The MFN thing seems similar to a credit card company demanding that retailers who accept the credit card not pass on the fees or give a discount to cash and/or debit customers. What is the state of the existing law and practice around that and what lessons does that hold for Amazon's MFNs?
The credit card issue is called "steering" and was the subject of a Supreme Court case called Ohio v. American Express Co. (https://en.wikipedia.org/wiki/Ohio_v._American_Express_Co). The Court ruled for Amex, arguing that the steering practice did not violate antitrust laws because the DOJ failed to prove that it harmed consumers. This was a 5-4 ruling, and I believe that the majority got this one wrong; I think banning merchants from offering a cash discount is clearly harmful to consumers. More relevant for our purposes here, I think that the test created by the Amex precedent — that in cases concerning two-sided platforms, the plaintiff must show that the alleged anticompetitive behavior harms both sides of the market and not just one — doesn't make sense.
That said, I am obviously not currently in a position to change the Supreme Court's mind on this and the precedent is what it is. This paper is a bit long, but it discusses how Amex might apply to antitrust action against Amazon, which is also a two-sided market: https://fordhamlawreview.org/wp-content/uploads/2019/10/Bloodstein_October_N_5.pdf
Well, that's unfortunate. Seems like it could be fixed by statute, though, no? That is, Congress could clarify that it really meant to make MFN clauses unenforceable throughout the economy, much as some states make noncompetes unenforceable.
The MFN claim seems facially plausible but (regardless of its legal status) the idea that the ad auction mechanism is actually raising prices seems hard to believe. As Amazon points out in their response to the court case, advertisers don't buy ads by working out their utility curves a priori, setting an appropriate price for a pure second-price auction, and then forgetting about it. The ad buys are a repeated game, and the actual thing that determines ad spending is what return you've seen on advertising in the past.
Amazon believed that its advertising customers focused primarily on their Return on Ad Spend, known in the advertising industry as “RoAS,” to measure the success of their advertising campaigns. Accordingly, Amazon’s use of soft reserve prices was focused on capturing the share of advertiser RoAS that it believed it could extract without a negative advertiser reaction. In a 2020 internal memorandum, the Sponsored Products auction team reported that “advertisers seem to be more directly responding to ad spend/RoAS changes as opposed to clicks being charged first price,” and “given the empirical data we have on advertiser response to RoAS changes, we will use it as the main signal to tune our pricing controls.”
The rest of the section on return on ad spend is heavily censored. But this suggests to me that Amazon was leveraging its market power to extract as much surplus from advertisers as possible without tipping them off to the change in auction type. We also know, from internal Amazon memos cited in the complaint, that Amazon thought its strategy on ad auctions increased its own revenues at the expense of advertisers. I think it is entirely plausible that higher advertiser costs weren't passed along to consumers. But the increased costs for advertisers is still bad.
I don't see what the argument is supposed to be there for the claim that this enables Amazon to misrepresent the advertiser's RoAS to the advertisers. It would certainly have to be a different argument than "advertisers thought they getting true Vickrey auctions, but they weren't".
There are a bunch of examples in the complaint of Amazon's official resources for merchants seeking to buy ads stating that they ran Vickrey auctions (pp. 36-56), and advertiser communications with Amazon indicating that they believed they were bidding in Vickrey auctions (pp. 78-95) in the complaint.
For example, paragraphs 137 and 138 on page 51:
137. These explicit false representations that Amazon Ads uses a second price auction were not limited to sales meetings. In February 2024, the Senior Vice President in charge of Amazon Ads was personally emailed by the CEO of an advertiser, who asked: “Regarding pay per click[,] If I put in a very high dollar amount per click, does that mean it will cost me that much per click, or will the system only charge me the going rate for that keyword at that time. So for example, if the competition is bidding between $1.50 and $2.00, if I bid $5, will it charge me 5 or will it charge me somewhere between 1.50 and 2.00[?]”
