Despite being well researched, auctions remain quite tricky. Lots of papers have been written on the topic. William Vickrey was even awarded the Nobel Prize in Economics for his work on the theory of auctions in 1996. He proved that the second-price auction (now often called the Vickrey auction) reveals the true value of the object, because it is in the individual’s best interest to pay a fair price for the item. For this reason, the second-price auction was adopted by the adtech industry as a default auction in the OpenRTB 2.0 protocol released in January, 2012. This means, unless specified otherwise, media buyers and sellers deal with the second-price auction.
However, the theory shatters on the rocks of reality. In real life publishers face a lot of auction-related issues and ask themselves: What if my inventory is not very good and there’s only one bidder? What if the fair price for my inventory is much lower than what I want?
And here is when hard and soft floor prices come to the rescue. Unfortunately, floor pricing is not covered in the OpenRTB specification, so you need to read the fine print in the contract with an SSP. SSPs may also take advantage of dynamic floor pricing, which means in case you are the only auction participant, SSP will try to set the floor price equal or close to your bid. Tricky, isn’t it?
But floor pricing is a relatively fair way to squeeze profit out of the second-price auction. In private auctions, SSPs can resort to some really dirty tricks, until they get caught with the hand in the cookie jar and another lawsuit breaks out.
At Getintent, we have analyzed over than 1 billion impressions in the US for suspicious auction mechanics. Specifically, we looked into how often SSP reported the auction type in the OpenRTB protocol and what the win/bid price ratio was.
First, we divided all auctions into three types:
Your expectations is your problem. So take time to study the contract carefully, because you will probably find a clause that says the SSP can change the auctions rules as it wants.
Here is what we got:
Then we plotted normalized histograms that illustrate the ratio between win and bid prices and distribution of the auction groups with approximately equivalent values.
The findings show 52% of the auctions have some anomalies. Below is the list of deviation scenarios.
Have a look at these guys. The SSP claims it applies the second price auction model, but the win bids can be divided into three buckets with values 0.8, 0.9 and 1.0.
The right most bar with the ratio of the bid/win price 1.0 signals that almost 30% of all impressions in the auction are bought at the first price.
As we move further, we see other SSP guys who take it easy and just break their clearing prices into 6 almost identical buckets — 0.89, 0.90, 0.91, 0.92, 0.93, 0.94 — with each roughly equal to 16.6%. Nowhere in the galaxy can this be called a fair auction.
The case is not a one-off. Below we see an auction with 10 buckets.
It gets worse and worse as it goes on. On the following SSP, despite the fact the second price auction is declared in the bid request, the win and bid prices are equal for more than 95% impressions.
Interestingly, there is also an opposite situation. Even though the first price auction was intended, Getintent bidder won with a lower bid. This looks like a software bug at the SSP level or missconfiguration on the publishers side.
So these have been the bad guys. But what should good SSP guys look like? Here is the profile of a fair auction:
“You see the buckets are continuously distributed, with some random spikes,”— explains Kirill Solokhov, Data Team Lead at Getintent. “These spikes can be accounted for by the big campaigns running on the Getintent platform, but there is only one bucket with the value greater than 3%. Compare it with the buckets on other histograms.”
All these schemes cast a shadow on the second price auction. No wonder there has been a tendency to switch to the first price auction mechanics recently. The key programmatic exchanges and SSPs like OpenX, Index Exchange, AppNexus, Rubicon Project, and PubMatic have already admitted starting experimenting with the first price auctions.
William Vickrey would say that the revenue would fall. Meanwhile the adtech community believes letting buyers pay what they bid will bring more transparency to digital advertising.