Does Amazon Lower Prices? Perhaps and Doubt Test It

Professor Perhapshas a theory

My framing: Amazon's effect on prices probably isn't one number but a curve, and the interesting question is its shape. Perhaps early on, scale and logistics push prices down, and later, as sellers depend on Amazon and rivals shrink, the same power lets it raise the toll on everyone selling there. Sellers pass that toll along, and if rules stop them undercutting Amazon elsewhere, shoppers can't escape it by switching. Here is a test. The research notes cite an analysis of Amazon's exit from certain toy markets, where prices rose about 5% afterward. If that holds, Amazon's presence was disciplining prices. But I'm assuming exit works like a clean experiment, and the notes themselves flag alternative explanations. So the question for Doctor Doubt: what would a convincing natural experiment look like, one where Amazon's presence changes and nothing else does?

Doctor Doubthas questions

The toy result has a problem before we even get to design. Look at the direction of the logic. If Amazon exits a market and prices rise 5%, that shows Amazon's presence lowered prices. That's the short-term efficiency story, the left-hand side of your curve. It says nothing about whether the toll rises later. Your curve needs evidence of the right-hand side: prices rising while Amazon grows. So the convincing experiment is different from the one you asked for. I'd want sellers who sell identical goods on Amazon and off it, tracked as Amazon's fee rises, to see whether off-Amazon prices rise too. That would be a pass-through test, and it separates "Amazon charges sellers more" from "shoppers pay more." I'm guessing here, but I suspect we don't have that data publicly. Do you know of any?

Professor Perhapshas a theory

Fair hit on the toy study: it supports the left side of my curve, not the right. I concede that cleanly. My memory serves up no public dataset that does your pass-through test, and I won't invent one. But here is a natural experiment in the research notes that sits closer to the right side. The notes say regulators allege Amazon's pricing rules restrict sellers from offering lower prices elsewhere. Suppose those rules are lifted or enjoined somewhere. If Amazon's power is a toll shoppers can't dodge, off-Amazon prices should fall for sellers who were previously constrained, and Amazon's fees should start losing sellers or drifting down. If nothing moves, the toll theory weakens. That test is cleaner than exit, because it changes one rule and leaves the stores open. My assumption: that sellers actually respond to rule changes by repricing.

Doctor Doubthas questions

Your rule-change test is better than exit, but I think it can fail in a way that hides the answer. Suppose the restriction is lifted and off-Amazon prices don't fall. That wouldn't weaken the toll theory, because sellers may have been slow to reprice, or may keep one price everywhere to avoid the hassle of running two. And suppose prices do fall. A rival platform might simply have been charging low fees all along, so the drop reflects rival margins, not Amazon's toll. Either result is ambiguous. So I'd add a requirement: the test needs a measure of the fees themselves, not just the prices. If fees fall and prices fall, the toll story gains weight. If fees fall and prices stay put, sellers are pocketing the savings, and shoppers never carried the toll.

Professor Perhapshas a theory

Your fee-and-price pairing is a real improvement, and I'll adopt it. But notice that it quietly changes which question we're answering. If fees fall and prices stay put, sellers pocketed the savings, and you say shoppers never carried the toll. Perhaps so in that episode, but the same logic cuts the other way: it would show sellers absorb fees only when they have slack margins. A 15 percent fee on a 40 percent margin product behaves very differently from the same fee on a 5 percent margin product. So my new hypothesis is that pass-through depends on margins, which means the right-hand side of my curve may show up first in thin-margin categories like commodity goods and groceries, and barely at all in high-margin ones. That is a prediction, and it's testable: compare price changes after fee increases across categories sorted by seller margin.

Doctor Doubthas questions

Your margin prediction is sharper, and I'll grant the mechanism: thin-margin sellers can't eat a fee. But the sorting variable may be doing hidden work. Thin margins usually mean commodity goods, and commodity goods are exactly where shoppers compare prices fastest and have the most alternatives, including Walmart and store shelves. So a seller who must pass the fee through into a price shoppers can easily check will lose the sale, and may simply exit. Then the observed price might not rise at all; the product just vanishes, leaving a survivor-biased catalogue that looks stable. Margin and competitive pressure move together, so your comparison can't separate them. Here's a cleaner handle: track seller exits and listing counts alongside prices in thin-margin categories after a fee rise. If prices hold but listings shrink, shoppers lost choice rather than paying more. Which would you count as harm?

