No model saw the COVID trough coming; SARIMAX came closest, and the rest caught almost none of the fall
View data table
SARIMAX follows the fall late: its lowest forecast comes two quarters after the actual trough. The other models dip later still or not at all.
descriptive Forecast protocols differ across models, so the figures are comparable within a model and not across models.
Details
Window: 2019 Q1 to 2025 Q3
Data through: 2025 Q3
Vertical axis: log(passengers)
Series: Actual; ElasticNet; LightGBM; Neural Ensemble; NeuralProphet; SARIMAX; XGBoost
Latest: Actual 19.48; ElasticNet 19.58; LightGBM 19.32; Neural Ensemble 19.58; NeuralProphet 19.54; SARIMAX 19.56; XGBoost 19.26
The title measures each model by the share of the fall in actual log passengers from 2019 Q4 to 2020 Q2 that its 2020 Q2 forecast captured. 'Saw it coming' means at least one half, and 'almost none' means under one tenth.
Canada's non-US international traffic is 4% above 2019 while domestic remains 8% below
View data table
Canada's non-US international traffic has passed its pre-COVID level while domestic and US transborder flying still lag, leaving the total slightly short of it.
Details
Window: 2017 to 2024
Data through: 2024
Vertical axis: Index (2019 = 100)
Series: Total; Domestic; US transborder; Other international
Latest: Total 96.2; Domestic 92.1; US transborder 98.8; Other international 104.4
COVID recovery by international region
View data table
Latin America is well above its pre-COVID level, while transpacific travel is still well short of its own and is the only region yet to recover.
Details
Window: 2017 to 2024
Data through: 2024
Vertical axis: Index (2019 = 100)
Series: Canada; Latin America & Caribbean; Transatlantic; Transpacific
Latest: Canada 100.2; Latin America & Caribbean 119; Transatlantic 104.4; Transpacific 81.8
COVID recovery by route type
View data table
Compared quarter for quarter with the last pre-COVID year, routes between two hubs carry about the same traffic and the other two kinds more. Point-to-point routes recovered last and have led only in the latest year.
Details
Window: 2019 Q1 to 2025 Q3
Data through: 2025 Q3
Vertical axis: Index (2019 = 100)
Series: Hub–Hub; Hub–Spoke; Point-to-Point
Latest: Hub–Hub 103.4; Hub–Spoke 113; Point-to-Point 110.7
Brazilian air travel fell 7.5% from 2015 to 2016 and regained its 2015 level in 2018
View data table
On full-year totals the fall took Brazilian air travel back to about where it stood four years earlier. Only part of the loss came back the following year, and a year later the total edged past the one before the fall.
Details
Window: 2012 Q4 to 2019 Q4
Data through: 2019 Q4
Vertical axis: Passengers in the latest four quarters (2015 total = 100)
Latest: 101.3
Domestic capacity fell 7.8% and domestic passengers fell 6.7% from the first two quarters of 2001 to the last two, and all 54 routes on the top-routes list kept service
View data table
Australian domestic seats and passengers both fell in the second half of the year Ansett collapsed, against a season that normally lifts them. The September attacks came days earlier, so the two cannot be separated here.
Details
Window: 1999 Q1 to 2004 Q4
Vertical axis: Index (1999 Q1 = 100)
Series: Domestic passengers (index); Domestic available seat-km (ASK, index)
Latest: Domestic passengers (index) 145.8; Domestic available seat-km (ASK, index) 152
The comparison is not seasonally adjusted: in the window's other complete years the second half ran on average 12.0% above the first in domestic passengers and 10.6% above in seat-kilometres, season and trend together, so the 2001 changes run against the usual pattern. The second half of 2001 also contains 11 September, 3 days before Ansett's collapse, so the figures describe the change between the halves, not what the collapse alone removed.
After COVID, the seasonal pattern was out of its pre-COVID range in 2022 and within it in 2023 and 2024
View data table
The post-COVID bars average three years. The first, when the winter quarter was unusually weak, fell outside the range expected from the pre-COVID years; the next two fell inside it.
descriptive A pattern in the data, not a causal estimate.
Details
Vertical axis: Quarter's traffic against its year's average (1.0 = average)
Series: Average of 2015 to 2019 (pre-COVID); Average of 2022 to 2024 (post-COVID)
Each post-COVID year is checked against the pre-COVID range (2015 to 2019): the interval that a further year like those would fall inside 95% of the time, on the first-quarter ratio and the Q3-Q1 swing. In 2022 the first-quarter ratio is below the range and the Q3-Q1 swing is above the range; 2023 and 2024 are inside on both. On the narrower span from the lowest to the highest of the pre-COVID years' own values, 2023 on the first-quarter ratio and 2024 on the Q3-Q1 swing also fall outside. The post-COVID bars average all 3 years, 2022 included.
The break test compares the whole seasonal regression, trend included, before and after the break, so its rejection of stability does not by itself say the seasonal pattern is still shifted.