Characteristics and introduction of GFP, GSP and VCG __ Ads

GFP (Generalized First Order price auction), the main feature is that the key words in the auction between advertisers is a repeat game, after each round of auctions, advertisers will be based on the last round of quotes to determine the next round of the price decision, and this war will be naturally divided into price climbing stage and price collapse phase. Can only be in search engine companies understand the premise of the evaluation of advertisers operating, otherwise, because there is no equilibrium, volatility will bring to a great extent the loss of auction efficiency.
GSP (Generalized second order price auction), in simple terms it is paid number = Click Number * Next bid. In practical applications, Google has merged this phenomenon into its newly designed GSP auction mechanism, which is the cost of each click of the ad in the first advertisement, equal to the ad bid for the I+1 advertisement plus a very small value (typically 0.01). There are two kinds of sorting strategies here: The first of which we call bidding ordering, in reverse order according to the advertisement, The Advertiser who is clicked pays his next fee; the second we call the revenue sort, which is sorted according to the maximum expected return, where the expected revenue is bid*ctr, and the cost of the clicked Advertiser is bid (i   +1) *ctr (i+1)/ctr (i). VCG (vickrey-clark-groves), by calculating the sum of the losses that an advertiser participates in an auction for other advertisers, pricing.

How to understand these nouns, let's look at the specific example:     assume that the main coffee will pay the most for the keyword, the main B will be the same keyword coffee the most paid $0.74. If B starts with a bid as low as possible, assuming $0.10, then a advertiser will bid 0.11 to get the first ad bit. Ad main B will make a 0.12 response to that in the past. Once a bids $0.75,b no longer follow the bid $0.76, because B thinks coffee this keyword is only worth 0.74. If you want to get a second ad bit, b just need to bid $0.10. A you just have to bid $0.11 to get the first ad bit, so the cycle begins. Google's AdWords platform developers have changed the bidding style by designing a more stable second-order auction (second price auction).      in the second-order price auction, the highest-priced advertiser only needs to pay a high price and add a fraction $0.01. At this point, The advertiser's interest in changing his or her price is much smaller, because the fees you pay are determined by the next bid. Consider a previous example where the advertiser a coffee the maximum bid for the keyword, while the ad main B bids $0.74 most. Only the search engine company knows the price of all bids in closed records. If a bids $1,b bid $0.74, then in a two-order price auction, a needs to pay $0.74 + $0.01 (0.01 is the minimum difference), B pays the lowest bid price $0.01. Now let's assume that A and B have more of a bidding skill. If a bids $0.78 (less than a thinks the value of the keyword itself) and B bids $0.80 (more than B considers the value of the keyword), then B gets the first ad at $0.79 price. This price is much $0.05 than B's highest bid. To put it another assumption, if a bids $0.70 and B bids $0.65, then a will get the first ad bit $0.66. But if B bids $0.74, he will get the first ad bit at $0.71 Price, which will save $0.03. When advertisers want to see or speculate on their competitors ' bids, they can do it. For example, if A's real bid is $1,b can bid $0.99 instead of $0.74, force A to pay for his ad bit instead of $0.75.    but the second-order price has not completely solved the problem, which is an example of Intel blocking AMD. In 2005, Google introduced the quality score factor to solve the problem. The key to Quality score is CTR (Click through Rate), withThe above example, Intel advertising because the above advertising language is Coca Cola, click Only those who shake hands, if the 0.01%,AMD ads if there are 0.5% of people will point, then calculate the revenue of the search engine, the assumption shows 10,000 times, If the display of Intel's ads can only revenue 10000 * 0.01% * $ = $, while the display of AMD ads will be 10000 * 0.5 * $ = $, obviously, should not be in the bidding rankings, only consider prices, but should be integrated prices and click-through. Google quality scores also include: click-through, keyword and advertising relevance (that is, to prevent Intel from Coca Cola), advertising account history, quality of the target page (for example, you use a completely unrelated to your site advertising language to entice user points), and so on.           When it comes to Google technology, we generally think that's the answer, but Google doesn't do the best thing. First, in the second-order price auction, The advertiser's bid can not reflect the user's real valuation of the ad, such as Intel and AMD are cast "CPU" This advertising word, Intel out of $2,AMD can not afford such a high price, it would have thought the second ad bit value $0.5, it would like me out of $ 0.5, that Intel only uses each pay $0.51, since cannot get the first advertisement position, I also cannot cheap intel, I $0.99 good. This leads to the theoretically famous vickrey-clark-groves (VCG) auction, which is priced by calculating the sum of the losses that an advertiser took to another advertiser.
Proceed to the first example, assuming that the ad master a bids $ per click, and the ad main B bids $ per click, and the Advertiser C bids $. Now there are only two ads, the first ad bit on average 100 clicks per hour, the second has 50 clicks. A gets the first ad bit, B gets the second ad bit. In a generalized second order price auction, a pays $500,b to pay $ $ (ignoring the fraction of 0.01). To use the VCG method to price, we have to consider if a is not involved in the auction will happen, a is not involved then B will get a 1th ad, will get 50 more clicks, and B think each click value of $, so A's appearance means that b loss of $, the same, if a does not appear, C will get 50 clicks, c think each click Value $1,a appear means c loss of $, so according to Vcg,a pay should be to B and C loss of the sum, namely $250+$50=$300.   Now consider B if it does not appear, a position will not be affected, c that is to lose 50 clicks per hour, for C loss is $, so B VCG pay should be $. Google did not adopt the VCG approach because the VCG approach is more flawed in practice than the generalized second-order price.

Generally speaking, the market design in the field of computing advertising needs to rely on the basic knowledge of game theory, auction theory and mechanism design, how to order advertisement, it is an integrated problem of information retrieval and economics, and the GSP is not a real price; VCG is a real quote, and a stable one, But in fact we do not use it; the mechanism design of search advertisement is still an open field, it is a very good research direction.