Professor Perhapshas a theory

I'd count lost choice as harm, but only when the missing listings were ones shoppers valued. Delisting a seventh near-identical phone charger costs almost nothing; losing the only seller of a cheap replacement part costs a lot. So the shrinking catalogue needs weighting by how closely the exiting products substituted for survivors. Here's a thought experiment. Two sellers offer the same $10 item. A fee rise pushes one out. The survivor now faces less rivalry on that page, and perhaps it nudges its price up, though that is my guess, not a finding. Prices were flat before the exit and creep after it. Both exit and price effects can matter, so we should watch the pair together. My assumption: Amazon's shoppers mostly buy from the top few listings. If so, a lost listing deep in the results costs little, while a lost top listing costs a lot.

Doctor Doubthas questions

Your weighting idea is right, but notice who gets to do the weighting. Amazon's ranking decides which listings count as "top," so the thing that determines the cost of a missing listing is itself under Amazon's control. Suppose Amazon's own Buy Box, the default one-click purchase button, favors sellers using Amazon's fulfillment service, which costs extra. Then a cheaper seller who declines that service can sit deep in the results, and when they exit, your method records almost no loss, because shoppers rarely clicked them anyway. The measurement understates harm precisely where the platform is steering. This is a hypothesis about design, not a finding; the research notes only say regulators allege fee and pricing-restriction problems. So here's a test: compare the price of the Buy Box winner against the lowest available price for the same product.

Professor Perhapshas a theory

Your Buy Box gap test is good, and I'll run with it, but it has an ambiguity that could flip its meaning. Suppose the Buy Box winner costs 8 percent more than the cheapest listing. That could be steering harm. It could also be a fair price for faster, reliable delivery that shoppers would knowingly pay. Same gap, opposite stories. So the test needs a second column: how often shoppers pick the cheaper offer when they have to click past the default to reach it. If most still take the default, perhaps they value the convenience. If they take it only because it's the default, that is the harm. This is speculation about behavior, not a finding, and the research notes don't settle it. My assumption is that defaults persuade more than they inform.

Doctor Doubthas questions

Your second column, how often shoppers click past the default, has a trap in it. The people who click past are self-selected bargain hunters. If 15 percent of them take the cheaper offer, that doesn't tell us the other 85 percent were steered rather than satisfied. It tells us only about the 15 percent. So we can't read the default's harm off observed clicks. A cleaner source is a natural experiment. Whenever Amazon changes how the Buy Box picks winners, some shoppers are suddenly offered a different default for the same product. If purchases move with the default, shoppers were following it. If the average price paid also moves, that shows what the default costs them. I'm guessing such changes happen; the research notes don't mention any, and the data would sit inside Amazon or in litigation discovery.

Professor Perhapshas a theory

The natural experiment is the best tool on the table, and I'll grant that it fixes the selection problem. But it measures the wrong horizon for our question. If Amazon changes the Buy Box rule and shoppers follow the new default, we learn what a default costs this month. We still don't learn whether the rule shaped prices over time. Here is how I'd stretch it. Sellers can read the Buy Box rule. If it rewards paid fulfillment, a seller who reprices their fees into the sticker price has, perhaps, been trained by the rule, not the shopper. That shows up after the change, in what sellers charge on Amazon versus their own websites, which the notes link to the price-parity complaint. Prediction, my guess, not a finding: the gap between Amazon and off-Amazon prices for identical goods widens after fee-heavy rule changes.

Doctor Doubthas questions

Your prediction has a hole in it: the gap between Amazon and off-Amazon prices could widen for a reason that has nothing to do with Amazon training sellers. Suppose a seller faces a fee hike and responds by pulling their cheapest items from Amazon and keeping their discounts for their own site, where no fee applies. Average Amazon prices rise and the gap widens, yet no individual product got pricier. The mix changed, not the prices. So the test has to follow identical products over time, not averages. And I'd add a harder rival hypothesis: sellers simply raise prices everywhere after a fee hike, which would shrink the gap while consumers pay more. A widening gap is one signature of harm, not the only one. Here's a question that would change my mind: do sellers who leave Amazon charge less on their own sites afterward?

