FlightCostairline route economicsRUENDE
FlightCost › Research

← FlightCost

FlightCost · Research series prologue

What does it take to start an airline?

Spoiler: first we need to know who will want to fly, where they’ll fly from and where they’ll go.

Prologue · Where it all beginsUp next: demand factors → demographics → economics → routes → aircraft

Imagine we’ve decided to start an airline. We already have funding and a team. The next question seems practical: where should we fly, and which aircraft should we use? Before answering, we need to understand the market. Even a carefully costed route is a risky proposition without an estimate of future demand.

The plan seems straightforward: pick promising destinations, calculate operating costs, estimate fares and compare costs with revenue. We can itemize the cost of a flight down to fuel and airport charges. But there’s no technical handbook that tells us how full the aircraft will be. We need to forecast that.

Millions of people may live in two cities without filling the planes between them. Some travel by train, some connect through a hub, and others don’t travel at all. Over five years, new nonstop service, rising incomes, an airport expansion or an economic crisis could change the picture.

That’s how a thought experiment about starting an airline turned into a study of air travel demand. Choosing an aircraft type could wait. First we needed to know who would fly, where they would fly from and where they would go.

The most important question came before choosing the aircraft: how many people will want to fly, and why will they choose this route?

I’m going to count you all

A reference to the animated film “The Little Goat Who Counted to Ten” (1968). © Soyuzmultfilm Studio.

We started with a simple hypothesis: when city populations grow, passenger numbers may grow with them. Population projections are available for many countries, and the hypothesis is relatively easy to test against historical data.

We then began adding incomes, fares, departures, the effects of crises, airport development and new routes to the model. Each factor can influence the number of trips, but their relevance varies by situation.

The testing began in selected European markets. The study then expanded to 22 subregions worldwide and three target years: 2029, 2031 and 2036.

22subregions in the global passenger traffic model
3 / 5 / 10years ahead in forecasts and historical tests
45,365directions and bidirectional pairs in the route sample

It’s just as important to understand the units. A connecting passenger is counted on each flight segment, while a single trip in a survey may include several flights. Global airline statistics often group traffic by the carrier’s country of registration rather than by departure airport. Mixing these measures can produce misleading results even with a carefully built model.

What we learn from getting it wrong

It’s easy to extend a historical curve and call it a forecast. For a practical decision, we need to ask whether the model could have predicted what happened later using earlier observations.

At each starting point in the historical test, we trained the model on observations up to and including the start year and compared the forecast with what happened later. We compared a simple population-based model with a more complex one that uses economic indicators and trends in departures.

Historical test · Five-year horizon

More complex models can be less accurate
Population onlyMacroeconomics and departures

Target years 2008–2019. The more complex model had a lower error in this sample. WAPE expresses the sum of absolute errors across countries as a share of actual traffic. Lower is better.

Source: historical test data (CSV). The figures refer to an earlier version of the macro model tested over five years. They do not measure the accuracy of future route forecasts.

During a period of relatively steady growth, the more complex model had a lower error. The advantage disappeared when crisis years were included, and the macro model fell behind the population model. Accuracy in one historical period does not guarantee that a model will hold up in new conditions.

That’s why FlightCost publishes its forecasts alongside their test results, including cases where they performed poorly. In airline strategy, the size of the error matters as much as the passenger forecast itself.

Global passenger traffic doesn’t tell us where to fly

Even a strong forecast of regional traffic growth does not automatically identify a profitable nonstop route. One market may grow through large hubs, another through new routes, and a third through more frequent flights on existing services.

A pilot study of international airport pairs with one endpoint in the United States examined how route history relates to the start of nonstop service. The sample included pairs without service in the two most recent years for which data were published. The rates of first observed launches and resumptions differed substantially.

0.75%

of pairs with no nonstop service on record gained service within three years

12.64%

of pairs that had nonstop service in the past regained it within three years

These figures apply to the pilot sample and cannot tell us whether a route will be profitable. They show why we need separate forecasts for restoring an old route and launching one for the first time.

We therefore need to distinguish the total passenger forecast, the forecast of new routes and an individual airline’s decision to launch a service.

Only then can we size the fleet

Estimating route demand gives us a basis for the engineering calculations. How many flights should we operate? What capacity and range does the aircraft need? How much time will it spend in the air and on the ground? How can we offer flights often enough and maintain a viable load factor?

01 / DEMANDWho wants to fly?

Population, income, reasons for travel and demand between cities, including connecting trips.

02 / ROUTEWhich service?

