Ghostbusters: Hunting abnormal flights in Europe during COVID-19.

The impact of the COVID-19 pandemic is unprecedented in airline history, with irregular flight bans, the inability for accurate demand estimation, several turns in the epidemiological evolution, and a wide range of downstream effects on all aviation stakeholders. While most airlines have increasingly entered a recovery stage, compared to the utmost disruption around April 2020, the airline business is far from back-to-normality. Throughout the past two years, recurrent statements have been made regarding the existence of so-called ghost flights, where airlines operate nearly empty aircraft on markets with insufficient demand, partially with the aim to avoid losing precious airport slots. This study investigates the extent of such abnormal market service during the COVID-19 pandemic through an explorative, data-driven analysis, based on actual load factor data of European airlines for the years 2017 to 2021. We break down the observed deviations by airlines, markets, and airports. We find that low-cost carriers are most-likely to have performed abnormal flights during the pandemic; and that abnormal flights have mostly occurred on frequently-served markets. In addition, we show that airline responses, in terms of departure and yield changes, are largely heterogeneous across the 24 airlines in this study. Our study is the first one to shed light on the important issue of load factor deviations, and we hope that our findings can contribute to a better understanding of the existence of abnormal flights during the pandemic, as well as deriving appropriate policies for dealing with the ubiquitous threat and impact of ghost flights in the future.

Introduction

The COVID-19 pandemic has caused inconceivable disruptions to all parts of our life. The aviation industry was hit particularly hard during the transformation from an epidemic outbreak to a global pandemic, around March 2020 to May 2020. Several existing studies have highlighted the tremendous impact of COVID-19 in this period, see Sun et al. (2021c) for a survey on the early impacts of COVID-19 on aviation. Among others, the largest impact on airlines was caused by a combination of three factors, which caused a major disruption to all aviation stakeholders. The first factor is a previously inconceivable degree of flight bans and aircraft groundings, which prevented airlines from medium-/long-term planning, a critical component of their business models. These flight bans, together with highly volatile demands led to an excessive number of aircraft groundings. The second factor is centered around the inability of most airlines to identify the actual demand during the COVID-19 pandemic, given that all the well-understood historical models suddenly were rather useless in terms of making accurate predictions. Without matching the demand and supply well, airlines faced tremendous operational and financial difficulties, particularly during the year 2020. The third factor is related to the evolution of the pandemic and the changing aviation environment, particularly due to emerging variants of concerns. The Omicron variant (Karim and Karim, 2021) is only the most recent player in this evolution, which might have a significant long-term impact on how we look at the COVID-19 pandemic. Finally, these three factors had severe effects not only on airlines, but also on all other stakeholders in the aviation value chain, including airports, aircraft manufacturers, and passengers.

Throughout the past two years, a recurrent topic in the news and popular literature concerns the existence of so-called ghost flights.1 2 3 Such ghost flights are presumably best characterized by load factors significantly lower than what would be required for an airline to operate the flight in a profitable way. In other words, if the number of revenue passengers is not large enough to cover the airline’s cost for operating a flight, then the airline will lose money from such flights (Janic, 2003). Notably, operating unprofitable flights is not a phenomenon exclusive to the COVID-19 pandemic. There are various motivations for airlines to operate unprofitable flights, several of which can be summarized as follows. First, airlines might operate unprofitable flights for the sake of keeping competitors out of a market, possibly through subsidies of various forms by governments or by cross-subsidization through highly-profitable flights. Second, a significant imbalance between outbound and inbound load factors might make airlines fly one direction under an unprofitable setup. Third, airlines might operate unprofitable flights in need to keep their slots at the busiest airports. Without serving a large-enough fraction of their slots, airlines will lose these slots to competitors. Given the eponymous grandfather rule (Grether et al., 1981), which states that airlines can keep their slots under a set of well-defined rules, airlines have huge incentives to keep their own slots, unless there is a major shift in the airlines’ business model or target market. While the system constructed around the grandfather rule has raised significant concerns from some parts of the aviation industry and from the research community, especially its effectiveness with respect to containing airport congestion and the associated tradeoff with potential reduction of airline competition, e.g., Zhang and Czerny, 2012, Gillen et al., 2016 and Dixit and Jakhar (2021), to date there is no significant change of the system in sight. In addition, there are requirements on the number of takeoffs for pilots to remain certified for specific aircraft types. The satisfaction of these requirements can lead to a certain extent of ghost flights as well. It should be noted that ghost-like operations were not exclusive to aviation. Operators of other transportation modes, e.g., railway, were operating nearly-empty services throughout some periods during the COVID-19 pandemic.

We discuss the second factor for unprofitable flights in context of the COVID-19 pandemic leading to the motivation for our analysis. Given the extensive disruptions of the aviation industry and concerns regarding the existence of ghost flights, the European Commission considered or communicated the lifting of slot rules on various occasions; see European Commission (2021) for announcements and rationale as described by the European Commission and see Fig. 1 for an overview. Specifically, it announced the waivers on March 30th, 2020, until October 24th, 2020 then extending the waiver until March 27th, 2021. From March 28th, 2021 to October 30th, 2021 another slot relief program was launched by the European Commission, stating that airlines would get a full slot series waiver for up to 50% of slot series they hold at an airport, provided that airlines returned 50% of their slot series to the slot coordinators for reallocation before the start of the season. The remaining slot series need to be operated at a 50% utilization rate. The suspension of slot rules is not unprecedented in aviation: similar measures were taken in earlier disruptions, e.g., after the 9/11 terrorist attacks in the United States, during earlier disease outbreaks, and during the global financial crisis in 2008. During the ongoing COVID-19 pandemic though, the discussion on airport slots led to a rather bizarre situation. Several larger airlines argued that they would need to perform a significant number of ghost flights if the slot rules were not waived. Such announcements were followed by a public outcry, especially supported by environmental protection groups (e.g., the ’flight shame’ movement) and related politicians, arguing that such ghost flights are contrary to all existing efforts to cut greenhouse gas emissions and turn aviation into a more sustainable transportation means.

