| Amazon.com Inc. sells advertising on its site: If you search for something on Amazon, some of the results you see are “sponsored listings,” that is, advertisements from companies that want to sell you their products. Approximately speaking, the way Amazon sells those ads is with a second-price auction [1] : Advertisers submit sealed bids for each sponsored-product slot, the highest bidder wins, but the price that the winner pays is a tick more than the second-highest bid. So if there are three bidders who bid $10, $8 and $6 for a spot, the bidder who bids $10 wins, but it pays $8.01, a penny more than the second-best bid. [2] Second-price auctions are the normal way to sell online advertising, and the theory is that they give bidders better incentives to bid their real valuations than a normal auction where the winner pays the highest bid. How well this auction works depends on how many bidders there are. If 100 companies bid for each advertising spot, the auction is probably pretty competitive, the second-highest bid is probably pretty high, and you get an efficient auction whose price reflects the value of the ad. On the other hand, if only like two or three companies bid, you might get some inefficient auctions where the clearing price is pretty low. This comes up because Amazon does not actually sell ads to the highest bidder. In showing users sponsored listings, Amazon considers two factors: (1) who will pay the most for the ad and (2) how “relevant” the ad is to the user’s search. When a user searches for “paper towels,” she wants to see results for paper towels. It might be more lucrative to show her advertisements for, you know, Ponzi schemes; advertisers might be willing to pay more to show her non-paper-towel ads than they would to show her paper-towel ads. But that would be bad for Amazon’s business overall: Users would get annoyed if their search results were always wrong. Amazon says: Our early approach was simple: advertisers enter a bid, and if they win, they would pay just enough to beat the next highest ranked ad. Our auctions took relevancy into account to a degree, but they were much more weighted toward the highest bid amount. While we could have decided to continue to favor the higher bids, we instead chose to focus on more relevant bids to ensure the best possible shopper and advertiser experience. As our advanced machine learning-based relevance models more heavily weighted relevance versus highest bid, we saw winning bids drop significantly. That was good for advertisers and shoppers but meant premium placements in our Store were being undervalued. The rough intuition is: 100 companies were willing to pay for ads on each search, which produced a price that Amazon liked, but only 2 or 3 of them were advertising relevant products, and those were not always the highest bidders. Amazon, over time, decided to prioritize the relevant products: Crudely speaking, it ignored the irrelevant bids and conducted an auction only among the relevant ones. This produced fewer bidders, less competitive auctions and lower winning prices. This is a problem, for Amazon: “Premium placements in our Store were being undervalued.” Here is a conceptual solution to this problem: Add one more bidder to the auction. The one extra bidder is Amazon itself, wearing a fake mustache, and it bids “what we estimate to be the true market value of the ad placement.” Where does it get that estimate, if not from the auction price? Shh, never mind, machine learning. [3] Anyway now instead of, say, three bidders, there are four. If Bidder A bids $10 and Bidder B bids $2 and Bidder C bids $1.50, but Amazon thinks the “true market value” is $9.50, then Bidder A wins the auction with a best bid of $10, and Bidder Amazon-in-a-Mustache comes in second with a bid of $9.50. So Bidder A pays $9.51, a penny more than the second-best bid, rather than $2.01, a penny more than Bidder B’s bid. Arguably this is still a second-price auction: The highest bidder still wins, and still pays the second bidder’s price. It’s just that there’s an extra bidder, Amazon itself, that sometimes provides the second-best price. And that second-best price is sometimes higher than the second-best outside bid. But what if it’s higher than the first-best outside bid? Like: Bidder A bids $9, Bidder B bids $2, Bidder C bids $1.50 and Amazon’s bid (the “true market value”) is $9.50. Amazon wins the auction, but of course Amazon doesn’t want to win the auction: It wants to sell ad space to a buyer, not keep it for itself. So the actual result is that Bidder A wins the auction, as the highest outside bid. What price does it pay? Well, the second-highest bid. The highest bid is $9.50 (Amazon’s), and the second-highest is $9 (Bidder A’s). So bidder A pays $9, its own bid. That’s kinda weird. Like: If you are bidding in a second-price auction, and you bid $9, and you hear back “good news, you won, that’ll be $9 please,” you might be puzzled. You might say “no, it’s a second-price auction; if I won with a bid of $9, that means I was first-best, and I should pay the second-best price. Why am I paying my bid?” Here is where it is perhaps relevant to mention that my description is vastly oversimplified