Nonstop service, frequency, fares, competition and alternatives through other hubs.

03 / FLEETWhich aircraft?

Seats, range, flying hours, constraints, the existing fleet, replacements and operating costs.

Dividing passenger traffic by aircraft capacity doesn’t tell us how many new aircraft are needed. We need to account for flights and flying hours first, then for the existing fleet, its age, replacements and production constraints. The economics of each flight require a separate calculation.

From forecasting to aircraft engineering

The next question concerns the aircraft themselves: which types could serve routes where demand is expected to grow?

The segment with 200–299 seats on routes longer than 4,000 kilometers is of particular interest. In one FlightCost route analysis, at an assumed 90% load factor, it accounts for 17.1% of additional flying hours and 8.25% of additional boardings on growing routes by 2036. The segment is large enough to test whether a dedicated aircraft type could serve that demand more effectively. Determining whether existing types can meet the demand requires a separate analysis of aircraft supply and flight economics.

Engineering hypothesis, no purchase recommendationWhat could bridge the gap between narrow-body aircraft and long-haul routes?

Future work will compare modern configurations of the Boeing 757-200 and Airbus A321XLR with an extensively modernized Tu-214. On the same routes, we’ll compare range with a commercial payload, cost per seat, scheduling options and actual demand. At this stage, there is no basis for declaring one option the best.

How the FlightCost research series is organized

That’s why we started with passengers. You can choose an aircraft from a catalog, but the future market needs to be estimated, tested against historical data and reassessed as new evidence becomes available.

That led to a series of eighteen studies on demographics, income, fares, recovery after crises, new routes and future aircraft requirements.

The series will start with the research question and tests of the hypotheses, move on to regional forecasts, and then examine routes, frequencies and aircraft. Methods, tables and limitations will accompany the findings.

First step

What exactly are we trying to forecast?

In the first study, we explain how we measure passenger traffic and how population, the economy and departures enter the forecast.

The research is open for discussion. If you work in aviation, design aircraft, study transport or spot an error in our figures, join the discussion. Substantive criticism helps improve the models. Sometimes a hypothesis that fails to hold up teaches us something useful.

All 18 studies

The articles will be released in stages. Forthcoming studies are marked as planned. The CSV files below contain the data from the historical tests discussed in this prologue.

  1. No. 1 · What can we forecast in aviation?
  2. No. 2 · From demographics to demand
  3. No. 3 · Historical changes
  4. No. 4 · Coverage by departure airport
  5. No. 5 · Regional macro model
  6. No. 6 · Shocks and forecast comparisons
  7. No. 7 · Air travel affordability
  8. No. 8 · Airfares in Brazil
  9. No. 9 · Validating Brazilian surveys
  10. No. 10 · Income at 3-, 5- and 10-year horizons
  11. No. 11 · Income and access to flights
  12. No. 12 · Transferring the model across regions
  13. No. 13 · Demand in the state of departure
  14. No. 14 · Launching nonstop service
  15. No. 15 · A gravity model for routes
  16. No. 16 · Route history and the start of ticket sales
  17. No. 17 · Nonstop service and concessions
  18. No. 18 · Evaluating investment
Definitions, sources and limitations

This is a thought experiment about starting an airline. It does not announce a new carrier or constitute investment advice.

For the 22 subregions and the 2029, 2031 and 2036 horizons, see the research question and assumptions. The count of 45,365 directions and pairs and the example of flying hours for aircraft with 200–299 seats were checked against the saved calculations. The figures are in the verification CSV. The full study is scheduled for publication. Its route sample mainly covers the United States and Europe, so the shares need further validation before they can be applied globally.

The five-year errors of 23.9%/17.9% and 29.3%/34.1%, and the nonstop launch or resumption rates of 0.75%/12.64%, are documented in the country-level test observations and the route test. WAPE divides the sum of absolute errors across countries by total observed traffic. It differs from the error obtained after aggregating traffic by region.

For the global forecast, passengers are assigned to the carrier’s country of registration. The route analysis counts boardings on each flight segment. The availability of historical and future schedules and gaps in regional coverage limit how confidently we can draw conclusions.

The study examines scenarios. Passenger traffic forecasts, estimates of new nonstop services and aircraft delivery requirements rely on different data. They do not establish whether a particular airline would be profitable.

Country-level observations from the historical test (CSV) · Nonstop service launch rates by route history (CSV) · Prologue figures and sample limitations (CSV)

Scenario-based research. The calculator does not apply these results automatically.