Fig. 1

Evolution of the number of worldwide commercial flights between the years 2019 and 2021 with annotations regarding slot waiver/relief decisions by the European Commission.

In this study, we investigate the impact of the COVID-19 pandemic on the load factors observed in European aviation system during the COVID-19 pandemic from the perspective of an explorative, data-driven analysis. Specifically, we aim to find out how the load factor of airlines changed throughout the pandemic, see Fig. 2 for a visualization regarding the region of interest. Taking pre-COVID-19 load factors of the 24 busiest European airlines as baseline, we evaluate their operational performance in the years 2020 and 2021. We define the notion of an abnormal flight based on a statistical measure over historical airline data.4 Based on the formal definition of an abnormal flight, we investigate system-wide abnormal flights as well as airline-specific abnormal flights and show their evolution throughout the pandemic. While our major focus is on the evolution of load factors, we connect this measure with other indicators, including airline types, departure counts, and yields, among others.

Fig. 2

Overview on airports in this study, together with slot level denotation, indicating the degree of slot control, while highlighting of the top-20 airports according to passenger volume pre-COVID-19 pandemic with their IATA codes. The size of circles corresponds to the number of passengers in the year 2019. Slot levels are defined as follows: Level 1 for uncoordinated airports, Level 2 for schedules facilitated airports, and Level 3 for all coordinated airports.

The remainder of this study is structured as follows. Section 2 provides a review on the related literature. In Section 3, we present our methodology used in this study. Section 4 introduces the dataset and reports the results of our data-driven analysis on 24 European airlines. We conclude this study with Section 5, summarizing our findings, discussing policy directions, and suggest several interesting directions for future work.

Literature review

While we are not aware of any study in the scientific literature that analyzes the subject of abnormal/ghost flights during the COVID-19 pandemic, there are a few other related studies which provide relevant scientific background and support for our work. Such studies comprise the recent literature of COVID-19 pandemic impact on aviation, earlier research on air transportation/airline networks, and research on slot-constrained airport management. We discuss these three related categories in detail below.

Research regarding COVID-19 impact on aviation: The extensive impact of the COVID-19 pandemic on the aviation industry has led to many scientific studies, which could be described as a paper hurricane; see Forsyth et al., 2020, Serrano and Kazda, 2020, Rothengatter et al., 2021, and Sun et al. (2021c) for recent overviews and reviews on that subject. Given the extent of the literature, we only revisit closely related studies here. Albers and Rundshagen (2020) derive a classification of airline reactions throughout the COVID-19 pandemic, leading to the following categories: Retrenchment, Persevering, Innovating, Exit, and Resume. Andreana et al. (2021) use Seasonal ARIMA econometric analysis for time series analysis and highlights that the impact of the COVID-19 pandemic is larger than the impact in any earlier crisis witnessed by aviation. Bombelli (2020), using techniques from network science, analyze on the role of integrators/freight airlines during the pandemic. Cui et al. (2021) decomposes the impacts of COVID-19 on the Chinese transportation sectors. Hrubỳ (2021) use multi-criteria decision analysis techniques to compare the financial impact across twelve airlines, identifying strong changes in airline profitability. Kuznetsova (2021) analyze the impact of COVID-19 on the Russian Federation aviation system using time series analysis, highlighting the shift towards domestic air transportation. Li et al. (2021) conduct a spatio-temporal variation of global air transportation networks, dissected by regions and time periods, mainly using time series and complex network techniques. Ng et al. (2022) reported on the performance of the Japanese aviation market based on yield analysis of airlines, emphasizing the differences between traditional full-service airlines and low-cost carriers. Suau-Sanchez et al. (2021) investigate the evolution in supply and demand during the pandemic, as well as fleet status changes, highlighting a reduced impact on domestic markets, compared to international markets. Similarly, Sun et al. (2020) use analysis tools from complex network science to explore the impact of COVID-19 on aviation networks, at different levels of fractality. Sun et al. (2022) provide an overview on how airline startups evolved under pandemic ramifications and what policy implications are required to ensure safe operations of these new airlines. Suzumura et al. (2020) analyze the number of flights during the first few months of the COVID-19 pandemic as time series, identifying critical phase transition points. Zhang et al. (2021) find that low-cost carrier networks increasingly overlap under the pandemic impact, which leads to increased competition during the COVID-19 pandemic.

Research regarding airline networks through complex networks and time series analysis: Prior to the onset of the COVID-19 pandemic, a wide range of studies have used techniques from the network science domain and time series analysis methods for analyzing air transportation networks at various scales, resolutions, and time periods, see Paleari et al., 2010, Zanin et al., 2018, and Wandelt et al. (2019) for overviews and surveys on that subject. Bagler (2008) was among the first to use complex network techniques to analyze air transportation systems, particularly the Indian domestic network. da Rocha (2009) reports the complex network characteristics for the Brazilian domestic airport network. Chi et al. (2003) reveal that the domestic airport network of the United States has small-world characteristics. Jia et al. (2014) and Lin and Ban (2014), independently from each other, analyze the temporal evolution of the United States domestic air transportation system and how the complex network facets changed over time. Neal (2013) assesses the United States business airport network for the years 1993 to 2011 at various levels of fractality. Sun et al. (2015) report on the temporal evolution analysis of the European air transportation system from two distinct perspectives, comparing the air navigation route network with the airport network. Zanin (2015), among others, investigate a multi-layer representation of the European airport network.

