13
Part III · Playing with people

Going once, going twice

One rare trading card, four bidders, and everyone's real price is a secret. Change how the auction is run and the right way to bid changes with it. In one kind of auction the unbeatable plan is simply to tell the truth, and by the end you will be able to prove it.

A rare trading card is up for sale. Four bidders want it: you, Robo, Robo 2, and Robo 3. Every bidder has a private value: the most they would pay for the card and still be glad they bought it. Your value is shown to you. The robots' values are secret. Each auction, every robot gets a fresh value somewhere between $10 and $100, every amount equally likely.

Winning is not the goal. Profit is. If you win at price p, your profit is your value minus p. If you do not win, your profit is 0. Paying $80 for a card you value at $60 is not a victory. It is a loss of $20.

The auction you probably picture is the English auction: the auctioneer calls a price, everyone willing to pay it stays in, and the price climbs by $5 until one bidder is left. That bidder pays the last price called. Play a few. Each time a price is called, stay in or drop out.

The English auction

Price climbs by $5. Last bidder standing pays the last price called.
Your value$60
Price called$5
Still in4
Auctions0
You won0
Total profit$0

Every robot follows one rule: stay in while the price is below my value, drop out the moment it is not. It is the right rule for you too. Say your value is $60 and the price called is $45. Staying in costs nothing yet, and if everyone else drops out the card is yours for $45, a $15 profit. Dropping out throws that away. Once the price called reaches $60, staying in can only win you the card at $60 or more: profit zero, or a loss. So the plan is: stay until the price passes your value. Nothing the other bidders do changes it.

Look at what the winner pays. Not their own value. They pay the price at which the second-to-last bidder gave up, which is roughly the second-highest value in the room. Hold onto that.

Sealed envelopes

Change the rules. Each bidder writes one number in an envelope. The envelopes are opened, the highest number wins, and the winner pays the number they wrote. This is a first-price sealed-bid auction.

The English auction had an easy rule. This one does not. Say your value is $60. If you write 60 and win, you pay $60 for a card worth $60 to you. Profit: zero, guaranteed. Bidding your full value can never earn you a cent. So you must bid less. But bidding less means you might lose to a robot that wrote $52 when you would gladly have paid $55. A lower bid gives you a bigger profit if you win and a smaller chance of winning. That tradeoff is the whole game.

The robots know this. Each one shades its bid: it writes down three quarters of its value. A robot that values the card at $80 writes $60. Your job: pick a bid, ten rounds in a row, and finish ahead.

First price, ten rounds

Highest bid wins and pays its own bid. Robots bid 3/4 of their value.
Round1 of 10
Your value$60
Total profit$0
Your bid $
Math corner

Expected profit. Chapter 4's tool is exactly what a sealed bid needs. Your expected profit from a bid b is

expected profit = (chance of winning) × (value − b)

The second factor is easy: bid $45 with a value of $60 and you earn $15 whenever you win. The first factor takes work. You beat one robot when its bid, 0.75 × its value, is below b, which means its value is below b ÷ 0.75. Robot values are spread evenly from 10 to 100, so the chance one robot's value lands below some number x is (x − 10) ÷ 90. Put those together:

chance one robot bids below b = (b ÷ 0.75 − 10) ÷ 90

kept between 0 and 1. The three robots draw their values separately, so the chance that all three bid below b is that number cubed. Here is the table for the value in the game above, right now ($60), with bids rounded to whole dollars:

The robots' three-quarters rule is no accident. With n bidders whose values are spread evenly from zero up to some top amount, bidding (n − 1)/n of your value is the equilibrium: the rule where no bidder gains by switching. Four bidders: 3/4. Our values start at $10 rather than $0, which nudges the exact equilibrium up by about $2.50; three quarters is close enough for a robot.

The clever auction

Here is a small change with a huge effect. Envelopes again, and the highest number still wins. But the winner pays the second-highest number, not their own. This is a second-price auction, also called a Vickrey auction (say: VICK-ree) after the economist who worked out what it does.

In a second-price auction, writing down exactly your true value is a dominant strategy, Chapter 2's word: it is at least as good as every other bid no matter what the other bidders write. No shading. No guessing what Robo will do. No tradeoff. Just tell the truth.

That is a strong claim, so test it before you believe it. Set your value V. Set R, the highest number anyone else wrote (only the highest matters, because that is the price you would pay). Then slide your own bid B around and watch your profit.

