Programmatic

Auction Theory Ph.D.s Share Five Things Buyers And Sellers Should Keep In Mind For First-Price Auctions

As all online advertising exchangesincluding Google – switch to a first-price auction format, marketers and publishers can trust they will be run fairly. The auction type ensures that the exchanges, acting as the auctioneers, have no incentive to cheat.

“First-price auctions wipe out a lot of ways auctioneers would be tempted to bend the rules,” said Shengwu Li, assistant professor of economics at Harvard University. “There aren’t secret ways [for auctioneers] to boost the price anymore.”

Li co-authored award-winning academic research on the trustworthiness of first-price auctions last July along with Mohammad Akbarpour, assistant professor of economics at Stanford University. Neither does paid consulting for advertising companies.

Most auction theorists focus on how to prevent the buyer and seller from cheating, but ignore scenarios where the auctioneer could have an incentive to run an unfair auction.

Li and Akbarpour discovered during their research that that scenario had been playing out in online advertising.

“We started thinking about this from a purely theoretical perspective – ‘What would happen if bidders are worried that the auctioneer might cheat?’ To see it play out in reality was almost surreal,” Li said.

AdExchanger spoke with Li and Akbarpour about what publishers, marketers and ad tech companies should consider when participating in a first-price auction.

Buyers need to think about “value” when bid shading

Bid shading algorithms created by DSPs and exchanges have touted double-digit decreases in CPMs. That’s the wrong way to evaluate the success of bid shading, Akbarpour and Li warned.

“As a buyer, you should care more about the average price of an impression,” Li said. “That’s only looking at one side of the equation.”

Instead, buyers should consider the value they’re getting for the impressions, not just price. Otherwise, they’ll end up buying bottom-of-barrel impressions that create little benefit.

“Ultimately, winning an auction doesn’t mean getting away with paying very little,” Li said. “It means maximizing the value of the impression minus the price that you pay.”

The best bid shading algorithms will account for value, not just price.

Full bid transparency leads to the best market outcome

Buyers need more information than they currently get to optimize their bids.

“There is too little information in the current state of the industry,” Akbarpour said. “You don’t have to know everything, but you have to at least be able to figure out, for each bid you might place, how likely that bid is to win.”

Buyers are flying blind without aggregated bid distribution information. “[Supplying bid information] makes it much harder to do any kind of cheating, and much easier to shade and optimize your bids,” Akbarpour said.

Not sharing bid information is short-sighted for exchanges.

“One view might be that if you give bidders very little information, maybe you can trick them into bidding more. But mature markets don’t work like that,” Li said.

In mature markets like the stock exchange, participants receive enough information that they can optimize buying and selling to minute degrees, Li said.

“As this industry matures, bidders will benefit from and should start demanding more transparency,” Li said. “It seems like a more sustainable business model to make sure bidders know what they’re doing and know the rules of the game.” 

Publishers shouldn’t A/B test their floors

While tech companies love to A/B test, the strategy won’t work in first-price auctions.

“When you randomly alter the bid floors, and buyers aren’t aware of it, their behavior won’t adapt in the short term,” Li said.

For the same reason, constantly raising or lowering floors won’t help publisher revenue, because a buyer doesn’t immediately react.

Both A/B testing and dynamically changing floors do work in second-price auctions, so publishers will have to abandon both tactics in a first-price auction world.

How to set floors is a question that must be answered with data, Li said. “You need to look at the distribution of bids you’re getting, and on a case-by-case basis decide how high that floor should ultimately be.”

A floor price can function as an extra, aggressive bidder in the auction that prevents buyers from lowering their bids too far. But when price floors are set too high, publishers lose out on revenue if other bidders drop out of the auction.

Auctioneers should publicly announce the floor.

“Eventually, the buyer figures out what the floor is,” Li said. “There isn’t much of a point to making the floor a secret.”

Today, that’s not often the case. While the the current OpenRTB spec for programmatic auctions has a field to reveal the floor, it’s not clear how often it’s used. In a second-price auction, publishers often submit one “hard” floor to an exchange, which then sets a higher “soft” floor designed to squeeze more revenue out of a publisher’s inventory.

Publishers have also observed that in private marketplaces, buyers sometimes bid below the floor without knowing, suggesting that they don’t have floor information. Best practices for flooring will need to change in a first-price auction world. 

“Buyer collusion” remains a problem.

Both first-price auctions and second-price auctions can be compromised by “buyer collusion,” or when buyers share information with each other to lower the price they end up paying.

In a second-price auction, if a DSP submits a single bid for $10 to an exchange and ignores two of its other advertisers’ bids for $9 and $8, it can save the buyer money. “You are taking the bids you know wouldn’t win and deleting them so you are not moving up the price that the winner pays,” Li said.

Collusion in a first-price auction requires a more “intricate” approach, since deleting lower bids creates no advantage. “In addition to not passing on the lower bid, you might need to adjust downward the high bid in light of the information the lower bids reveal,” Li said.

Another form of buyer collusion happens when buyers “take turns” to win, which can also drive down prices.

“That would be a worry, that DSPs can … blunt price competition,” Li said. “In an extreme version of this, where there is only one DSP, that DSP could decide the winner on its side and submit only that bid. That prevents a real auction from even happening.”

This post was syndicated from Ad Exchanger.