Research regarding airport congestion and capacity management: During the past decades, the tremendous growth in aviation demand was matched by limited support in terms of supply, mostly due to under-development of airport infrastructures around the world. The incurred congestions and delays are critical challenges to address and, accordingly, this topic has received much attention in the scientific literature, see Daniel, 1995, Daniel and Pahwa, 2000, Brueckner, 2002, Pels and Verhoef, 2004, Morrison and Winston, 2007, Zhang and Czerny, 2012, Gillen et al., 2016, and Dixit and Jakhar (2021) for overviews and surveys on this subject. We refer to selected studies here, with a strong emphasis on slot policies. We distinguish two types of studies here: (a) studies which are focused on non-pandemic contexts and (b) studies which have a strong emphasis on COVID-19.

Airport slot policies pre-COVID-19: Since the first release of so-called High Density Rules in the United States more than 50 years ago, several studies have investigated the rationale and usage of slot policies; see Czerny et al. (2016) for an excellent overview on airport congestion and slot-allocation schemes; and particularly the chapter by Gillen (2016) for a primer on the concept and history of airport slots, Ulrich (2016) for an overview on the present slots rules, Forsyth (2016) for an discussion on policy-related issues, and Bauer (2016) for a discussion on the effective usage of slots by airlines. Madas and Zografos (2008) emphasizes that the primary policy concern around airport slots is in the compatibility of slot allocation strategies in context of various airport settings. Based on a multi-criteria evaluation framework, the authors provide a list of policy recommendations for European airports. Sieg (2010) compared the use-it-or-lose-slot rule against an unrestricted slot-ownership plan. It is found that the former are profitable for airports, but decrease airline profits and social welfare. Fukui (2010) finds that is that the high-density rule for airports can be beneficial, but is subject to improvements, given that the number of slots for new entrants is usually too small for the entrants to compete effectively. Ball et al. (2018) highlights the opportunity for market mechanism as a control measure, and specifically the conduction of periodic combinatorial slot auctions for time-limited leases based on the case of the United States. Sheng et al. (2019) model the effects of slot-hoarding behaviors, finding that airlines will only hoard slots when the demand/capacity ratio is low. Czerny and Lang (2019), based on simulations over a stylized model, finds that equilibrium policies involve slots once airport profits do not matter and pricing policies when airport profits do matter (see also Brueckner (2009) and Basso and Zhang (2010) for related results). Lang and Czerny (2022b) use a case study to identify the distinct contributions of local and non-local passengers for local welfare-maximizing airport congestion policies. Czerny (2020) performed numerical simulations on a case study, showing that use-it-or-lose it requirements should be decided based on a specific case, depending on the degree of competition and available airport capacity. Lang and Czerny (2022a) report on the impact of substitute destination choices on equilibrium slot quantities, aiming for maximization of welfare of all airport regions. Ali and Sharunova (2022) investigate the durations for which airlines are meeting the minimum requirements, using exogenous removal of slot control at the Newark Airport in the year 2016, attempting to find the transition between policy intention and policy violation. Cavusoglu (2022) compares different airport slot allocation approaches and proposes a new multi-criteria slot allocation auction model.

Airport slot policies during the presence of COVID-19:Truxal (2020) provides a discussion of Article 10(4) of Regulation 95/93, which governs European airports, in context of the COVID-19 pandemic onset. It is stated that the announcement of the COVID-19 slot rules, on March 13th, 2020, is good news for airlines, and several airlines had immediately started to cancel some of their flights. However, it is also argued that the adaption of slot rules is only one facet of the problem, cancellation regulations are a natural follow-up hurdle of this problem. Relatedly, Ariane (2020) highlights the importance of temporarily neutralizing the ’use it or lose it’ rule, with the major goal of giving airlines a legal certainty. Haanappel (2020) provides an early discussion of the slot allocation procedures under the COVID-19 pandemic and also provides an outlook to possibly slot allocation scenarios post-COVID-19. European Airport Coordinators Association (2020) discusses the possible extension of the use-it-or-lose-it slot rule extension for the Winter 2020/2021 schedule season. The member argue that the time between slot cancellation and day of operation should be at least four weeks and highlight the need to discourage the intentional non-use of slots, specifically in context of long-term strategic changes in the planning of airlines. Air Traffic Controllers European Union Coordination (2020) emphasize that the original purpose of slot waiver programs should re-evaluated throughout the pandemic, in order to ensure a free movement and a high level of consumer protection. Without such continuous reevaluation, airlines might see incentives to rather arbitrary last-minute cancellations without explanations and alternatives. Maughan (2021) explores the intrinsic difficulties for making informed decisions concerning slot relief programs, covering pre-COVID-19 disputes and potentially over-optimistic forecasts for recovery. In addition, the ability to form a united voice inside the Worldwide Airport Slot Board during the COVID-19 pandemic is highlighted as one of the major positive takeaways. Palinckx and Kemper (2021) discusses the introduction of a Level 4 slot, which would be targeted specifically for saturated airports, which are presumably the first to recover towards the end of the COVID-19 pandemic. Centre For Aviation (2022) focuses on the conflicting interests of between traditional carriers and low-cost airlines concerning the extensions of slot waiver programs, particularly in the context of public threats as those by Lufthansa’s CEO Carten Spohr, making claims in the magnitude of 18,000 unused flights in the sky if the slot waiver programs are not extended.

