The Challenge: Predicting Demand for Air Traffic in a Highly Disruptive Environment
The current Corona pandemic has devastated many industries, but only a few with the vigor that struck the aviation sector. Under the impression of international travel bans and national lock-downs, customers were unable or unwilling to travel anymore. Thus, airlines were forced to ground the vast majority of their fleets. While this was extremely painful, the upcoming re-launch of flights will be even more challenging, since no airline has ever been in a situation like this before.
It will be highly seductive for the ailing airline community to reactivate idle capacities as soon as the lock-down is released, and demand for air transport seems to return. However, being too quick in this race may ultimately be lethal, because ramping-up flight operations after the shut-down unavoidably boosts operational costs which need to be covered by respective revenues – a task that is intimidating in the light of the frenetic volatility of the current situation.
Well understood, forecasting has always been a crucial function throughout the entire process flow of network management, since airlines need to allocate very expensive assets under significant uncertainty. Furthermore, it is without any doubt valuable to identify market opportunities earlier than competitors. However, these statements have become even more valid in disruptive scenarios such as the current Corona pandemic.
In this critical situation, demand forecasting becomes a make-or-break capability for airlines
The Status Quo: Performance of Current Demand Forecasting Models
Unfortunately, the current generation of demand forecasting models is close to the end of its technological life cycle.
Four major weak spots have become particularly disadvantageous in times of an unprecedented upheaval
- Traditional forecasting models depend on macroeconomic trend information, but even more on historic and current booking data from sources such as Sabre’s well-known MIDT (Marketing Information Data Tapes) service. MIDT data are based on actual transactions reported by Global Distribution Systems (GDS). If this input is unavailable or invalid (e.g. for new routes or in case of disruptive events), the results are almost useless. Furthermore, the global coverage of the MIDT samples has eroded from historically 80 % to currently below 30 %, since low cost airlines try to avoid the expensive GDS channel. Still a solid sample size, but much less valid than 15 or 20 years ago!
- Although transaction data such as the MIDT sample are O&D data, they reflect sold tickets, so they represent an airport-to-airport view — real demand occurs door-to-door (‘true O&D’). For clients living close to a major airport, that may be a minor distortion, but there are travelers living far away from any airport, or having the choice between various airports
- Current forecasting models focus on demand for air transportation, which may be a constrained demand if no appropriate flight exists on a certain O&D. Unconstrained demand contains all other transportation modes as well, particularly on short- and mid-haul O&Ds. Ignoring intermodal competition may at the same time over- and underestimate the real demand for air travel, and is in any case a distortion
- Current models basically assume an inside-out view, looking at the world from the perspective of a single airline’s hub. This view may lead to blind spots with respect to competing hubs, underestimating their bypassing potential
Furthermore, traditional forecasting models are designed for a world of incremental changes, not for the extremely disruptive scenario that we witness right now. The reactivation of 40, 60, or 80 % of the initial capacity within a few weeks or months is by no means incremental, but the most fundamental relaunch in the history of aviation.\nEven worse, traditional data sources are corrupt or useless in this specific environment: historic O&D demand is invalid for the current phase, booking data are outdated, and external sources such as requests on meta-search engines include a massive bias, since travelers will not look for flights – even if they could and would like to travel – if they believe that flights are not available or not safe. And last, but not least, national governments are forced to frequently adapt their travel bans to the current pandemic situation (e.g., the most recent U.S. restrictions for Brazilian travelers), which adds another element of uncertainty.
Consequently, the traditional demand forecasting models seem to be poorly prepared to answer the most pressing question for airlines in the current setting:
At every point in time – with or without a disruptive change – this question can be broken down into two sub-questions:
- Overall demand: how many passengers will in total use air transportation on a certain O&D or route?
- Airline-specific demand: which share of the overall demand will choose a certain airline offering flights on the respective routes?
The second sub-question, i.e. the traffic share of a certain airline, is a function of the attractiveness of this airline’s offering compared to the services of all relevant competitors on the same route. This share is usually predicted with the help of so-called logit models – a mathematical approach that is quite robust even under high volatility. Opposed to this, the first sub-question is the critical one in times of radical shut-down and ramp-up, since the weak spots described above become fully effective.
The Remedy: Next Generation Airline Demand Forecasting Models
Airlines try to mitigate the weaknesses described above by intensive re-calibration of their models, as well as by applying various correction factors. However, the current generation of models leaves room for improvement, and has particular shortcomings in case of non-incremental changes.
In contrast, next generation forecasting models emerging right now immanently address these weak spots by:
- Less dependence on historic and actual booking data (e.g., from MIDT), instead consistent and comprehensive modelling of needs to travel and readiness to travel based on macroeconomic relations between countries and regions worldwide
- Door-to-door perspective (‘true O&D’) instead of airport-to-airport (as supported by airport statistics and MIDT data)
- Unconstrained demand (incl. intermodal competition) instead of constrained demand (air traffic only)
- Comprehensive, outside-in view on traffic flows and competition between airlines and hubs instead of inside-out perspective of individual airlines (often underestimating bypassing potential of competing hubs)
Well understood, new demand forecasting models also struggle with the unprecedented volatility of the current crisis; however, their philosophy makes them more resilient against non-incremental changes than traditional forecasting models. Their lacking dependence on historic booking data, as well as their holistic view on external factors, enables them to include all available information on pandemic dynamics like any other macroeconomic driver (or constraint) shaping global air travel demand.
New ways of demand forecasting are of the essence to master the enormous risk posed to airlines by the upcoming operational ramp-up. Traditional demand forecasting models are badly prepared for the radical changes ahead and have reached the end of their lifetimes anyways.
Exploring the benefits of next generation forecasting models now comes at a cost; however, it may kill two birds with one stone: first, the new models are much more robust and flexible in facing the challenges of the upcoming ramp-up, and second, they fix traditional weak spots of the current methods. Having a more accurate view on demand over the next few months, implicitly adapted for COVID-19 effects (by travel reason and customer segment), will in any case pay off also short-term, since it will ensure the best possible match between returning demand and reactivated capacities.