Empire State Bidding: RTB Auctions Research 2.0
Historically, programmatic has rested on a second-price auction model, where the second highest bid determines the price the winner pays (one cent more than the second highest bidder offers). As a result, buyers usually paid far less than what they were bidding. The gap between the first and second highest bid enticed some exchanges and SSPs to make up for the missed profit. The strategic-minded ones secretly resorted to tricks like phantom bidders (to create the appearance of competition) or soft price floors, where they would raise the floor without buyers’ knowledge. This created the climate of mistrust in the industry — the issue so much publicized about.
The situation got even worse with the rise of header bidding — a complex, multi-tier auction model where several SSPs bid on the same inventory simultaneously. The header bidding setup implies the bid that wins in an SSP’s second-price auction is to compete in another auction, a first-price one, in the publisher’s ad server. So if you’ve won in the first round of auction, but your second bid passed from the SSP to the publisher’s ad server is not the highest across all SSPs, you will lose the final auction, in other words, your win rate will drop.
What happens to the law-abiding SSPs that hold fair second-price auctions in the header bidding environment? They find themselves with low win rates, which means low revenue. So they face a choice — to miss out on valuable impressions and lose money, or to hustle to survive.
Survival of the swiftest
To ensure higher revenue for themselves, SSPs either have to cheat or move to the first-price auction model, where the winning bid is the amount paid. Moving to first-price auctioning, advocates say, is the route to transparency and a win-win game for all the parties involved: advertisers pay what they’re ready to pay, publishers get a better price for their inventory, the marketplace becomes more reliable. Yet, things aren’t always what they seem.
Some major SSPs are playing fair and square. For example, OpenX informs DSPs on the final auction type in the bid request. Rubicon Project automatically suggests the most appropriate auction dynamic for each impression. Index Exchange also notifies DSPs about its shift to first-price.
But some SSPs mislabel the type of auction they run, leaving advertisers in the dark to play the guessing game. As a result, programmatic buyers can’t design a proper strategy to optimize their media spend. If buyers think they participate in a second-price auction while it isn’t, they overpay. On the other hand, if they think they are playing according to the first-price rules while in fact it’s a second-price auction, buyers bid too low and miss out on the inventory they want.
The issue has been highlighted in a recent survey. More than 60% of marketers worldwide prioritize understanding auction pricing in 2018, and about a quarter of respondents identify it as a major priority. So the present RTB landscape is well worth investigating.
Bad guys and good guys in profile
It’s been more than six months since our first research into auction dynamics. Last October we witnessed a prevalence of second-price auctions with anomalies (52%) over fair second-price (35%) auctions across 39 SSPs from Getintent’s inventory. The share of first-price auctions was 13%.
Now that on media there are even more theoretical debates over auction pricing, it’s interesting to know how actually the situation has changed in the programmatic market.
This time we have examined the auctions run on the same 39 US-based SSPs in March, 2018. Again, for visualization purpose, we plotted normalized histograms that illustrate the win/bid prices ratio. This helps to analyze the nature of the bidding logic and spot deviations.
How do we tell what type of auction this or that SSP applies? A fair second-price auction on the Getintent platform is usually characterized by a more or less balanced distribution of the bid groups on the win CPM/bid CPM axis. This looks pretty much like Chicago skyline — dense buildings, with spikes of skyscrapers now and then.
In a case where the win CPM is very close to the bid CPM, we assume it’s either a first-price auction (if the SSP has stated so in a special field on the OpenRTB protocol) or a second-price auction with an anomaly (if the auction type is declared as second-price). The profile of these two resemble a timeline that roughly depicts how the city of Dubai was built: a continuous desert with a lonely Burj Khalifa tower on the far-right.
Here we’ve spotted a cohort of SSPs with a weird distribution where all Getintent win CPMs lie within ratio 0.9–1 or 0.67–0.97:
Another example, not as impressive as the cases above, but still a very suspicious distribution:
To put it all together, here’s what the split of RTB auctions across 39 SSPs by pricing looks like in March:
The share of SSPs conducting first-price auctions has increased by 7% since October 2017, which might seem not that significant. But you’ll see the first-price auctioning is a trend indeed if you look at the auction split at the impression level.
Our analysis of more than 336 billion US impressions over a study period in March has shown the share of the impressions available via a first-price auction is over 43%:
Compare it with the picture for December 2017 — the share of the impressions available via a first-price auction hardly made 6%. The growth of the first-price share over the past four months is dramatic!
We can suggest that many of those SSPs that offered inventory via a second-price auction in December have fairly switched to the first-price, which resulted in an increased share of the first-price inventory in March. Meanwhile, the share of the impressions offered in auctions with anomalies has not changed significantly over this period.
The following diagrams show auction type dynamics across a few of Getintent’s key SSPs. All of them have increased the share of first-price auctions in their traffic, and we can expect the trend to pick up the pace in the future.
We can suggest that the rise of first-price auctions is driven by steady growth of header bidding adoption. With the number of publishers embedding header bidding tags growing on a monthly basis, the share of SSPs that have at least partially moved to first-price increases too. Yet the hypothesis is hard to check since the OpenRTB protocol does not specify whether the auction takes place in the publisher’s header or it’s a good old waterfall.
Expert note
“First-price auctions indeed can bring more transparency for DSPs, provided the auction type is clearly communicated to buyers,” — says Kirill Solokhov, Data Team Lead at Getintent, who performed the analysis. — “However, we can’t say it’s a step forward for the industry. We shouldn’t dismiss the auction dynamics driven by common sense and business interest — buyers’ bidding strategy won’t remain the same”.
“It’s natural that DSPs want to win the auction at a lowest bid,” — he explains. “Sooner or later buyers might get a feeling they are paying more than necessary and start testing lower bids, which means practicing bid shading. Such logic debunks the wide-spread argument that publishers benefit from first-price auctions. Publishers may make more money from first-price auctioning, but only in the short run. In the long-term, first-price will actually lower CPM. As another side-effect, it will be harder for them to understand the true value of their inventory”.