138. The head of Amazon Ads replied with the same false representation that Amazon consistently makes to its advertising customers: “Generally, the amount you pay is governed by a generalized second price auction - which means that the amount you pay for a click is dependent on what the next lowest bid is. So, if you bid $2.00 and the next lowest bid is $1.00, you win the auction and pay $1.01.” The advertiser then confirmed that he had executed his bidding strategy in reliance on this false representation: “Yes that is perfect[.] So I just went ahead and put 5 dollars so that I for sure win every keyword. Thanks for everything.”
Sure, which may well be sufficient for establishing actionable misconduct by amazon. But what I am not seeing there is "then it charged him 5 after all and he didn't notice for some reason". The problem that this is a repeated game where ad purchasers get to observe the results of their spending remains.
It doesn't matter if they maintained a positive ROAS for their advertisers. If you are running an auction, you still ought to adequately and accurately disclose the mechanisms of it.
You'll note that my comment did not say anything about whether Amazon acted rightly (they didn't), or whether the FTC should sue them over it (maybe?). It still matters whether the claim that this meaningfully increases consumer prices is true or not.
As someone who works on online ad systems, that part of the article is a little weak. Most ad systems (especially as complex as Amazon’s) will use a form of auto bidding with a pacing mechanism, where bids are implicitly conditioned on:
- Spend rate over the lifecycle of the campaign
- Remaining budget over the lifecycle of the campaign
- Probability of click and probability of order
- Observed ROAS for this merchant
Second price auctions tend to actually introduce bias and unfairness, because the advertisers with very high probability of click (typically, megabrands) tend to bid disproportionately high but get larger “discounts” than smaller merchants. Moving to first price doesn’t necessarily imply unfairness, oftentimes it improves ROAS for smaller merchants because it unlocks more opportunities as larger brands spend budget quicker and free up supply.
Otherwise, a great article though! just a minor quibble maybe
If you found this interesting, I encourage people to read Matt Levine's explainer into the Amazon auction dynamic: https://archive.is/PWcMY
I'm somewhat persuaded that reality is messier than simple auction models and that Amazon allows advertisers to see if they're actually profiting off these ads or not.
I agree with the overall point of this article, but I’m skeptical this point belongs here:
> Amazon effectively requires sellers to use its in-house fulfillment services (which, according to the FTC, are more expensive than competitors such as FedEx) if they want their products to have the “Prime” badge
Amazon is a logistics company as well as an e-marketplace. The reason I buy products with a “Prime” badge is usually because I want to be certain that they arrive on time and undamaged—something that I have consistently had issues with with FedEx.
It seems reasonable for Amazon to indicate prominently whether they handle logistics or whether another company is doing so.
Still, isn't the Khan concern that when a company has 40%+ market share they are able to repeatedly engage in new and varied forms of anti-competitive behavior and you end up playing whack-a-mole?
Well done, you got me at least sympathetic when I started out with a high prior that an anti-Amazon article would be a "fig leaf for crude anti-business views or a stalking horse for socialism." But I'd lead, or at least tease, with the allegation that they're running a first-price auction while telling bidders they're running a second-price one.
I'm much more skeptical of the MFN position and even if true, would want a general "MFN ban" applicable to all companies (e.g., like banning non-compete agreements) and not have this be part of an anti-trust suit against Amazon. Is the case tight enough though that MFN hurt competition to warrant general bans?
I think so. MFNs are a type of "contract referencing rivals" (see https://www.justice.gov/archives/atr/file/518971/dl?inline=). In theory, there may be some situations where MFNs are efficiency-promoting — they can help solve certain kinds of hold-up problems when one party is considering making a large relationship-specific investment — but there is a good case that MFNs are generally anticompetitive.
Well-argued and informative piece. Loved it.
Oh man Milan, OG vintage take assuming the point of antitrust is to protect consumers from artificially high prices ;)
Really strong article and agreed.