Methodology

The main goal of our research is to identify abnormal flights during the COVID-19 pandemic. However, to the best of our knowledge, there is no formal definition of abnormal flights in the scientific literature, let alone a definition of ghost flights5. The reasons for an absence of a formal definition are presumably twofold. First, these ghost flights have become more of a common public term for oversimplifying an observed or conjectured phenomenon in air transportation. Second, it is rather hard to develop a binary (yes/no) decision for the identification of ghost flights. During the early days of the COVID-19 pandemic, several airlines have used their aircraft as so-called preighters, a combination of passenger and freight aircraft, in order to keep their aircraft flying (Thorn, 2020). There is no global data available regarding which aircraft have been used as preighters, for which period of time, and on their distributions of passenger and cargo load factors. A related, more formal criterion would be to assess whether a specific flight is profitable for an airline or not. Nevertheless, such decisions can hardly be done based on a single flight, as it is known that airlines frequently perform cross-subsidization, where the loss in one market can be temporarily compensated by other highly profitable markets. Moreover, state subsidies can be used to suppress competitors on specific markets, while temporally operating aircraft at unprofitable parameters (Gössling et al., 2017, Wu et al., 2020). Accordingly, since aircraft operations are complex and airline-specific, we need to find an alternate measure for the potential of ghost flights. After some consideration, we decided to use a statistical measure instead of economic ones. The underlying assumption of our statistical method is that airlines will not operate ghost flights on many markets over a longer period of time. Therefore, we estimate the likelihood for ghost flights through the computation of load factors deviations across markets in Europe, comparing pre-COVID-19 levels with those experienced throughout the pandemic. The rationale is that extreme deviations towards lower load factors indicate a higher possibility of airlines operating ghost flights on specific markets. Such an assumption only holds if the ticket price is not changed significantly; an assumption which will be verified at the end of our study.

Within the field of descriptive statistics, various measures for identifying statistical dispersion and outliers have been proposed. One of these measures is the separation of distributions into quartiles, which uses the so-called interquartile range to identify the core of a statistical distribution. While this notion is straightforwardly defined and used in several recent studies (Zhang et al., 2020, Dai and Chang, 2021), we provide an example here, because it is essential to understand our methodology for the identification of abnormal flights during the pandemic. The concept is visualized in Fig. 3. Given a Gaussian distribution with mean value 0.0 and standard deviation 0.5, covering 100,000 samples, two chart elements represent the frequency distribution. The upper subchart shows a histogram revealing the typical bell-curved shape of the distribution. The lower chart is a corresponding boxplot, which highlights the median and first/third quartile of the data (labeled Q1 and Q3) within a blue box. The interquartile range, abbreviated IQR, is the difference between Q3 and Q1. Values outside the interval defined by [Q1-1.5*IQR,Q3+1.5*IQR] are considered outliers of the distribution. Within the emerging data science domain, the interquartile range is often used to perform descriptive analysis of data distributions.

Fig. 3

Histogram (top) and boxplot (bottom) for a Gaussian distribution with 100,000 samples. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

With the interquartile range providing a means to identify normal (or expected) values inside a distribution, we proceed to define the actual notion of abnormal flights next. Intuitively, we define the property of being an abnormal flight based on the historical distribution of load factors encountered on specific markets for airlines. In order to provide an example of the methodology and its rationale, we have created four independent series from a Gaussian distribution. The three series , and are centered around 0.8 and the fourth series is centered around 0.5, simulating the magnitude of values for load factors. The standard deviation of the Gaussian distribution is set to 0.2 each. The four series are given below:

Taking the three series , , and as reference, we compute the median (Quartile 2) of the three series, obtaining the reference median 0.78. The values of Quartile 1 and Quartile 3 of the three concatenated series are 0.61 and 0.85, respectively. Accordingly, the cutoff for the outliers is computed as 0.61-1.5*(0.85-0.61)=0.37. Values below 0.37 of the test series are considered as outliers with respect to the reference series. This situation is depicted in Fig. 4. Intuitively, the -axis in Fig. 4 can be considered a temporal axis, evolving from historical/reference scenarios towards recent ones. Accordingly, we argue that the statistical distribution of values in to can be used as a baseline for assessing the degree of normality in . Here, data values which have been historically abnormal (towards the left), should also be considered abnormal in later samples (on the right).

Fig. 4

Stylized example visualizing the statistical notion chosen for the identification of abnormal flight outliers in the load factor data. The area shaded in red color indicates the region of abnormal values, as determined by the reference series. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Data and results

We have obtained the data for this study from Sabre Market Intelligence, a global aviation data provider. The data consists of monthly passenger numbers, load factors and yields across all markets and for all airlines between the years 2017 and 2021, i.e., covering three years pre-COVID-19 pandemic and two years with COVID-19 pandemic. In addition, Sabre Market Intelligence contains information on all schedules flights and their flight numbers. We have selected those markets where origin and destination airports are located in Europe, including Great Britain, Iceland, and Ukraine; please refer back to Fig. 2 for the covered airports. Based on the markets between the airports of interest, we extracted data for the busiest airlines in these markets, measured in the total number of transported passengers before the onset of COVID-19. Table 1 reports the top-24 airlines identified across these markets together with essential airline information and three indicators: load factor, number of departures, and number of passengers. These indicators are computed as median over the three years before the COVID-19 pandemic, covering years 2017–2019. The airlines are sorted by the descending number of passengers. We can observe a wide range of median load factors for these airlines between 69.1% for Eurowings and 92.6% for Transavia Holland, a low-cost airline owned as a subsidiary of KLM. To summarize, all data analysis in the remainder of this study concerns these 24 airlines across the 195 airports shown in Fig. 2.

Table 1

Summary of the top-24 airlines in the area of interest, sorted by the total number of passengers.

IATA code	Airline name	Country	Load factor	Departures	Passengers
LH	Lufthansa	Germany	73.8%	122,485	15,766,654
U2	easyJet	United Kingdom	85.8%	104,058	14,981,239
VY	Vueling Airlines	Spain	84.8%	86,027	13,458,548
AF	Air France	France	84.3%	105,319	14,307,373
SK	Scandinavian Airlines System	Sweden	69.6%	101,410	11,240,737
FR	Ryanair	Ireland	86.4%	63,031	10,123,141
BA	British Airways	United Kingdom	76.6%	67,198	8,783,755
AZ	Alitalia	Italy	77.0%	66,941	8,503,622
W6	Wizz Air	Hungary	91.3%	37,607	6,795,842
KL	KLM Royal Dutch Airlines	Netherlands	86.2%	44,673	6,520,567
IB	Iberia Airlines	Spain	83.2%	44,772	6,300,975
TP	TAP Portugal	Portugal	81.1%	42,038	5,522,607
A3	Aegean Airlines	Greece	85.6%	33,047	4,976,642
OS	Austrian Airlines	Austria	73.2%	41,757	4,752,124
UX	Air Europa	Spain	87.4%	27,031	4,447,849
WA	KLM Cityhopper	Netherlands	83.9%	51,759	4,173,673
LX	Swiss International Air Lines	Switzerland	73.1%	27,511	3,747,649
NT	Binter Canarias	Spain	76.5%	41,199	2,333,604
DY	Norwegian Air Shuttle	Norway	67.9%	24,727	3,116,096
LO	LOT Polish Airlines	Poland	74.9%	37,907	2,776,214
EI	Aer Lingus	Ireland	78.9%	19,711	2,633,637
I2	Iberia Express	Spain	90.0%	14,500	2,279,382
HV	Transavia Holland	Netherlands	92.6%	10,778	1,816,562
EW	Eurowings	Germany	69.6%	15,082	1,783,745

