An econometric study of why people fly, across twelve markets.
Flying grows with income, roughly in step with it in the
United States, faster in Australia and far faster in Brazil.
Over the long run, a one percent rise in national income goes with about 0.9 percent more flying in the United States and about 2.8 percent more in Brazil.
The United States figure is an estimate, and it depends on how the model
is set up. Choices the data cannot settle, such as how the pandemic
quarters are treated, give figures from about 0.9 to about 1.3. Partway through the years studied, more small airlines began
reporting to the US traffic survey. The reasonable ways of allowing for
that change give figures from about 0.8 up to the figure quoted above, which uses the survey's numbers as published.
Four of the twelve markets studied
are ranked on the first chart below. Open "How strong is this evidence?" beneath
it to see how firmly each estimate holds. China is shown beside them without a rank,
and six more were tested and left off. Each market
counts passengers its own way. Singapore counts passengers passing through
its airport, and the United States counts each time a passenger boards a flight.
For that reason, markets whose figures sit close together are not ranked against
each other. Europe gets its own country-by-country charts.
Ticket prices are not in these models. After inflation, fares fell over
the same years, so where cheaper fares and rising incomes went together,
each figure includes the effect of both. Fleets, airline costs and route
networks are separate questions with their own charts below, and some of
them remain unsettled.
Every chart on this page, and every number in its text, is produced from
public data, from the models fitted to that data and, for the
projections, from scenario assumptions. No number is typed in by hand.
The United States data run through quarter 3 of 2025. 2024 is the
last full year, and 2025 is in the data only up to that quarter. Other markets state their coverage
on their own charts.
Singapore and the US are too close to rank. China, inconclusive on a shorter pre-COVID window, is shown but not ranked, and Canada is left off. Every ranked market fails at least one pre-COVID re-check of the long-run link
Data through 2024 Q4 (earliest sample ends 2019 Q4)
airtraffickonomics.com
The ranked markets are in order of how strongly air travel moves with income. For each one, at least one repeat of the test on pre-COVID years alone disagrees about that long-run link.
How strong is this evidence?
Each market carries a status label, and the glossary explains what each label means. “Confirmed” rests on the main
long-run test, which controls for the pandemic. That test is also re-run on
pre-COVID data alone, and a market that passes only with the pandemic quarters
in the sample is published as provisional. A second, different test of the same
link (Johansen) runs without those controls. Its pre-COVID check asks whether
it still passes once the quarters from the pandemic onward are removed.
Market
Status
Pre-COVID main test
Pre-COVID Johansen check
Singapore
Confirmed existence
passes
uninformative (the two samples disagree)
United States
Confirmed
passes
fails pre-COVID (full-window agreement depends in part on the pandemic quarters)
Australia
Provisional
fails (passes only with the pandemic quarters in the sample)
passes at published lag (2 of 4 lag cells clean)
Brazil
Confirmed
inconclusive (inside the test's bounds)
fails pre-COVID (full-window agreement depends in part on the pandemic quarters)
China (pre-COVID 2011–2019) (not ranked)
Inconclusive
pre-COVID by construction
pre-COVID by construction
Tested and left off the chart: United Kingdom, Canada, India, Japan, Mexico, and South Korea. The glossary gives the grounds.
The pre-COVID Johansen check qualifies the corroborating second test only;
it changes no elasticity or status. The status of each market and the
tests behind it are set out in the elasticity ladder field note.
Passengers by quarter in the airlines' traffic reports to the US Bureau of Transportation Statistics (T-100), 1990 Q1 to 2025 Q3, with a rolling average and one nowcast quarter from airport security counts (TSA)
The nowcast scales airport security counts (TSA) up to the official passenger count (BTS), which runs about 1.2 times as high. It fills in quarters the official data have not completed and is not a demand model
TSA screening counts arrive about three months ahead of the official passenger figures and move with them at a steady ratio. The nowcast applies that ratio to the quarter still being counted and is not a demand model.
Distance-fare scatter: route-level average fare vs distance, single-quarter snapshot of up to 500 routes, dot size proportional to passengers; headline is the log-log slope of the plotted routes (the full 2020-25 route panel runs flatter)
Each dot is a route, sized by its passengers, and the line is the fit to the dots. Among short and medium routes, fares at a given distance still differ several-fold.
Sum of per-market central paths, 2025–2044. Each market counts passengers its own way and a trip between two covered markets is counted in both. The fan spans income-path and population scenarios, not model uncertainty
The ten markets combined over the coming two decades, as the sum of the income model's projections and Singapore's time trend. Supply-side and network growth are left out, so this is not a full forecast.
Each bar links one route feature to the average fare, others held fixed, in a regression on 83,815 route-quarters (a 10% sample of 838,162) in 22,983 city pairs. R²=0.299 measures how closely it fits; these are associations, not causes
Each bar is how the fare moves with one route attribute when the others are held constant. Fares in this fit are also lower on busier routes and on more concentrated ones.
Route counts fall 4,209 to 990 across the distance bands, while traffic per route is 20% higher in the longest band than in the shortest; descriptive band means, not a tested effect
Low-cost carriers turn 0.56 cents profit per seat-mile against 0.58 cents for legacy carriers. Unit margin is operating revenue less operating cost per available seat-mile, each carrier counted by the seat-miles it flies
Low-cost carriers spend less per seat-mile but also earn less, and the gap between the two carrier types' pooled unit margins cannot be told apart from zero.
Next-quarter forecast error (RMSE), each model refitted quarter by quarter, for the 2017 to 2019 holdout and after COVID, in ascending order of holdout error. Models see the quarter's actual macro data, and ARDL has no post-COVID score
Before the pandemic the leading models were close together. Afterwards the seasonal model predicts best and the error-correction model worst of those scored here; each sees the quarter's actual economic data.
An aircraft has left when it is no longer filed at a year-end census under any tail number, and a conversion to freighter is not an exit; counting only aircraft last filed as withdrawn puts every median later (see caveat)
Here, regional aircraft leave US service youngest, wide-bodies oldest and narrow-bodies in between. Every median is later if only aircraft reported as withdrawn count, so the result depends on how an exit is counted.