It is important for government to make sure companies do not employ anti competitive business practices
But government, shouldn't it be either pro or anti big business
The MFN thing seems similar to a credit card company demanding that retailers who accept the credit card not pass on the fees or give a discount to cash and/or debit customers. What is the state of the existing law and practice around that and what lessons does that hold for Amazon's MFNs?
The credit card issue is called "steering" and was the subject of a Supreme Court case called Ohio v. American Express Co. (https://en.wikipedia.org/wiki/Ohio_v._American_Express_Co). The Court ruled for Amex, arguing that the steering practice did not violate antitrust laws because the DOJ failed to prove that it harmed consumers. This was a 5-4 ruling, and I believe that the majority got this one wrong; I think banning merchants from offering a cash discount is clearly harmful to consumers. More relevant for our purposes here, I think that the test created by the Amex precedent — that in cases concerning two-sided platforms, the plaintiff must show that the alleged anticompetitive behavior harms both sides of the market and not just one — doesn't make sense.
That said, I am obviously not currently in a position to change the Supreme Court's mind on this and the precedent is what it is. This paper is a bit long, but it discusses how Amex might apply to antitrust action against Amazon, which is also a two-sided market: https://fordhamlawreview.org/wp-content/uploads/2019/10/Bloodstein_October_N_5.pdf
Well, that's unfortunate. Seems like it could be fixed by statute, though, no? That is, Congress could clarify that it really meant to make MFN clauses unenforceable throughout the economy, much as some states make noncompetes unenforceable.
Yes
The MFN claim seems facially plausible but (regardless of its legal status) the idea that the ad auction mechanism is actually raising prices seems hard to believe. As Amazon points out in their response to the court case, advertisers don't buy ads by working out their utility curves a priori, setting an appropriate price for a pure second-price auction, and then forgetting about it. The ad buys are a repeated game, and the actual thing that determines ad spending is what return you've seen on advertising in the past.
Per the FTC complaint (paragraph 191 on page 70):
Amazon believed that its advertising customers focused primarily on their Return on Ad Spend, known in the advertising industry as “RoAS,” to measure the success of their advertising campaigns. Accordingly, Amazon’s use of soft reserve prices was focused on capturing the share of advertiser RoAS that it believed it could extract without a negative advertiser reaction. In a 2020 internal memorandum, the Sponsored Products auction team reported that “advertisers seem to be more directly responding to ad spend/RoAS changes as opposed to clicks being charged first price,” and “given the empirical data we have on advertiser response to RoAS changes, we will use it as the main signal to tune our pricing controls.”
The rest of the section on return on ad spend is heavily censored. But this suggests to me that Amazon was leveraging its market power to extract as much surplus from advertisers as possible without tipping them off to the change in auction type. We also know, from internal Amazon memos cited in the complaint, that Amazon thought its strategy on ad auctions increased its own revenues at the expense of advertisers. I think it is entirely plausible that higher advertiser costs weren't passed along to consumers. But the increased costs for advertisers is still bad.
https://www.ftc.gov/system/files/ftc_gov/pdf/AmazonAds-Complaint.pdf
I don't see what the argument is supposed to be there for the claim that this enables Amazon to misrepresent the advertiser's RoAS to the advertisers. It would certainly have to be a different argument than "advertisers thought they getting true Vickrey auctions, but they weren't".
There are a bunch of examples in the complaint of Amazon's official resources for merchants seeking to buy ads stating that they ran Vickrey auctions (pp. 36-56), and advertiser communications with Amazon indicating that they believed they were bidding in Vickrey auctions (pp. 78-95) in the complaint.
For example, paragraphs 137 and 138 on page 51:
137. These explicit false representations that Amazon Ads uses a second price auction were not limited to sales meetings. In February 2024, the Senior Vice President in charge of Amazon Ads was personally emailed by the CEO of an advertiser, who asked: “Regarding pay per click[,] If I put in a very high dollar amount per click, does that mean it will cost me that much per click, or will the system only charge me the going rate for that keyword at that time. So for example, if the competition is bidding between $1.50 and $2.00, if I bid $5, will it charge me 5 or will it charge me somewhere between 1.50 and 2.00[?]”
