estimate dispersion. 

• Championship data repeatedly shows a near-dead zone in the ADP 61–100 band—almost no high-share championship contributors emerge from it in recent seasons. Managers who “win” these players by taking them highest are systematically the most optimistic and, on average, the most incorrect about median outcomes.

• Rounds 12–20: Extreme raw uncertainty (high coefficient of variation), but low absolute cost. Most players already carry near-replacement expected value. Being the most optimistic simply means selecting a higher-variance lottery ticket. 

• True league-winners do appear here, yet the opportunity cost of the pick itself is small.

Player- and team-level residual uncertainty amplifies the effect. High residual variance after conditioning on Vegas implied totals, pace, and ADP is a reliable flag.

Quantifying the bias

Think of each manager’s estimate as

Vi=V+ϵiV_i = V + \epsilon_i

V_i = V + \epsilon_i

where ( V ) is true expected value and

ϵi\epsilon_i\epsilon_i

is mean-zero error. The winner is the manager with the largest

ϵi\epsilon_i\epsilon_i

. Conditional on winning,

E[ϵi∣win]>0E[\epsilon_i | \text{win}] > 0E[\epsilon_i | \text{win}] > 0

. The size of that positive bias grows with the variance of the

ϵ\epsilon\epsilon

distribution and the number of competing managers. In the middle rounds that variance is largest.

Practical countermeasures

1. Condition explicitly on winning. Before locking a pick, ask: “If I am the only manager willing to take this player here, what does that information tell me about my estimate?” Shade the valuation downward by an amount proportional to estimated dispersion (larger shade in rounds 5–10).

2. Treat ADP as a Bayesian prior. Form your ranking, then update it with the market’s average. Only deviate when you possess genuine private information (scheme fit, recovery timeline, target-share trajectory the room is missing). Reaches without private edge are pure curse exposure.

3. Enforce tier discipline. Prefer the last player in a coherent tier over the first player in the next, more uncertain tier. Letting a positional run occur often returns higher-value assets later and avoids the last-in-tier overpay.

4. Separate floor from ceiling by region. Early: prioritize high-floor difference-makers. Middle: demand a realistic path to a tier jump or avoid. Late: maximize ceiling and private information; the absolute cost of optimism is low.

5. Use external signals to shrink residual variance. Condition projections on Vegas team totals and player props. Where ADP and Vegas diverge, or where last-year performance clusters show high within-group residual variance, increase the shade.

The managers who consistently leave snake drafts with fewer overpays are those who treat every mid-round selection as if they already know they are the highest estimate in the room. 

That single mental adjustment converts the curse from a hidden tax into a controllable bias.