Note: The data shown is calculated based on yearly averages for the year 2017–2019, i.e., pre-COVID-19 levels.

Fig. 5 provides an overview on the system-wide load factor evolution inside the European market of interest for the years 2017 to 2021, i.e., covering three years of pre-COVID-19 periods and almost two years facing the impact of the COVID-19 pandemic. Each box corresponds to the load factor distribution across all markets and airlines, highlighting median, quartiles, and outliers. The blue line represents the seasonal reference load factor median extracted from the years 2017–2019, projected until the end of 2021. The red line corresponds to the cutoff threshold for abnormal flights, as obtained from the seasonal reference load factor first quartile minus 1.5 times the difference between third quartile and first quartile.

Fig. 5

Evolution of system-wide load factors between the years 2017 and 2021. The blue line is the extracted monthly reference median based on the years 2017–2019 and the red line visualizes the cutoff for ghost flights as computed by the interquartile range. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 5 leads to a set of noteworthy observations. First, the pre-COVID-19 load factors in the region of interest were around 80% with a rather strong seasonality: In summer months, median load factors were found at about 90%, while during the winter nadir, median load factors can be around 75%. There is a rather strong deviation of load factors across markets, as can be seen by the boxes, highlighting 25% (Q1) and 75% (Q3) quartiles. In the period before COVID-19, there were few abnormal flights, as indicated by the small number of outliers. With the onset of the COVID-19 pandemic, the load factors plummeted significantly, reaching a clear low in April and May 2020; the two months which can probably considered to be the hardest months for aviation stakeholders, given the large number of grounded aircraft and largely reduced demands. Since then, a gradual recovery can be observed, visible in the increase of median load factors. One can vaguely perceive a seasonality trend in the recovery as well as the median load factors tend to go slightly down again towards the end of 2020. Finally, towards the end of 2021, the median load factors reach their peak during the COVID-19 pandemic, at about 70%.

The system-wide statistics can only provide a rough picture for the existence of ghost flights in the European aviation system. Accordingly, we break down these results by airlines next. Fig. 6 reports the evolution of the load factors for the 24 airlines in our study in period between year 2017 and year 2021. The gray vertical bars correspond to the boxes representing Q1 to Q3 of the load factors of that airline across all its markets. The red line visualizes the cutoff for ghost flights based historical data. It should be noted that the cutoff is computed as airline-specific here and, therefore, is different from the system-wide cutoff. The rationale is that the reference load factors (based on years 2017–2017 ) have very distinct heterogeneous patterns among these airlines. For instance, low-cost carriers, e.g., easyJet, have a much higher historical load factor and partially less seasonality than traditional full-service carriers, e.g., Lufthansa. Therefore, to identify the degree of deviation for specific airlines, we believe it is more appropriate to use the airline-specific reference cutoff. A set of interesting observations can be made based on Fig. 6. First, the number of ghost flights in the reference years is negligible. This is consistent with the assumption that airlines would not arbitrarily execute ghost flights, and partially due to our definition based on descriptive statistics for the reference years. Second, the highest degree of abnormal flights can indeed be found between March 2020 to May 2020; almost all airlines have a significant number of flights located below the red cutoff line. Some airlines, such as Austrian Airlines or LOT Polish Airlines have avoided extremely low load factors even during that period of extreme difficulty. We have investigated the case of Austrian Airlines further: Between June 2020 and December 2020, the number of departures of AUA was usually at about 30% of pre-COVID-19 level. In the year 2021, the number of departures of AUA reached approximately 50%–60%, compared to the pre-COVID-19 level, in most months. Accordingly, except from the two rather extreme months April/May 2020, the airline seems to have rather successfully mastered the COVID-19 pandemic, from the standpoint of our study. Third, some of the airlines hesitate to recover to higher load factors throughout both years, 2020 and 2021, for instance: Aegean Airlines, Aer Lingus, and Transavia Holland. Especially the case of Transavia Holland is interesting, given that this airline had extremely high load factors before the outbreak of the COVID-19 pandemic. The high load factors of Transavia Holland throughout the reference seasons can partially be explained by the fact that the airlines had mutual leasing agreements with Sun Country Airlines in the United States, allowing both airlines to maximize their aircraft usage during peak seasons.

Fig. 6

Evolution of load factors and potential ghost flights for 24 European airlines. Note that the cutoff threshold for abnormal flights – indicated by the red line – is computed airline specific. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 2 reports detailed results for each airline, grouped by four periods of six months each: January 2020–June 2020 (labeled 2020 Q1/Q2), July 2020–December 2020 (labeled 2020 Q3/Q4), January 2021–June 2021 (labeled 2021 Q1/Q2), and July 2021–December 2020 (labeled 2021 Q3/Q4). It can be observed that the low-cost carriers are consistently ranked at top of the table, having the largest absolute number of abnormal flights. Notably, for some of the airlines, the fraction of abnormal flights can be rather small, despite being ranked towards the top half, for instance, Lufthansa. Eurowings and Binter Canarias, two typical seasonal tourism airlines, are ranked at the bottom of the table.

