Forecasting avg. flight occupancy I have about 15 month time series data of average flight occupancy. I wish to forecast how the average flight occupancy would look like 1 year from now. There are alot of work done on this it seems (Example: https://www.icao.int/Newsroom/Pages/Latest-air-traffic-forecasts-illustrate-encouraging-recovery-and-higher-growth-in-global-air-travel.aspx)

*

*How can I approach this?

*can I assume what happen in past to repeat in future as well, basically past average occupancy to repeat?

*how the recovery of market can be taken into account?

*how leap year to be taken into account? given this mean 366 days instead of 365?

*How can we take into account special days like Christmas where people travel more?

Any pointers?
 A: It sounds like you have daily data, based on your asking about leap years and Christmas.

*

*A sensible baseline is a time series model. For daily data, you may have multiple-seasonalities, specifically intra-weekly and intra-yearly. The tag wiki contains pointers to relevant algorithms and papers. Then again, the seasonal pattern in occupancy may not be as strong as in passenger numbers, since airlines try to keep their planes full and schedule more during high season. So I would also use a simpler method as a benchmark.


*You can assume this, and a time series method will do so. Whether this is a reasonable assumption depends on what you believe will happen in the future. What will happen to the economy? Will we have constraints on travel due to COVID again?


*I assume experts in this field use a lot of judgmental input and modify time series forecasts. They probably prepare multiple scenario forecasts, one for a strong recovery, one for a smaller one, one for a recession, to help decision makers.


*Your time series method should account for this. To be honest, given the current uncertainties, worrying about a single leap day does not seem like the best use for an analyst's time.


*Special days that are fixed in time, like Christmas, should be captured automatically by your time series algorithm. Days that shift in time like Easter or Labor Day may require something like a regression of your time series on dummies, followed by a model applied to residuals.
There are a couple of relevant articles in the International Journal of Forecasting, none of them recent, though. You may also be able to find some relevant literature by searching for "tourism forecasting" or similar.
