Some Auction Strategies

In online ad industry, ad inventory is increasingly sold through Real Time Bidding (RTB). In this article, we describe three of the most prominent auction strategies used in RTB. All these auctions are sealed bid auctions. (i.e. bidders do not know about values bid by each other)

Second price auction

This is an auction method where there is one object to be sold and multiple buyers. Each buyer bids a value. The highest bidder is awarded the object and he pays the price equal to the second highest bid value. For instance, if a pen is being auctioned and P1, P2, P3 bid 70, 50, 40 respectively then P1 wins the auction and pays 50.

Truth telling is dominant strategy in second price auction

What it means is that each participant has an incentive to bid the true value that he attaches to the object.

Let's see this using an example. Let's say an auction for a pen is taking place. I value that pen at 50. Then, it is best for me to bid 50. Why?

As can be seen from above, truth telling is never worse off than bidding higher or lower and sometimes it is better than alternative strategies. Hence, truth telling is a dominant strategy in second price auction.

Why auction systems where truth telling is dominant strategy are good

In case participants are not incentivized to bid their true values, their bids would be dependent on their guess about other participants' bids. This would lead to differing bid values by participants even when inventory quality remains the same. Thus, revenues of seller would be variable, and that is not desirable by the sellers.

Generalized second price (GSP) auction and Vickrey Clarke Groves (VCG) auction

They are auction strategies where

There are various ways to understand GSP and VCG. In this article, let's consider the setting that is suitable for understanding auction in ad space.

Example below is inspired from [this](http://en.wikipedia.org/wiki/Generalized_second-price_auction) wikipedia article. Let's say there are two ad slots with CTRs of 0.9 and 0.4 respectively (irrespective of the ad that is shown here). Let's say there are three bidders A, B and C who value the clicks at 7, 6 and 1 units respectively. They need to pay the seller only if an ad gets clicked.

Let's say the participants bid values 7, 6 and 1 respectively.

GSP

Here is how GSP conducts the auction. Highest bidder, A, wins first slot and pays the second highest bid (6). Second highest bidder, B, wins the second slot and pays the next highest bid (1). Net value realized by A is (7 - 6) * 0.9 = 0.9 Net value realized by B is (6 - 1) * 0.4 = 2 Net revenue of seller is 6 * 0.9 + 1 * 0.4 = 5.8 GSP has a shortcoming that revealing their true valuations in bids is not always optimal for the participants. For instance, in above example A can bid 2, which leads B to win first slot at 2 units and A to win second slot at 1 unit. In this case, Net value realized by B is (6 - 2) * 0.9 = 3.6 Net value realized by A is (7 - 1) * 0.4 = 2.4 Thus, we see that A can realize more value by bidding lower than his true value.

VCG

Here is how VCG conducts the auction. As in GSP, highest bidder, A wins the first slot and second highest bidder B wins the second slot. To calculate the bid value of A, we find how much penalty does A cause to other participants by being present in the system, and winning the first slot. If A were not present in the system, B and C would win first and second slots and derive a total value of 6 * 0.9 + 1 * 0.4 = 5.8. Once A gets first slot, B derives a value 6 * 0.4 = 2.4 and C derives a value 0. So, A should pay 5.8 - 2.4 = 3.4. Similarly, B should pay 1 * 0.4 = 0.4, since this is the value by which C is harmed by his presence (A is not harmed by B's presence). In this case, total revenue realized by seller = 3.4 + 0.4 = 3.8

A few notes

Where are various auction mechanisms used