138. The head of Amazon Ads replied with the same false representation that Amazon consistently makes to its advertising customers: “Generally, the amount you pay is governed by a generalized second price auction - which means that the amount you pay for a click is dependent on what the next lowest bid is. So, if you bid $2.00 and the next lowest bid is $1.00, you win the auction and pay $1.01.” The advertiser then confirmed that he had executed his bidding strategy in reliance on this false representation: “Yes that is perfect[.] So I just went ahead and put 5 dollars so that I for sure win every keyword. Thanks for everything.”
Sure, which may well be sufficient for establishing actionable misconduct by amazon. But what I am not seeing there is "then it charged him 5 after all and he didn't notice for some reason". The problem that this is a repeated game where ad purchasers get to observe the results of their spending remains.
It doesn't matter if they maintained a positive ROAS for their advertisers. If you are running an auction, you still ought to adequately and accurately disclose the mechanisms of it.
You'll note that my comment did not say anything about whether Amazon acted rightly (they didn't), or whether the FTC should sue them over it (maybe?). It still matters whether the claim that this meaningfully increases consumer prices is true or not.
As someone who works on online ad systems, that part of the article is a little weak. Most ad systems (especially as complex as Amazon’s) will use a form of auto bidding with a pacing mechanism, where bids are implicitly conditioned on:
- Spend rate over the lifecycle of the campaign
- Remaining budget over the lifecycle of the campaign
- Probability of click and probability of order
- Observed ROAS for this merchant
Second price auctions tend to actually introduce bias and unfairness, because the advertisers with very high probability of click (typically, megabrands) tend to bid disproportionately high but get larger “discounts” than smaller merchants. Moving to first price doesn’t necessarily imply unfairness, oftentimes it improves ROAS for smaller merchants because it unlocks more opportunities as larger brands spend budget quicker and free up supply.
Otherwise, a great article though! just a minor quibble maybe
If you found this interesting, I encourage people to read Matt Levine's explainer into the Amazon auction dynamic: https://archive.is/PWcMY
I'm somewhat persuaded that reality is messier than simple auction models and that Amazon allows advertisers to see if they're actually profiting off these ads or not.
I agree with the overall point of this article, but I’m skeptical this point belongs here:
> Amazon effectively requires sellers to use its in-house fulfillment services (which, according to the FTC, are more expensive than competitors such as FedEx) if they want their products to have the “Prime” badge
Amazon is a logistics company as well as an e-marketplace. The reason I buy products with a “Prime” badge is usually because I want to be certain that they arrive on time and undamaged—something that I have consistently had issues with with FedEx.
It seems reasonable for Amazon to indicate prominently whether they handle logistics or whether another company is doing so.
This is really persuasive.
Still, isn't the Khan concern that when a company has 40%+ market share they are able to repeatedly engage in new and varied forms of anti-competitive behavior and you end up playing whack-a-mole?
Well done, you got me at least sympathetic when I started out with a high prior that an anti-Amazon article would be a "fig leaf for crude anti-business views or a stalking horse for socialism." But I'd lead, or at least tease, with the allegation that they're running a first-price auction while telling bidders they're running a second-price one.
I'm much more skeptical of the MFN position and even if true, would want a general "MFN ban" applicable to all companies (e.g., like banning non-compete agreements) and not have this be part of an anti-trust suit against Amazon. Is the case tight enough though that MFN hurt competition to warrant general bans?
I think so. MFNs are a type of "contract referencing rivals" (see https://www.justice.gov/archives/atr/file/518971/dl?inline=). In theory, there may be some situations where MFNs are efficiency-promoting — they can help solve certain kinds of hold-up problems when one party is considering making a large relationship-specific investment — but there is a good case that MFNs are generally anticompetitive.