Table 2

Airline-specific percentage of abnormal flights for distinguished periods of time and the total number of abnormal flights.

Code	Airline	2020 Q1/Q2	2020 Q3/Q4	2021 Q1/Q2	2021 Q3/Q4	Total
W6	Wizz Air	33.2%	50.8%	68.6%	39.2%	972
U2	easyJet	29.9%	36.4%	41.8%	16.3%	698
VY	Vueling Airlines	42.2%	65.7%	66.2%	15.9%	628
FR	Ryanair	34.5%	44.6%	41.4%	27.0%	601
WA	KLM Cityhopper	38.3%	65.9%	51.9%	6.5%	400
KL	KLM Royal Dutch Airlines	50.0%	63.6%	75.8%	28.4%	362
SK	Scandinavian Airlines System	43.2%	45.1%	36.1%	19.3%	350
AF	Air France	41.9%	35.5%	50.0%	12.6%	337
IB	Iberia Airlines	42.4%	76.7%	78.3%	34.4%	306
LH	Lufthansa	36.5%	27.8%	25.5%	3.8%	301
BA	British Airways	49.2%	52.4%	73.9%	32.0%	301
A3	Aegean Airlines	47.4%	39.2%	53.8%	32.9%	290
HV	Transavia Holland	52.2%	91.3%	77.3%	70.5%	266
UX	Air Europa	41.0%	37.2%	45.7%	20.0%	169
TP	TAP Portugal	34.4%	19.6%	48.9%	20.7%	158
OS	Austrian Airlines	28.1%	34.0%	36.1%	9.0%	151
EI	Aer Lingus	56.5%	75.0%	92.0%	56.2%	144
I2	Iberia Express	60.0%	64.1%	62.3%	56.4%	140
AZ	Alitalia	48.6%	9.4%	19.8%	0.6%	123
DY	Norwegian Air Shuttle	66.7%	85.4%	100.0%	67.5%	120
LX	Swiss International Air Lines	30.9%	27.5%	38.8%	9.6%	111
LO	LOT Polish Airlines	31.6%	25.6%	9.4%	3.4%	94
EW	Eurowings	28.3%	45.3%	56.9%	12.3%	86
NT	Binter Canarias	43.1%	10.7%	28.6%	10.0%	71

Note: Airlines are ranked by the total number of abnormal flights descendingly.

The results on load factors and conclusions on abnormal flights can be understood only in context of the revenues of airlines. The underlying question is whether airlines could have accepted to operate flights at significantly lower load factors through an increase in ticket prices. If an increase of yields was visible during the COVID-19 pandemic, it would mean that the observed abnormal flights could indeed have been profitable. Fig. 7 reports the evolution of airline yields over the period between the year 2017 and year 2021. Overall, we can see that the yields of airlines have been relatively stable in the periods before and during the pandemic. There are only a few airlines, for which significant changes can be observed. Binter Canarias, the airline that had the lowest number of abnormal flights, seems to have increased the yield since the onset of the pandemic. When closely investigating its historical yield, however, it seems like the yield simply happened to be in a nadir for the years 2018/2019. Another airline with significant changes in the yield during COVID-19 is Aer Lingus. Starting from the end of 2020, the yield has a wide range with peaks reaching 170% of pre-COVID-19 yields. This airline was among the hardest hit during the year 2020 regarding the load factor results presented above. This airline might indeed have used the recovery to increase its revenue to compensate the lower load factors. Iberia Airlines and Iberia Express also show slight indications for yield increases. Most airlines, however, have a stable yield, highlighting that the abnormal flights might indeed have been unprofitable for these airlines.

Fig. 7

Evolution of airline yields for 24 European airlines.

To better understand airline reactions, we further investigate the relationship between changes in departures and yields during the COVID-19 pandemic. The results are reported in Fig. 8. Red arrows indicate changes from 2019 to 2020 and blue arrows indicate changes from 2020 to 2021. It can be observed that for most airlines, the reaction in 2020, compared to 2019 data, is synchronized: Reducing the number of departures significantly, as indicated by the large number of red downwards arrows. Some of the airlines also had a reduced yield, particularly some of the low-cost carriers such as Wizz Air and KLM Cityhopper. The reactions in 2021 are more heterogeneous. Several airlines show clear trends to simply increase the number of departures, without affecting the yield, for instance, Iberia Express, Norwegian Air Shuttle, or Alitalia. For other airlines, we can observe much richer reactions, but often contrary to the reactions observed for the year 2020, however. A good example for such an airline is Ryanair, whose counter reactions in 2021 are seemingly made on a per-market bases, as indicated by the blue arrows pointing into all four quadrants.

Fig. 8

Overview on airline reactions during the COVID-19 pandemic. Each arrow visualizes the changes in the number of departures versus the changes in yield for a specific route on a year-by-year difference, i.e., red arrows represent the changes from 2019 to 2020 and blue arrows from 2020 to 2021. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

In another experiment, we are interested in the spatial distribution of markets with abnormal load factors during the COVID-19 pandemic. For each market, we have computed the fraction of abnormal flights during COVID-19. All markets are shown in Fig. 9, with a transition for green (normal flights only) to blue (abnormal flights only). We can see that the markets with a high ratio of abnormal flights are centered around a few selected airports, including Amsterdam Airport Schiphol (AMS), Josep Tarradellas Barcelona-El Prat Airport (BCN), Athens International Airport (ATH), Adolfo Suárez Madrid–Barajas Airport (MAD), Stockholm Arlanda Airport (ARN), and Dublin Airport (DUB).