and, as it were, slowed down. These auctions happen bajillions of times per day, electronically, every time someone runs a search on Amazon. No advertiser is submitting a bid package and getting back a result and comparing its bid to the price it pays. It’s using algorithmic tools to automatically submit lots of bids, paying some aggregate cost for the ones it wins and getting a bill at the end of the month. [4] It might never notice that it’s winning the auction and paying its winning bid. Here’s how Amazon describes this system: With relevant ads increasingly winning at prices below market value, we began to test a concept called “soft reserve prices,” a real-time minimum value that seeks to better reflect what each placement is actually worth. We also introduced what we call a “hard reserve,” which is the minimum a bid must surpass to enter an auction. The hard reserve helps cover our costs whereas soft reserves represent what we estimate to be the true market value of the ad placement. Reserves like these are common across the industry. Our auction looks at a combination of which ad is most relevant to the customer and the price an advertiser is willing to pay. Here’s how it works: Advertisers bid a maximum price for a placement. When the winning advertiser’s bid exceeds both the hard and soft reserve, they pay the soft reserve, which is less than they were willing to pay. When the winning advertiser’s bid exceeds the hard reserve but doesn't meet the soft reserve, we still grant the placement to that advertiser and they pay their bid. In no scenario does an advertiser pay more than their bid. … With this approach, in 2024, approximately 92% of selected Sponsored Products ads were not the highest bid, often by a wide margin. The mean winning advertiser’s bid is typically about the 128th bid by amount. This means the winning advertisers' cost is almost always lower than if we had selected ads on bid alone. The “soft reserve” is what I have described as the Amazon-in-a-mustache bid, the second-best (or sometimes first-best) bid submitted by Amazon itself. If the winning bidder is higher than Amazon’s bid, it pays the second-best price (i.e. the soft reserve); if it’s lower, it pays its own bid. Here, on the other hand, is how the US Federal Trade Commission describes it: [Yesterday], 22 states joined the Federal Trade Commission in filing suit against Amazon, alleging that the company engaged in deceptive and unfair practices that secretly inflated prices in its online search advertising auctions. The complaint alleges that, for over seven years, Amazon has covertly and substantially increased the prices that more than one million brands and sellers were required to pay to advertise on its platform. As a result, the complaint alleges that Amazon’s scheme has likely extracted tens of billions of dollars from its unwitting advertising customers. … As described in the complaint, Amazon has represented to prospective advertisers for years that Amazon runs “second price” auctions where the winner of the auction would only pay “one cent more than the next highest bidder” for each successful bid for an advertising keyword. However, in practice, the complaint alleges that Amazon has charged its Sponsored Products advertisers their own winning bid close to 80% of the time and has effectively converted its nominally “second price” auction into a first price auction. … Amazon told advertisers it ran a GSP [generalized second-price auction], but for years its auction pricing had “a surcharge hidden in it,” in the words of one internal Amazon document. The complaint alleges that, beginning in 2019, Amazon changed its auction rules without notice by adding an undisclosed surcharge that Amazon referred to internally as a “soft reserve price.” This resulted in advertisers paying substantially more than the price determined by the GSP auction. The FTC and 22 state attorney generals sued Amazon yesterday. Here is the complaint. The theory is that Amazon lied to advertisers about how its auctions work: It told them that it was running a second-price auction, which led them to submit relatively high bids, but actually it “has charged its Sponsored Products advertisers their own winning bid close to 80% of the time and has effectively converted its nominally ‘second price’ auction into a first price auction.” If you say “we’ll charge you the second price,” and actually charge the highest price, that does seem deceptive. Amazon disagrees, for kind of an interesting reason. You and I and the FTC can sit around and talk about a schematic idealized version of how Amazon’s ad auctions work, a version in which advertisers submit bids and Amazon picks the highest bidder and then either does (good) or does not (fraud) charge the highest bidder the second-highest price. But in the real world, Amazon’s ad allocation process is a gigantic machine-learning-driven optimization problem designed to optimize both relevance and revenue, a process that runs entirely electronically many times per day and that is functionally a black box to everyone involved. “Advertisers often do not even know the amount of their bids because they adopt tools provided by Amazon to