The World Cup Effect

The Hypothesis

Before we looked at a single row of data, we had a guess, and it was the same guess most people would make. A mega-event like the World Cup brings hundreds of thousands of visitors into a handful of cities over a few short weeks. The obvious hypothesis is that short-term rental hosts in those cities would see their calendars fill up. More visitors, more nights booked, more occupancy. That was the prediction we set out to test.

If that hypothesis were true, we'd expect to see three specific things in the data:

The hypothesis didn't survive contact with the data. What we found instead was close to the mirror image of what we expected, and the rest of this report is the evidence trail that got us there.

How We Tested It

Here's the short version of the method, so you can trust what follows. The full model lives in Appendix A and Appendix D.

We compared ten U.S. host metros against seven similar cities that hosted no matches, using those seven as a control group. The closest control city sits 330 km from any host metro — far enough that the tournament in one city couldn't itself be inflating the numbers in a nearby “control.”

We measured three things independently: the rate hosts charged, the share of nights they filled, and the resulting income per listing. Each was modeled separately, with no shared assumptions, so that agreement between them would mean something rather than being built into the math.

Most importantly, we measured change within each individual listing — comparing a listing to itself before the tournament, rather than comparing one group of listings to a different group. That turned out to matter a great deal. An earlier version of this analysis skipped that step and produced a false positive (Section 14).

A Pricing Event, Not A Booking Boom

Here's what we actually found, against what the hypothesis predicted.

Rate is the solid estimate. Nights filled and income are exploratory — the nights interval touches zero. Use them for the shape of the story, not for precise magnitudes.

Three separately-modeled channels. Rate and occupancy reconcile with the independently-measured income change to within less than 1% in log terms.

• Nights filled: down about 4%, not up. The hypothesis predicted this would climb. It fell.
• Asking rate: up about 20%. The hypothesis treated this as a secondary effect. It turned out to be the entire story.
• Income per listing: up about 14%, driven almost entirely by rate, not occupancy.

This isn't a small miss on the original hypothesis; it's an inversion of it. Hosts didn't fill more nights — they charged more for the nights they were already going to fill. Three separate models point at the same conclusion.

The honest caveat: the occupancy and income figures are softer than the rate figure, since the underlying booked-nights data carries a quality issue in 2024 that touches every metro equally. The rate number is the one to trust down to the point estimate.

Every Host City Showed A Positive Lift

Every one of the ten host cities showed a positive income lift. None showed a loss.

Not a leaderboard: the intervals overlap almost completely and three of the ten still reach below zero. Section 13 explains what actually drives the spread.

It would be easy to read this as a ranking. It isn't one. The ordering here mostly reflects each city's baseline price, not how well any of them “performed.”

New York / New Jersey

+13.2%

Income per listing, vs. matched controls

Los Angeles

+8.4%

Income per listing, vs. matched controls

Miami

+12.8%

Income per listing, vs. matched controls

Where The Hypothesis Broke Down

This is the section where the original hypothesis fails outright. If the World Cup were a demand-volume event, this is the chart that should have lit up. It didn't.

Hosts raised rates ~20% and lost ~4% of nights filled. That is a description, not an elasticity — no price/demand trade-off is quantified here.

Occupancy fell in nine of the ten host metros relative to control cities, by about 6 percentage points on average. San Francisco was the only exception. A second, independent measurement — scraped booked-nights rather than scraped availability — agrees on both direction and size.

This was never a story about cities filling up. It was a story about the price of the nights that were already filling.

We're not going to tell you that means demand was “inelastic,” because we can't isolate a real elasticity here: supply, listing mix, and the event were all shifting together. What we can say plainly is that this was never a story about cities filling up.

Six Estimators, One Answer

Since the rate finding is now carrying the whole hypothesis, we stress-tested it as hard as we could before trusting it.

The tinted band marks +16% to +20% — every estimator falls inside it. This is why we call the pricing result a finding rather than a hunch.

Where A Listing Sat Mattered More Than Which City It Was In

Once we accepted that this was a pricing story, the next question was obvious: was the increase spread evenly across a host city, or concentrated somewhere specific? It was concentrated, and heavily.

Write the near-venue figure as a range: +42% (Bayesian, pooled), +35% (frequentist within-listing) and +30% (winsorised) are all defensible. Distance to downtown is flat over the same data — this is a venue effect, not a city-centre effect.

The closest ring is the most sensitive number in this report: a small number of outlier listings can swing it several points.

Which host city you were in mattered less than where in that city you were. The responsible way to state the near-venue effect is a range — roughly +30% to +42%. We'd rather give you the honest range than round up to the biggest number.

Only The Stadium Slope Is Distinguishable From Zero

Before accepting “proximity to the stadium” as the explanation, we had to rule out a more boring one: maybe this was never about the stadium at all, just about being downtown. Several venues sit well outside city centres — AT&T Stadium is in Arlington, Gillette in Foxborough, MetLife in East Rutherford — so the two are easy to confuse.

