There has been some debate recently on AdExchanger about the benefits of first price versus second price auctions. Esco Strong, the Director of Display Marketplace Strategy for Microsoft Advertising wrote an article that basically said second price auctions didn’t work well for single unit RTB auctions and we should get rid of them:
Comparatively, first-price auctions are competitions where there is no reduction in clearing price for the auction winner; instead, the winner simply acquires the good they have won by paying the price of their bid. The dynamics of this type of marketplace would become much more straightforward and predictable, enabling more parties to participate and experience stable results, as well as manage their businesses to a of set expectations that won’t require constant revision.
While I am and remain a fan of Esco, and his piece was elegantly argued, I strongly disagree with it. As I see it, there are two options in building a long-term business: by pricing transparently, or by taking unfair advantage of your customers. Only the first seems sustainable to me.
He then went on to touch on some topics relating to first and second price auction mechanisms with some input from the Business Intelligence group at Criteo. The problem with Esco Strong’s original article is that it shows a strong lack of understanding on some critical basics of auction theory and mechanism design. Things like the Bayes-Nash Equilibrium, Revenue Equivalence Theorem, and construction of optimal floor prices were not really mentioned for some reason. So, what are you to do if you want to design an auction mechanism to participate in the RTB space? I’m glad you asked. The short answer is to use a sealed bid, second price auction with a revenue-optimal reserve price. This kind of auction is nothing new and has been around since Myerson’s crucial 1981 paper Optimal Auction Design.
Let’s say you want to sell something and you need to know which auction mechanism will provide you with the move revenue. There is a result in Auction Theory dating back to Vickery in 1961 (that Myerson generalized) that says if the auction has the following criteria:
The bidder with the highest signal/valuation/whatever wins the auction.
The bidder with the lowest signal/valuation/whatever expects zero surplus (i.e., they win nothing).
All the bidders are risk-neutral.
All bids are drawn from a strictly increasing and atomless distribution.
then your choice of auction mechanism will not have an impact on your revenue as the seller. As long as those conditions are met then your choice of mechanism is not relevant to your overall revenue. Bid Shading and Revenue Volatility So, why bother with any debate about first versus second-price auctions if the revenue to the seller is equivalent? Well, because there are some problems with using a first-price auction. First of all, bidders don’t have any reason to bid their true value and instead are more motivated to engage in bid shading, which just means bidding slightly less than you think the item will sell for. This kind of bidding strategy leads to more volatility in revenue for the seller, even though the long-term revenues would be equivalent by the Revenue Equivalence Theorem. Most of the people in the Finance and Controlling departments don’t really like increased volatility in the revenue, so it’s not so cool for them or anyone else involved on the business side to have to deal with it. Additionally, if you are on the buy side then you have to do all kinds of trickery and magic in order to figure out what the true market price is. It’s a waste of time and energy and it doesn’t make anybody happy.
One thing that was touched on by Jonathan Wolf was the use of what he called dynamic pricing, which is just a different way of saying price discrimination. In theory, price discrimination comes into play when you have a market for goods, the goods cannot be transferred easily or at all after being sold, and there is either only one place to get the goods or a few limited sources. In this way it is possible to charge different buyers different prices for the same item. There are different kinds of price discrimination that we encounter in our every day lives. For example, the concept of buying in bulk in order to pay a lower cost per unit is a standard example of so-called second degree price discrimination. If you have a monopoly on certain publishers and you are entering that inventory into an RTB system and you also have really fantastic data analysis skills, then you can take advantage of monopolistic effects and do things like first degree price discrimination, which basically means you charge the buyer exactly what highest price that they’re willing to pay. However, price discrimination may not be advisable as a long-term business strategy as the eventual result is that people would rather do business somewhere else than deal with your games. This leads to a discussion on . . .
Most of my experience comes from the financial services industry, and almost all of that was in doing consulting work for mutual funds and investment advisors in the US. This industry is heavily regulated (although it needs more if you ask me) and many of these regulations are designed to provide transparency and accountability. We will need the same two things in the RTB space as computational advertising evolves. Any company that is going to provide an RTB system for others to participate in should be as clear as possible about how their auction system works, the design of the mechanism, how any conflicts of interest are handled, and so on. A lot can be learned here from the transparency and accountability rules of the finance space. The more transparent you are with regard to your RTB system, the more comfortable people will be participating. When I was at the AdTrader Conference in Hamburg last month, one of the things I discussed with quite a few attendees was concern about and desire for transparency in RTB systems. I can tell you all that any sort of black box implementation just isn’t going to cut it. I spoke with people from all sides who were skeptical about participation in such an exchange. You will have to be open and transparent about your systems (to extent possible) if you want people to buy in and be comfortable doing business with you. This is not some new groundbreaking business philosophy, but some companies will be tempted to take advantage of the newness and lack of understanding about RTB systems in order to try to make a quick buck. You probably wouldn’t buy or sell stocks on an exchange that didn’t provide transparent pricing information. So why would you try to sell people on a black box RTB auction system?
So, the key takeaway here is that even though first-price and second-price auctions generate equivalent revenue for sellers, first-price auctions come with a lot of baggage that doesn’t make sense to deal with. If you’re a buyer, then first-price auctions are a pain because you have to employ strategies for true price determination/bid shading, and it just adds complication. Additionally, if you are designing an RTB system, transparency and accountability are the name of the game. Be honest about how things work and the peace of mind your participants have will be rewarded with loyalty.