adjust their bids automatically,” says the FTC complaint, which kind of undermines the argument that Amazon’s disclosures tricked them into making higher bids. Amazon says: The FTC claims advertisers were harmed because they didn’t understand how our auction worked and therefore overpaid. Not only do we properly describe our pricing and auctions to advertisers, but this claim fundamentally misunderstands how advertisers behave. Advertisers adjust bids based on real-world outcomes, not descriptions of auction mechanics. … As a customer-centric advertising company, our advertising team, including senior leaders, communicates with advertisers on a regular basis and it’s not unusual to simplify a description in these instances. Back to how advertisers really form their bids, ad buyers today use highly sophisticated, automated, programmatic advertising platforms that have a deep understanding of how auctions work across different providers, enable bid experimentation and are used to maximize results and return on investment. Reserve prices are common in the industry and our use of them is consistent amongst industry leaders. The presence of reserve prices actually doesn't change how advertisers bid in practice. Advertisers optimize their campaigns based on actual auction outcomes — what they pay, what they win, and the performance they see. They don't bid based on simple descriptions of auction format. Even if an advertiser wanted to factor reserve prices into their strategy, they couldn't easily do this because reserves are determined in real time and aren't predictable in advance by anyone, including Amazon or the advertiser. Advertisers of all sizes actively manage their bids using a range of tools and data. Many use automated bidding tools from Amazon or third-party services to manage campaigns based on real-time performance data. These tools monitor clicks, purchases, cost per click, and return on ad spend, and automatically adjust bids to get the best results. Amazon’s basic case is that it charges advertisers prices, overall, that are pretty good, and gives them a pretty good return on their advertising spend and pretty good visibility into their results, and advertisers can look at their overall spending and revenue and be satisfied, and that the market microstructure of all of this is Amazon’s own black box that it continually tweaks and is really nobody else’s business. “Simple descriptions of auction format” might be helpful to give new customers an intuition about how the auctions work, but they are not legally binding, and everyone understands that the reality is messier and more proprietary. I come to this from the financial industry, where things are different. The auction rules of, say, a stock exchange tend to be fixed and binding and subject to regulatory approval, and when a platform also operates its own secret bidder that is often scandalous. [5] “We use a machine-learning-based model to match buyers and sellers in the optimal way, and our users trade based on real-world outcomes, not descriptions of auction mechanics” is not a standard argument in financial markets, the way it (apparently) is in internet advertising. I wonder if we’re heading that way, though. The process for matching buy and sell orders on the stock exchange, which has its roots in a time when the stock exchange was a physical location, is just about graspable by humans. The processes for matching online advertising orders on Amazon are more complicated and “aren’t predictable in advance by anyone, including Amazon.” As finance becomes more AI-driven, perhaps that will be the future of markets. Here’s a weird timeline. GoPro Inc. is a camera company that has had a rough few years; as of last week its stock price was around $0.60 per share, for a market capitalization of about $110 million. By early July, a YouTuber named Mark Fischbach (who goes by Markiplier on YouTube) had bought 13.5 million GoPro shares, about 8.5% of the company. He disclosed that stake in a filing with the US Securities and Exchange Commission on Aug. 20. This disclosure does not seem to have caused much of a ripple at the time. This past weekend, Bloomberg’s Lucas Shaw published a newsletter with the title “YouTube Star Markiplier Is Now GoPro’s Largest Shareholder,” containing an interview with Fischbach. As far as I can tell, this is how the market noticed Fischbach’s position (which, again, he disclosed on Aug. 20). The stock, which closed at $0.5999 on Friday, went up over the course of the day yesterday and closed at $0.8762, a 46% one-day gain. Small beaten-down consumer-facing company, significant short interest, new (or newly noticed) big investment from a celebrity, stock ripping: This all sounds like a meme stock. “A YouTube star is GoPro’s biggest shareholder, and meme stock traders are loving it,” Business Insider reported this morning. “GoPro Stock Rockets 140% as Markiplier Money Meets a Short Squeeze,” reported Benzinga. Fine. But then this morning GoPro announced a merger? GoPro, Inc. (NASDAQ: GPRO) and Starman Optical, Inc. ("Starman"), a privately held optical-photonics company, today announced that they have entered into a definitive merger agreement. In connection with the proposed transaction, GoPro shareholders will receive an aggregate cash payment of $285 million, or $1.14 per share, subject to potential adjustment based on GoPro's net working capital at closing and will maintain ownership of approximately 10% of the outstanding shares of the Company. GoPro's outstanding debt of approximately $92 million will be repaid in full at closing, resulting in a clean, substantially debt-free balance sheet.