Fig. 9

Spatial distribution of abnormal (blue) and normal (green) markets. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

In the next experiment, we investigate the extent to which these six airports with a notable fraction of abnormal flights have operated close to their capacity limits during the pandemic. Notably, all six airports belong to the highest category of slot level 3. We have obtained the hourly maximum aircraft movements from the Airport Corner of Eurocontrol (https://ext.eurocontrol.int/airport_corner_public) and computed the average number of observed departures during the period of our study. The results are reported in Fig. 10. For four of these airports (AMS, BCN, ATH, and DUB), we can observe that they have operated close to their maximum capacity at peak times pre-COVID-19. After the initial shock of the pandemic, the number of aircraft movements has gradually recovered to a degree of 60%–80%. Neither of the four airports has recovered towards pre-pandemic aircraft movements though. Please note, however, that Fig. 10 reports a highly aggregated perspective by months; with significant daily and hourly deviations, these airports might have well hit their maximum operation conditions for various times during the COVID-19 pandemic, especially during summer 2021. The results for the other two airports are not conclusive, as our data for pre-COVID-19 does not indicate that these airports have reached their optimal capacity limits. There are various explanations, mainly concerning limitations regarding the available data. Another explanation, especially for Stockholm Arlanda Airport (ARN), would be the operation of private/business jets (general aviation), which are not included in our data. Finally, the data from Eurocontrol holds under optimal conditions, which might not be often realized at these airports. Nevertheless, according to the data available to us, the observations from the four other airports carry over, regarding a peak recovery towards about 60%–80% of pre-COVID-19 operations.

Fig. 10

Average hourly aircraft movements at six selected airports (blue line). The dashed horizontal line indicates the capacity limits for each airport under optimal conditions as reported by Eurocontrol. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

To better understand whether there are slot level-specific effects on airports during the COVID-19 pandemic, we report the monthly aggregated sum of departures and arrivals from airports in Fig. 11, with respect to the year 2019 as baseline. A ratio of 100% means that the operations at an airport have reached the maximum of the year 2019 baseline. We find that there are no significant differences among airports from the three different categories. Notably, the vast majority of airports is operating significantly below their 2019 reference

Fig. 11

Monthly evolution of departure ratios for airports with respect to the year 2019, grouped by slot level.

Fig. 12 reports the temporal evolution of abnormal flights at airport pairs, grouped by the three slot-induced categories used above. Specifically, we have grouped all airport pairs into three categories: (a) “No Level 3”: Neither of the two airports is slot-controlled (slot-controlled as defined by Level 3), (b) “One Level 3”: exactly one airport is slot-controlled, and (c) “Two Level 3”: both airports are slot-controlled. We have derived the time series of abnormal flight fractions throughout the period of our study. We observe that there are no tremendous differences for the time series of the three categories. In general, it seems like the fraction of abnormal flights was larger among slot-controlled airports in Summer 2020. In Summer 2021, on the other hand, the fraction of abnormal flights was slightly higher among the non-slot-controlled airports. One potential explanation could be that the LCCs have pushed more aggressively into recovery in Summer 2021, serving preferably flights between secondary airports.

Fig. 12

Abnormal flights at airport pairs, grouped based on the slot properties (no Level 3, only one airport Level 3, both airports Level 3).

In a final experiment, we specify a simple linear regression model as follows: where indicates whether the market was flagged as abnormal flight (1.0) or not (0.0), is the distance (in km) of market , is the dummy variable for low-cost carriers, is the average pre-COVID-19 load factor of airline on market , is the average pre-COVID-19 number of departures of an airline on market , and is the average pre-COVID-19 yield of an airline on market . Finally, three variables indicate the airport types of the market as follows: (both airports are slot-controlled), (exactly one airport is slot-controlled), and (neither of the two airports is slot-controlled), with slot-controlled being defined as Level 3 airports. For all continuous explanatory variables, except from the low-cost carrier variable and the load factor variable, we take the logarithm in the estimation. We report a set of descriptive statistics for the variables of this experiment in Table 3. Please note that the value of continuous variables are reported as before taking the logarithm.

Table 3

Descriptive statistics of the variables in this study.

	Variable	Observations	Min	Mean	Median	Max	Std. Dev.
0	Y	21482	0.000	0.334	0.000	1.000	0.472
1	Dist	21482	111.832	1015.882	928.724	3057.516	585.576
2	LCC	21482	0.000	0.431	0.000	1.000	0.495
3	LoadFactor	21482	0.538	0.813	0.822	0.973	0.078
4	Departures	21482	7.917	109.794	89.806	542.861	85.013
5	Yield	21482	0.677	12.057	9.928	51.242	7.948
6	TwoLevel3	21482	0.000	0.561	1.000	1.000	0.496
7	OneLevel3	21482	0.000	0.355	0.000	1.000	0.479
8	NoLevel3	21482	0.000	0.084	0.000	1.000	0.277

The results of our regression experiments are shown in Table 4. The observations from the table are consistent with the results reported earlier in this study. The variable has a strong positive impact on the degree of abnormality of a flight under consideration, which indicates that low-cost carriers were more likely to execute abnormal flights during the COVID-19 pandemic. The variable has a strong positive impact as well, indicating that airlines preferably executed abnormal flights on markets which they served frequently before the COVID-19 pandemic. The p-values of the other independent variables do not suggest a significant contribution towards the presence of abnormal flights.

Table 4

Results of ordinary least squares computed by the Python package statsmodels.

Indicator	coef	std err	t	P>|t|	[0.025	0.975]
Intercept	−0.0904	0.059	−1.521	0.128	−0.207	0.026
Dist	0.0141	0.008	1.814	0.070	−0.001	0.029
LCC	0.0789	0.009	9.197	0.000	0.062	0.096
LoadFactor	0.0448	0.052	0.861	0.389	−0.057	0.147
Departures	0.0719	0.006	11.585	0.000	0.060	0.084
Yields	−0.0075	0.008	−0.994	0.320	−0.022	0.007
TwoLevel3	−0.0357	0.021	−1.669	0.095	−0.078	0.006
OneLevel3	−0.0579	0.021	−2.715	0.007	−0.100	−0.016
NoLevel3	0.0032	0.020	0.160	0.873	−0.036	0.043

Discussion

This study has investigated the impact of the COVID-19 pandemic on the load factors of 24 European airlines in their European markets. Mainly motivated by the circulated claims and rumors about ghost flights, specifically during the early phase of the pandemic, but also recurrently at later stages, we wanted to identify how much this phenomenon existed in reality and whether there were significant deviations in terms of airlines and temporal/spatial distributions. With data for the years 2017 to 2021, we used three years of pre-COVID-19 airline data as a reference baseline which was then used to identify potentially abnormal flights during the pandemic in the years 2020 and 2021. Our major findings are summarized as follows.