Professor Perhapshas a theory

Your exit question is the right one, and I'll concede your mix-shift point cleanly: following identical products, not averages, is the only honest way to run my test. But the exit question has its own selection problem. Sellers who can leave Amazon are the ones with a brand, a loyal following, and cheap alternatives, so their own-site prices tell us little about what everyone else would charge. The sellers most dependent on Amazon are exactly the ones who can't leave, and they are the ones whose prices matter most. Here is a sharper cut. The toy study in the research notes found prices rose about 5 percent after Amazon exited those markets. Read that against your question: if Amazon's presence disciplines rivals, then leaving should raise prices elsewhere. That is the opposite of the harm story. So which way does the exit data point?

Doctor Doubthas questions

You've asked which way the toy exit data points, and I think it points at a different question than the one we're asking. The notes describe prices rising about 5 percent after Amazon exited those toy markets, with other explanations possible. Steelmanned: Amazon's presence disciplines everyone, so removing it raises prices. I accept that. But it only shows Amazon beats having no Amazon, a comparison nobody in this debate disputes. The harm story says Amazon beats the world with no Amazon, yet loses to a world where it faces a rival platform and charges sellers lower fees. Exit data can't see that third world. Worse, an exit is a shock to the whole market, so the 5 percent could also reflect lost logistics and stock, not lost discipline. Here's a handle: has any retailer ever gained a comparable rival to Amazon in a category, and did prices fall further?

Professor Perhapshas a theory

Your three-world framing is the best thing said so far, and I'll take it. I can't point to a clean case of a comparable rival arriving, and the research notes name none, so I won't invent one. But there is a partial test that sits inside Amazon's own marketplace: the experiment the harm story needs is already running every time Amazon changes a fee. If the third world is real, sellers should pass fee increases through to shoppers, and the notes list pass-through as unresolved. So I'd measure pass-through, meaning how many cents of each fee-dollar land in the sticker price. My guess, speculation only: when sellers have nowhere else to sell, pass-through approaches a full dollar. And here is where I could be wrong: if pass-through is low, sellers are absorbing fees, and the harm story weakens. Do you accept that number as the arbiter?

Doctor Doubthas questions

I'll accept pass-through as evidence, but not as the arbiter, because the number can mislead in both directions. First, a high rate doesn't prove harm. If every seller raises prices a dollar per fee dollar, that is also what a perfectly competitive market does when a cost goes up. Gas stations pass through a tax hike too, and nobody calls that market power. The harm lies in whether the fee itself is inflated above what a competitive platform would charge, and pass-through can't see that. Second, a low rate doesn't clear Amazon either. Sellers who absorb fees get thinner margins, and over time the thinnest-margin sellers quietly stop listing, so consumers pay in lost selection rather than in sticker prices. So my handle is this: what happened to seller counts and product variety after the last big fee increase?

Professor Perhapshas a theory

Both of your cases are right, and they change my claim. Full pass-through is the competitive baseline, and low pass-through can hide exit. So pass-through only tells us who pays a fee, never whether the fee is too high. Seller counts have the same flaw, since a shrinking seller base could mean thinner margins or a maturing market. Perhaps the arbiter is the fee itself, set against what a platform would charge under rivalry. My idea is to use the places where Amazon is the rival. Amazon sells its own goods with no commission to itself, and the research notes say regulators allege its price-parity rules stop sellers from undercutting it elsewhere. So compare the cheapest price for the same product on Amazon versus a seller's own site, in categories where Amazon can't enforce parity, such as outside the US, if rules differ.

Doctor Doubthas questions

Your price-parity comparison is a good test, but I think it is confounded by what shoppers are buying. A seller's own site charges the same or less, and I'd bet shoppers still pick Amazon, because Amazon bundles things the own site can't: next-day delivery, easy returns, one checkout, a trusted refund. So an off-Amazon price gap measures the fee plus the cost of replicating those services, and you can't read all of it as inflation. Here's the cleaner version, offered as a thought experiment: imagine Amazon's fee were exactly what a competitive logistics-plus-marketplace firm would charge. We'd still see sellers listing cheaper elsewhere, because their own site skips the service. So a gap proves nothing alone. What would? A seller who ships through Amazon's warehouses but sells on a rival platform for less. Does any seller do both, and does the notes' parity allegation cover that?

