Second-Price Auctions And The Potential For Gaming The System

michaelnecheporenkoThe Sell Sider” is a column written for the sell side of the digital media community.

Today’s column is written by Michael Necheporenko, chief technology officer at Roxot.

Currently, publishers are discussing emerging server-to-server header bidding solutions and latency issues but ignoring the biggest problems with website revenue: Why are publishers still relying on second-price auctions to sell online advertising? Publishers owe it to their bottom lines to re-examine this strategy.

Like many good theories, the second-price auction model loses its advantages when implemented in the real world. The idealistic concept of a fair auction where buyers bid their true private values starts to crack when the programmatic advertising environment adds new variables to the equation. These inefficient auction mechanics cause an imbalance in the market, jeopardize publishers' positions and lower their potential revenue.

The logic behind truthful bidding, the main theoretic advantage of the second-price auction model, is only true if a singular, unique item is auctioned. However, impressions auctioned in programmatic advertising are not as unique as you would think.

Ad auctions are replayed over and over again against the same visitors, websites, inventory units, ad sizes and more. It doesn't take long to analyze auctions won and lost to find a way to game the system.

For example, to uncover the actual market value for an auction aimed at a specific user type, all one needs to do is bid unrealistically high. Besides winning the auction and paying (and learning) the price the second-highest bidder was willing to pay, if done across millions of impressions you'll get a map of competitive bids for your target audience over time. Armed with the knowledge of true market value and the auction parameters, an advertiser can influence future auctions, such as bidding lower than the actual market value to drive down bids for future auctions.

In the end, advertisers adopt new bidding strategies where bidding their true private value is no longer the most profitable approach. In an attempt to buy the sum total of impressions needed at the lowest rates possible through substitution, advertisers programmatically vary bids from impression to impression and auction to auction.

When bids unpredictably and rapidly fluctuate, huge gaps between a winning bid and a second bid may occur. Inaccuracy in and discrepancies between advertisers’ private values, which are usually based on customer acquisition costs and different attribution models, increase gaps between the advertisers' bids even further. As a result, these gaps discount publisher inventory prices, lower their potential revenue and jeopardize their position in the market.

For better or worse, RTB functions on the second-price auction model with crucial limitations. Because advertisers employ bidding strategies that result in low bids and/or huge gaps between the winning and second-highest bids, the current state of the industry is proving untenable.

As it evolves, where will the industry look? Will it be to first-price auctions? One unified second-price auction with first bids submitted by all demand partners? Or something else entirely?

Follow Roxot (@roxot_team) and AdExchanger (@adexchanger) on Twitter.


  1. Jed Nahum

    Good Article. I’m always interested to hear the thoughts of experts on 2nd price auctions. I think your thesis is that they are no longer appropriate, or never were appropriate, for digital ad auctions. I’m struck that you investigate the original set of flaws WRT 2nd price that were accurate in 2009 when AdEx launched. It would be interesting, I think, to examine the most recent assault on 2nd price, which is undoubtedly header bidding. For various technical reasons, in a standard header bidding arrangement, auctions have already devolved to first price. Because the same auction technology powers both header bidding and standard RTB, I suspect that the last nail is effectively already in that coffin and that bidders are morphing to first price by default. Would you agree?

    • I don't see it that way. Actually, I think header bidding is a step towards unified second-price auctions rather than industry-wide transition to the first-price auction model. I'd say client-side header bidding have introduced a hybrid auction style. Header bidding wrappers collect clearing prices (already a result of second price auctions) from demand partners and conduct a first-price auction. This is mostly caused by technical limitations - collecting all bids to a client-side header and conducting a second-price auction there would likely result in insane latency times. The hybrid auction style, unfortunately, doesn't solve second-price auction's inefficiencies. It can prevent the highest bidder from winning an auction. The logical solution is to find a way to conduct one unified auction in header bidding. It seems like S2S header bidding solutions endeavor to accomplish that considering the news that AppNexus is planning to collect first and second prices from integrated demand partners in their S2S Header Bidding solution. How efficient will this be?

      • Michael, the "hybrid" model is not due to latency. HB is first price because of the 2-step decisioning (within HB, THEN within DFP).
        Server side will not change that, as long as there is still an ad server.
        Going DIRECT to demand source from HB will make it a true first price auction. I believe this is what will happen.

  2. Second price auction in an inefficient market indeed drives the price down compared to a fixed buy for publishers. PMPs and header bidding directly or indirectly address this gap as these mechanisms take the clearing price closer to the first price auction. However, this is alarming the buy side players on header bidding. Genearating market consensus from both sides to move from second price auction to first price auction explicitly is probably lot harder than doing so implicitly thru header bidding and PMPs 🙂


  3. GSP enables the right buyer/bidding behavior when the seller is able to see all possible bids (i.e. multi-bid from a DSP.) If DSPs each passed their top two bids and buyers knew 2nd price auctions were the reality, pubs would gain their best yield from that. As it stands, because we (Goodway) know games are being played with auction mechanics, we've constructed ways to avoid over-paying. Because our bidding behavior now suppresses our bid prices, publishers will struggle to get prices they could have gotten from us before when we trusted that auctions were GSP.


Add a comment

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>