3.4. Step 4. Prediction comparison

The last step is to apply the best‐performing forecasting model to predict demand for each passenger segment in each scenario. We first obtain the prediction in the business as usual scenario, using historical demand for each segment. Comparing the prediction in the business as usual scenario with the real situation, we derive the twofold impact of COVID‐19 on demand for passenger air transport, assuming that passengers will follow their behavior pattern if there was no pandemic. The time series prediction uses univariate flight data and the availability of flight routes is not considered. To obtain the prediction in the pandemic scenario, we first identify passengers' behavior pattern in terms of flight route choice and then adjust the prediction in the business as usual scenario, considering the impact of each flight route restriction on demand. Not every flight route restriction has the same impact on demand. Restrictions on popular routes result in severer impact than restrictions on less‐popular routes. In addition, restriction on a specific flight route may impact demand for different passenger segments differently because not every segment will fly on the route with the same frequency.

The flight route choices of each passenger segment can be identified by calculating the frequency at which the segment flies on each flight route, using historical data. In the pandemic scenario, the impact of a route restriction on demand for a segment considers both the availability of the route and the previous frequency at which the segment flew on this route. For example, if a segment flew from AMS to BRU, 30% of the time previously and in the pandemic period flights on this route were canceled 90% compared to the previous schedule, then the impact of this travel restriction on demand for this passenger segment, measured in the number of flights reduced, will be equivalent to 30%×90% of the demand forecast in the business as usual scenario. The prediction in the pandemic scenario is given by Equation (2):

where Fsp is the forecast on the number of flights for passenger segment s in the pandemic scenario, Fsb is the forecast on the number of flights for passenger segment s in the business as usual scenario, Ls is the number of flight routes on which passenger segment s has flown previously, Ar∣s is the number of flights on route r passenger segment s has flown previously, As is the total number of flights passenger segment s has flown, and DrNr calculates the availability of flight route r in the pandemic period (see Step 3 of the method).

The impact of a flight route restriction on demand for a passenger segment can also be evaluated by directly multiplying the availability of this route with the demand forecast for this route. However, it will then require forecasting to be done on the flight route level, that is, demand forecasting for each flight route for each passenger segment. The potential disadvantage of this approach is that there may be a lack of time series data on the flight route level within a passenger segment, and thus the forecast may not be accurate. By aggregating the flight data on different flight routes, generating an overall demand forecast for the entire set of flight routes, and considering the weight of each flight route in the demand forecast, we maintain good accuracy of the forecasting model and extend the applicability of our approach to a broad range of industries.

In the pandemic scenario, the impact of the availability of flights on demand is considered. Comparing the prediction in the business as usual scenario with the prediction in the pandemic scenario, we derive the impact of COVID‐19 associated with supply restriction. Comparing the prediction in the pandemic scenario with the real situation, if there is a lower number of flights in the real situation, it can be attributed to a low willingness to fly. Therefore, the difference between the prediction in the pandemic scenario and the actual number of flights is the impact of COVID‐19 associated with demand depression.