Professor Perhapshas a theory

Your fulfillment-only seller is the right test, and I think the research notes point toward it without naming it. Amazon offers a service where it ships orders placed on other sites, and my recollection, not something in the notes, is that Amazon has charged different fees for that than for orders on its own marketplace. If so, the logistics price is observable without the marketplace layer attached. That gives a decomposition. Take the fee Amazon charges to pack and ship an order placed elsewhere, and compare it to the fee when the same box ships against an Amazon sale. The first approximates the competitive price of delivery. Whatever the second adds is a charge for the shopper traffic and the Buy Box. Perhaps that gap is the toll. Here is where I could be wrong: if the gap is small, the toll story shrinks to the parity allegation alone.

Doctor Doubthas questions

I like the decomposition, but I think the gap undercounts the toll. The fee for shipping an order placed elsewhere is Amazon pricing against competition: it has to beat other shippers to win that business. The marketplace fee has no such discipline, so the gap would measure how much the toll exceeds the cheapest delivery, but it still leaves out the part you can't see. A seller can't take the traffic and leave the box. Here's the wrinkle: by your own logic, a small gap would clear Amazon, yet a small gap could also mean Amazon prices the bundle low and recoups through advertising. The research notes don't cover ad fees, so I can't say how large that is; I'm guessing it matters. So the better test is total take per sale, fees plus ads plus fulfillment, as a share of price over time.

Professor Perhapshas a theory

Total take per sale is the right unit, and I'll adopt it. But a rising take rate still isn't harm on its own. Imagine take climbs from 30 to 45 percent of price because sellers buy ads to win the Buy Box, the default winning offer. If every seller must pay for placement, the ad bill works like rent on a shelf that can't hold everyone. Then the ad share is the marketplace's scarcity price, and my assumption is that this price is set by Amazon, not by competition among shelves. So the sharper split is between the take that buys access and the take that buys visibility. A seller can leave a pricey mall and sell elsewhere; a seller buying ads inside the mall is bidding against neighbors for the same eyes.

Doctor Doubthas questions

Your access-versus-visibility split is clean, but there's a wrinkle: ad spending is partly sellers bidding against each other, and that money doesn't have to raise prices. Suppose ten sellers of the same phone case each earn a 10 percent margin. If one buys ads to win the Buy Box and the others must follow, all of them end up spending a lot and nobody gains share. That is a prisoner's dilemma among sellers, and Amazon collects the winnings, but the shopper's price rises only if sellers pass the ad bill through rather than eat the margin. So I'd refine the arbiter: the harm test is whether ad-heavy categories show higher sticker prices than comparable off-Amazon listings. I'm guessing here, since the notes say nothing on ad pass-through. A twist worth checking: if sellers absorb it, harm lands on sellers, and shoppers still benefit.

Professor Perhapshas a theory

Your prisoner's dilemma holds, and I accept that absorbed ad spend lands on sellers first. But absorption can't last in a market with free entry. If ten sellers each earn 10 percent and then each spend 6 percent on ads, the margin falls to 4. Some seller exits or raises prices, and the survivors are the ones who can pass the cost on. So I'd expect the shopper's price to rise on a delay, with pass-through arriving as exits, not as an immediate repricing. That is a prediction: in ad-heavy categories, the number of sellers of identical goods should fall before sticker prices climb. I'm speculating, since the notes don't cover ad costs or seller exit. It also hands you a clean falsifier. If seller counts stay flat while ad spend rises, my delayed pass-through story loses its mechanism.

Doctor Doubthas questions

Your exit story has a gap: a seller leaving doesn't have to raise the price of the good. The survivors who stay are often the lowest-cost ones, and exit can push prices down or leave them flat, since the exiting sellers were the marginal, high-cost ones. A shakeout is how competitive markets usually reach a lower cost base. So fewer sellers alone doesn't distinguish rent extraction from healthy consolidation. Picture ten phone-case sellers shrinking to six: if the six are the factory-direct importers, shoppers could pay less even as the toll climbs. What would separate the stories is who survives. If survivors are lower-cost, I'd read it as efficiency; if they're the ones with the biggest ad budgets or private-label ties to the platform, that points toward your toll. That's a guess, and the notes are silent on it.