The stadium interval sits entirely left of zero; both downtown intervals straddle it. There is a 99.8% probability the stadium slope is genuinely steeper than either downtown comparison.

Distance to the stadium came back clearly negative. Distance to downtown didn't — in host cities or in the control cities that have no venue at all. The alternative explanation didn't hold up. This really is a venue effect.

Every City Has Its Own Halo

The pooled curve in Section 7 averages across ten cities whose rental inventory sits in very different places relative to their venues. Four things are worth knowing before reading your own city's numbers.

Treat the pooled curve in Section 7 as the number to trust. City-level variation is for understanding your own inventory, not for building a ranking — the ordering mostly reflects baseline price and where inventory happens to sit.

The Premium Arrived And Left With The Match Calendar

If proximity was the “where,” this is the “when.” The premium wasn't constant — it rose and fell with the match schedule itself.

This is anticipation in pricing for future stay-dates, not evidence about when guests booked. Booking dates are not observable here.

Intervals (94%) span roughly ±13 points on each week; weeks −6 to −2 include zero, weeks −1 through 5 do not.

There's a measurable run-up before the first match: rates for future stay-dates were already climbing weeks in advance. That's anticipation baked into pricing — a sign this was priced in ahead of time, not a surprise that hit all at once.

The Premium Was Rented, Not Owned

This is the most decision-relevant chart in the report, because it shows location and timing interacting rather than acting on their own.

Going from zero to two matches in a metro-week widens the near-versus-far gap by roughly 17 percentage points. Proximity paid most in the weeks the venue actually hosted matches.

A within-metro spatial comparison, not a control difference-in-differences: it says how much more a listing at 2.9 km gained than one at 52.8 km in the same week, not the absolute lift.

Being close to a venue was worth roughly three times more in the heaviest match weeks than in the quiet ones. Proximity was rented, not owned.

It wasn't a fixed asset that paid off evenly for six weeks — the halo expanded and contracted with the match calendar.

One caveat that makes this conservative rather than inflated: because several venues sit outside their city centres, “beyond 30 km” is sometimes a city's best neighbourhood. In Boston, 82% of listings sit beyond 30 km from Gillette Stadium. There, the comparison is really “near the stadium versus the best part of town,” which understates the effect if anything.

What You Listed Mattered Far Less Than Where It Was

Two more predictions from the original hypothesis didn't hold up, and both are worth stating plainly.

Studios, private rooms, and five-bedroom homes all moved within about 10 percentage points of each other.

We expected more matches to mean a bigger city-level lift. We couldn't detect that relationship at all. Every measure of match volume produced an interval spanning zero, and the sign flipped depending on the method. That means we couldn't detect an effect — not that there definitely isn't one. Within the tournament, heavier match weeks clearly saw bigger lifts (Section 11); across only ten cities there simply isn't enough variation in match count to detect a city-level relationship, even a modest real one.

Expect A Dollar Premium, Not A Percentage

Match volume explained none of the spread between cities. Something else explained most of it.

The event added roughly a fixed dollar amount per night — about +$55, varying only ~$9 between cities. A fixed dollar premium is a big percentage in a cheap market and a small one in an expensive one. Baseline price explains about 80% of the between-city variance; match volume explains none of it.

Do not build a league table from this: the per-city intervals overlap almost completely, and the ordering reflects baseline price, not performance. Spearman(baseline, % lift) = −0.83.

Cheaper cities showed bigger percentage lifts. Los Angeles looks like a modest performer and Philadelphia like a strong one, but they made similar money on very different starting prices. This is the most useful mechanism in the report for a host: expect a dollar premium, not a percentage. What percentage you personally see depends entirely on what you already charge.

A Year With No Tournament Returns Essentially Nothing

Before you believe any of the above, the method itself has to survive a null test. We re-ran the entire analysis pretending 2024 was the “treatment year” against a 2023 baseline — two ordinary summers, no tournament. A sound design should return approximately zero.

An earlier version of this study, run on a smaller set of internally-managed listings, failed this same test — returning a false +16% “effect” in a year with no World Cup.

That earlier failure is why the analysis looks the way it does. A small portfolio of listings churns between years, so a simple metro-level average ends up comparing different properties to each other rather than the same properties over time. The fix was measuring change within each individual listing, across a full market-wide dataset. Every number in this report comes from that corrected, tested version.

What This Data Does Not Support

We never observe willingness to pay in this data. We observe posted rates on nights that sold. That number bundles genuine guest demand together with however hosts and their pricing tools responded, and the two can't be separated here.

The distance gradient, timing pattern, and metro-level lifts are treated as causal, conditional on three things holding: parallel pre-tournament trends (tested, Section 14), no spillover into control metros (checked — the nearest is 330 km away), and no other 2026-specific shock hitting host metros differently (not tested, see Appendix C).

Three Things To Carry Away