Why honesty wins

Second price: if B beats R you win and pay R. Ties go to the other bidder.

bid = valueyour bid Brival bid R
Quiz: your value is $60. What should you bid? $

Slide R anywhere. The yellow dot, your profit when you bid V, is never below any other point on the curve. Other bids sometimes tie it. None beat it. Here is the proof, in two cases.

  1. Case 1: R is below V. Bidding V wins, and you pay R, so your profit is VR, a positive number. Any other bid above R gives the exact same result: you still win, you still pay R. Any bid at or below R loses and earns 0. So no bid beats V.
  2. Case 2: R is above V. Bidding V loses: profit 0. Any other bid below R also loses: 0 again. A bid above R wins, but you pay R, which is more than the card is worth to you, so your profit is VR, a negative number. So no bid beats V.
  3. If R equals V, every bid earns 0: lose and get nothing, or win and pay exactly what it is worth. So in every case, bidding V is at least as good as anything else. That is what dominant means.

Notice the trick. Your bid decides whether you win, but never what you pay. Since your bid cannot touch your price, there is nothing to gain by lying about it. Now play it for real. The robots bid their true values; they have read the proof.

Second price, ten rounds

Highest bid wins and pays the second-highest bid. Robots bid their true values.
Round1 of 10
Your value$60
Total profit$0
Your bid $

Going down

One more. In a Dutch auction the price starts high and falls. A clock ticks down from $100, and the first bidder to shout "Mine" takes the card at whatever the clock says. Flower sellers in the Netherlands have sold their flowers this way for more than a century, which is where the name comes from.

It looks like a different game from sealed envelopes. It is not. Before the clock starts you have to decide the number you are waiting for: the price at which you will shout. Shout too early and you pay too much; wait too long and a robot beats you to it. That waiting number is your sealed bid. The first-price auction and the Dutch auction are the same game in different clothes: the highest number wins, and the winner pays their own number. The robots below wait for three quarters of their value, just as before. Try it by hand first, then type a waiting number and let the clock shout for you.

The falling clock

Price falls $1 at a time. First to shout takes it at that price.
Your value$60
$100
Total profit$0
Or wait for $
Big idea

How the auction is run changes how you should bid. When the winner pays their own number (first-price, Dutch), you have to shade below your value and gamble on the tradeoff. When the winner pays a price set by someone else (English, second-price), the right bid is simply your true value, and no cleverness can beat it.

Try it on paper

1. A second-price auction. The bids are 40, 65, and 80. Who wins, and what do they pay?

Answer

The 80 bid wins and pays the second-highest bid, 65.

2. Second-price again. Your value is 60 and you bid 60. The others bid 30 and 70. What happens? Would bidding 75 have helped?

Answer

The 70 beats you: profit 0. Bidding 75 would have won, but you would pay the second-highest bid, 70, for a card worth 60 to you: profit −10. Losing was the better result.

3. A first-price auction. Your value is 80. If you win with a bid of 60, what is your profit? If you win with a bid of 78?

Answer

80 − 60 = 20 with the low bid, 80 − 78 = 2 with the high one. The high bid wins far more often, but each win is worth only 2. Which is better depends on how often each wins: the expected-profit question from the Math corner.

Challenge

So far every bidder had a value of their own. Now suppose the card is worth exactly the same to everyone: it will sell next week for one fixed price, and nobody knows what. Each bidder makes a guess; the guesses scatter around the truth, some high, some low. Everyone bids from their own guess in a first-price auction. Strange fact: the winner usually loses money, even though nobody bids above their guess. Why?

Then test it. Put a handful of coins in a jar without counting. Four people guess the total and bid their guess, sealed; the highest bid takes the jar. Count the coins. Do five jars and keep score of the winner's profit.

Answer

Winning means your guess was the highest of four. The highest of several guesses that scatter around the truth is usually above the truth: for the top guess to be too low, all four guesses would have to be too low at once. So winning is bad news about your own guess. This is called the winner's curse. The cure: bid as if you already know you will win, so assume your guess is the high one and shade it down. Petroleum engineers gave the curse its name in 1971 after watching oil companies overpay for the right to drill in the Gulf of Mexico.

True story

William Vickrey published the second-price idea in 1961. In October 1996 he won the Nobel Prize in economics, partly for that paper, and died three days later, at 82, on his way to a conference. The idea was older than anyone knew: stamp collectors were running second-price auctions by mail in the 1890s, decades before the theory existed. eBay's automatic bidding, where you type in the most you would pay and the site bids for you, is a second-price auction in disguise. Google sells its search ads with a generalized second-price auction, and that auction runs billions of times a day.

Stars in this chapter

Earn them by doing the clever thing, not by clicking around.