In a previous blogpost, we talked about DSP’s, SSP’s and Ad exchanges and how they fit together. Many DSP’s use second price auction mechanics, but how do these auction mechanics work? You’ve probably heard the terms first price auction, second price auction and soft floor, but what are they? How do they differ from each other and how do they work? Let’s take a closer look through the eyes of Optimizer, Quality and Control Manager, Edward Gerits.
Auctions in a Programmatic World
Many DSP’s like DoubleClick Bid Manager (DBM) and AppNexus use second price auction mechanics for advertisers buying media programmatically. In a second price auction, the winner (the bidder with the highest bid) doesn’t pay his actual bid, but the second highest bid; this is also called “price reduction”. In a first price auction, by contrast, the winner ends up paying his actual bid, hence there’s no price reduction in a first price auction.
In theory, a second price auction should reveal the true value for the bidders and should induce truthful bidding, i.e. if an impression has a value for you of $1.00, you should also bid $1.00. For a rational bidder, there’s no incentive to bid more, because if you bid more, for instance $1.50, and you happen to win, you’ll end up paying the second price thanks to price reduction. If this second price is lower than $1.00, you would pay the exact same price as if you would have bid $1.00. Is the second price higher than $1.00, for instance $1.40, you would end up paying $1.40 for a $1.00 valued impression, which doesn’t sound like a good deal, right? Lowering your bid also doesn’t help, because you decrease the chance of winning, while you don’t increase the chance of paying less for the exact same impression.
In a second price auction, the worst that can happen when you win is paying the price you have bid – this happens in the case that someone else bids the same price as you. In all other cases, you’ll have a positive pay-off, paying less than the value of the impression. In a first price auction, paying the price you’ve bid is actually the best what can happen. This results in a zero pay-off, hence there’s no positive pay-off when you bid your true value in a first price auction. The only way to get a positive pay-off when a first price auction is in place is to lower your bid below the value of the impression. In other words, in a first price auction, there’s no incentive for the bidder to bid truthfully.
Soft Floors introduce First Price Auctions in the Programmatic World
It seems that many DSP’s have chosen the perfect auction mechanics to induce truthful bidding. They have also created bidding algorithms to help advertisers calculate the true value of an impression in a second price auction. However, these DSP’s themselves or connected SSP’s have also introduced yield management tools for publishers, like for instance soft floors, to increase revenue for publishers. A soft floor is a tool which disables price reduction from a certain price point set by the publisher, which in essence turns a second price auction in a first price auction. This means that if you bid below the soft floor, you end up paying the price you bid if you win the auction. We’ve just seen what happens when an advertiser keeps on bidding truthfully in a first price auction: there’s no positive pay-off. And that is exactly what’s happening when an advertiser keeps on using the readymade algorithms provided by their DSP’s! These algorithms are made to determine the true value and bid truthfully – perfect for a second price auction, but unfortunately, they’ll often operate in first price auctions.
Protect your Advertiser’s Budget in First Price Auctions
Now we know that we often operate in a first price auction and we know that bidding truthfully isn’t efficient from an advertiser point of view, the question arises – ‘What should we bid to maximize the potential pay-off for the advertiser?’ There’s no easy answer to this question: bidding too low results in reducing the chance of having the highest bid and winning the auction while bidding too high results in a minimal pay-off. To answer this question, we refer to a blogpost of Chris from Mathalicious, where he uses math to determine the price one should pay when there’s a first price auction in place. Although certainly not all assumptions hold true in programmatic bidding, it gives a clear view about the fundamental difference in bidding when the rules change from a second price auction to a first price auction. While in second price auctions the value to the bidder is the only aspect we need to know, for first price auctions it’s also important to know external factors like how many others will participate and how they’ll calculate their bids. Based on the assumptions made in the blogpost, Chris shows that in order to maximize pay-off, you should bid n/(n+1) of your value, where n is the number of opponents. So when an impression has a value for you of $1.00 and you expect that besides yourself only 1 other opponent will bid, your optimal bid should be: 1/(1+1) * $1.00 = 1/2 * $1.00 = $0.50. From this example, it’s clear that the more opponents, the higher the price that you bid should become, which results in lower pay-offs for the advertiser.
Turn Theory into Practice
Knowing that the outcome of your DSP’s optimization algorithm isn’t ideal when soft floors are applied is the first step, the next step should be to act appropriately to increase your expected pay-off. Your first practical step should be to identify when and at which price point soft floors apply. This is something you can quickly identify by using log level data, by talking to your publishers or, the more labor-intensive way, by creating test campaigns. Some publishers use dynamic price floors, which for instance change every hour. This will obviously make it harder to identify a soft floor. However, once identified, you can tweak your algorithms to bid lower for inventory within a certain price range, which can be achieved by setting a lower campaign goal as needed in your DSP. By lowering your goal, your bid will be reduced below the true value, which is your strategy in these circumstances. The disadvantage of creating multiple campaigns for each publisher can be overcome in AppNexus by using AppNexus Programmable Bidder (APB), where you can simply adjust the calculated algorithmic bid by a certain percent. When you know you have in general 4 opponents for a certain domain, you can, for instance, chose to multiply the algorithmic bid by 4/(4+1) = 0.8 in order to reduce your bid and use the optimal bid strategy in a first price auction.
What about the Publishers?
From a publisher point of view, we have 2 competing forces. In a second price auction, the price paid by the winner will be reduced to the second price which means the publisher earns less than the buyer’s initial bid. In a first price auction, the buyer’s will bid lower than their true value, which also means that the publisher will earn less than the buyer’s value. When one compares the theory of a first price and a second price action the result is that the expected pay-off from a publisher point of view is the same for both auction types! Setting a soft floor is only beneficial for a publisher as long as advertisers don’t change their bidding strategy as outlined in this article.
What have we learned?
A soft floor set by a publisher changes the auction type from a second price auction to a first price auction. Remember that in a first price auction it is not wise to bid your true value, instead, you should bid lower. How much lower you should bid, depends largely on the number of expected competitors. The more competitors compete in an auction, the closer your bid should be to the real value of the impression.
It’s best to adapt your bidding strategy when rules change from a second price auction to a first price auction when soft floors apply. In the end, you want to maximize the potential pay-off for your advertiser, right?
https://www.bannerconnect.net/wp-content/uploads/2018/07/MvM-rising-star-2.png400400Deesi Lihttps://bannerconnect.net/wp-content/uploads/2015/09/logo_bannerconnect_72.pngDeesi Li2018-07-03 10:01:362018-07-25 16:37:04Programmatic Talent Rising Star Mart van Melick
https://www.bannerconnect.net/wp-content/uploads/2016/08/Alice-in-Dataland-02.jpg400400Alice Bezetthttps://bannerconnect.net/wp-content/uploads/2015/09/logo_bannerconnect_72.pngAlice Bezett2018-04-03 10:00:372018-05-16 17:11:29Regression Models for Trait Prediction Part II