Auction Dynamics and the Prisoner’s Dilemma

June 12, 2018 — by Travis Barnes

A hypothetical scenario:

Two criminals from the same gang are arrested and imprisoned without the ability to communicate with each other. The prosecutor lacks the evidence to convict either one of the primary crime, so is holding each one on a lesser crime, which has a one-year prison sentence. Each prisoner is asked to betray the other, in which case one will go free and the other will serve a two-year sentence.

If you recognized the classic Prisoner’s Dilemma that has existed as a principal problem in game theory since 1950, then you probably know that rational actors always betray each other in the end, no matter how many scenarios you act out, because good faith cooperation is not rewarded in situations in which two players are not transparently communicating. What the heck does this have to do with digital marketing? A lot more than we usually care to admit, it turns out.

No transparency means no trust

In the world of online advertising auctions, there is almost no direct communication between buyers and sellers—it’s mediated by an exchange. Buyers and sellers are forced to anticipate what the other will do and, just like in the Prisoner’s Dilemma, they are incentivized to dime out each other.

Historically, RTB auctions have been conducted as second- price, or “Vickrey”, auctions, in which the highest bidder pays $0.01 more than the second highest bid or any floor set by the publisher. Second-price auctions incentivize each participant to bid exactly what the impression is worth to him or her. If you would be willing to pay up to $5.00 CPM for a certain impression, you can bid $5.00 in a second-price format and be confident that you’ll never overpay.

Although almost all exchanges nominally operated as Vickrey auctions in which every participant understood the rules of engagement, they really operated more as hybrids of first- and second-price auctions, using mechanisms such as “dynamic floors” to cause the impression to clear higher than it would in a true second-price format.

Advertisers have understood that they are not seeing the full picture about where and why their bids cleared when they did and, therefore, have no reason to believe that a $5 bid that cleared at $3.01 did so because of another $3 bid or an invisible price floor.

Hence, the Prisoner’s Dilemma: since neither the buy side nor sell side knew fully what the true price of the impression should be or what an advertiser was willing to bid, both were incentivized to overcorrect—resulting in less efficient outcomes for both the buy-side and the sell-side transaction participants. Buy-side players are unable to get the best ROI for marketers on media because they are unable to optimize to the true dynamics of a given auction. Sell-side participants, meanwhile, end up blindly raising and lowering floors to try to find the right balance for selling the most inventory at the highest CPM.

First-price auctions encourage transparent cooperation

Advertisers are justifiably suspicious of first-price auctions. But if implemented between the buy side and sell side cooperatively, they can restore some degree of transparency between both parties. First-price auctions provide this transparency by removing the opportunity to dynamically adjust floors on the sell side without letting the buy side know. Some exchanges have announced plans to fully migrate towards a 100-percent first-price format, both for their own interests and to implement the only truly “transparent” auction model available. Vickrey auctions would be more transparent, but in a world without a mechanism for verifying that a true-second price auction has occurred, it is the closest thing marketers can get to price transparency, and they’ll be able to adjust their methodologies for evaluating bids accordingly.

Read more about MediaMath’s study of the trends driving exchanges towards first-price auctions and the impact it is having on advertisers in our whitepaper HERE.