Okay? The stock was at about $1.31 per share as of noon today, which is higher than the cash merger price, though I suppose that reflects the fact that current shareholders will keep 10% of the company. “The company said it was exploring strategic alternatives in May,” noted Shaw, so it is perhaps not quite a surprise that it announced a merger today. Still, odd timing. Like, by this point there is a traditional playbook for a public company that suddenly becomes a meme stock. Mostly you sell stock, in an at-the-market offering or maybe a floating-strike convertible. I suppose doing a merger is a version of that: “Oh, our stock is shooting up, we should sell 100% of it to an acquirer.” [6] I don’t think I’ve ever seen that version of the playbook executed so quickly: GoPro became a meme stock on Monday and announced that it was being acquired on Tuesday.But surely GoPro didn’t call up Starman Optical yesterday to propose this deal; surely this deal has been in the works for a while. In that sense, becoming a meme stock yesterday was arguably inconvenient. When GoPro was trading at $0.60 per share and drifting downward, selling for $1.14 or $1 or whatever the deal price was last week would have been relatively easy. Once you’ve got a meme-stock rally, it might be harder to strike a deal: An acquirer will only want to pay the fundamental value of the company, but the meme shareholders are not similarly constrained. if GoPro had traded to $10 per share (!?) on Markiplier’s investment, it would have been hard to get a deal done: Meme shareholders’ expectations would outstrip a buyer’s willingness to pay. Once GoPro became a meme, getting the deal done became more urgent. Bloomberg’s Ira Boudway has a story about a gambling addict who relapsed into gambling on Kalshi. (“This is a cherry-picked case,” says Kalshi.) We have frequently discussed the weird situation that Kalshi is (1) obviously an online gambling site but also (2) a federally regulated financial market. Federal financial regulators care about stuff like market manipulation and fraud and insider trading, while state gambling regulators care about stuff like consumer protection and gambling addiction. Kalshi’s federally regulated status means (probably!) that it is exempt from the state rules designed to limit the harms of gambling addiction. But Kalshi is an online gambling site, and would like to mitigate the harms of gambling addiction for altruistic or marketing or political reasons, so it sort of voluntarily follows some gambling-site best practices. (Not all! It offers gambling to 18-year-olds, for example. And in Utah, which prohibits gambling, and where Boudway’s protagonist lives.) Boudway writes: The company offers a variety of risk management tools, the spokesperson says, including allowing users to temporarily restrict their own activity, self-exclude and set deposit limits. Kalshi maintains a nationwide list of users who have self-excluded, the spokesperson adds, and has asked state gaming regulators to share their opt-out lists, so far, it says, to no avail. It also limits potential losses on each market and works with the National Council on Problem Gambling and with addiction treatment providers to help address irresponsible behavior. “We’ve prioritized making Kalshi the safest venue for people to trade on,” the spokesperson says. “Has asked state gaming regulators to share their opt-out lists” is the weird one there. Like: State gaming regulators try to make Kalshi comply with their rules, and Kalshi says “nope, we’re not a gambling site, your rules don’t apply.” And then it turns around and says “hey can you send us your list of self-excluded problem gamblers so we can responsibly exclude them from our gambling site,” and the states are understandably confused. Bobby Jain’s Hedge Fund Made $1.8 Billion as Millennium Cash Rolls In. US Gets Board Veto, Right of First Refusal in Venezuela Oil Deal. Trump’s Venezuelan Oil Company Plans Massive Drilling Push. Revolut’s mission to dominate banking. Companies Plow Tariff Refunds Into Price Cuts, Appealing to Stretched Consumers. Private Markets Set for Delay in Tapping German Pension Cash. How Aston Martin’s latest financing sparked a bondholder revolt. Citadel Says Marshall Wace ‘Stonewalling’ in Recruitment Spat. A $140 Million Ponzi Fraud Is Haunting Georgia GOP Ahead of Midterms. A Lucrative Crypto Contract Sets Off New No-Bid Fight for Trump. Ernst & Young Is Giving $100 Million in Bonuses to Staff for ‘Human’ Skills. “There is no plutonium market and it doesn’t exist in nature.” If you'd like to get Money Stuff in handy email form, right in your inbox, please subscribe at this link. Or you can subscribe to Money Stuff and other great Bloomberg newsletters here. Thanks! |