First, we confirmed that the peak of abnormal flights indeed occurred during the early phase of the COVID-19 pandemic, mostly in March 2020 and April 2020. The first month, March 2020, was characterized by global chaos, probably plus the use-it-or-lose-it slot policy for March, caused mainly by transition from an epidemic outbreak to a pandemic, as declared by the WHO on March 11th, 2020. Noting the significant problems airlines were facing, the European Commission quickly waived slots, as per regulation published on March 30th, 2020. Presumably, because of this decision of the European Commission and since it became clear that the pandemic has just started, airlines then significantly reduced the number of unprofitable flights over the next months. At the same time, summer season 2020 (in the Northern hemisphere) led to a significant increase in the number of flights as well as a partial recovery of load factors for several airlines. Since that time, until the end of 2021, we can observe that the load factors have gradually recovered to about 70%, which is close to the pre-COVID-19 winter nadir of 77%.

Second, our analysis confirmed that the historical load factors of European airlines are very heterogeneous, which prevents from using a fixed threshold for the identification of abnormal flights. Accordingly, in the absence of detailed additional economic data for these airlines, we introduced a statistical measure based on the seasonal interquartile range of historical load factors, which gives us an estimation regarding the expected load factors for each airline at a given point of time. With this estimation, we identified a wide range of load factor deviations during the pandemic, ranging from rather short-term deviations (mostly for traditional full-service carriers) and other long-term deviations (several low-cost carriers). When deriving this observation, one must consider that low-cost carriers have a much higher expected load factor and, accordingly, presumably a harder time to fill up their aircraft.

Third, given the observations regarding the load factor evolution during the COVID-19 pandemic, we performed further analysis regarding airline responses, which were mainly targeted at the reported yield and the number of departures for each market in the study period. For few airlines we have observed major changes in the yield per market during the pandemic, which means that airlines have tried rather not to compensate for lower load factors with higher ticket prices at a larger scale. Nevertheless, we could observe differences in airline reactions, when looking at the evolution at specific markets. Our analysis revealed that the reactions in 2021 are rather diverse across airlines and markets.

Fourth, our regression analysis indicated that some of the major drivers for abnormal flights are the airline type and the service frequency. During the pandemic, it was specifically the low-cost carriers (LCCs), which operated abnormal flights. This observation needs to be understood in context of the generally much elevated load factors for LCCs compared to full-service carriers (FSCs), i.e., the low-cost carriers abnormal load factors might still be partially in a range where FSCs would operate normally. Moreover, our experiments indicated that markets with high pre-COVID-19 service frequencies had a higher likelihood for having abnormal flights during the pandemic. This observation can be possibly explained by airlines who want to proceed serving their dominant markets. It should be noted, however, that the yield of airlines did not have a strong effect on the abnormality of flights, indicating that it is not necessarily the high-yield market that continued to be served in abnormal manners.

The results in our study lead to a set of policy implications and directions for future work. First, concerning the slot waiver and slot relief decisions of the European Commission, our results indicate that there were wide-spread abnormal flights across airlines and markets in March/April 2020. Accordingly, it seems to be the right decision to have implemented the first slot waiver program on March 30th, 2020. Afterwards, the aviation industry has seen a gradual recovery, with a concurrent reduction of abnormal flights. When major full-service carriers threatened in December 2021 that the sky could be full of ghost flights without another slot waiver/relief extension, our results show that the aviation system was widely on a track towards recovery, reaching flight counts close to comparable seasonal indicators pre-COVID-19, while showing a significantly reduced number of abnormal flights compared to those at the peak during March/April 2020 (Sun et al., 2021a). Accordingly, the European Commission and other involved policy makers should carefully fathom opportunities for stopping the slot relief packages and return to a more competitive setting in the European aviation system. These findings are in line with other studies which emphasize the need for recovery preparedness in aviation (Zhu et al., 2021).

Second, as one of the rationales for airlines’ running ghost flights is to fulfill their slot policy (e.g., the use-it-or-lose-slot rule), a more systematic, long-term slot policy should incorporate the policy changes when the industry faces a major negative demand shock (e.g., epidemic/pandemic, wars, recessions, natural disasters). While ghost flights might increase airport revenues (via extra landing fees) and keep employment during the shock, there are socially inefficient overall. Thus, the benefits of slot policies, such as the use-it-or-lose-slot rule, should be weighed against the social waste of ghost flights. Third, the apparent problems for airlines to navigate successfully through the COVID-19 pandemic and the wide range of seemingly uncoordinated actions, lead to the necessity for developing better policies and decision/optimization frameworks targeted towards pandemic-resilient aviation (Sun et al., 2021b). Traditional demand estimation, which is largely based on simple regression models under consideration of seasonality and GDP growth (among others), do not work well in a highly volatile environment of flight bans and changing passenger fears. Therefore, we need to develop a better understanding of how airlines should react in the sense of an optimal decision and develop the right policies to safely convoy airlines through such events. The COVID-19 pandemic is not over, despite large recoveries visible in many sectors, including aviation. Presumably, the next pandemic outbreak cannot be avoided, unless we significantly revise our way of understanding and operating aviation. In sum, our study suggests that in addition to the traditional policy objects behind a slot policy – airline competition, airport demand management, airport regulation (e.g., Yang and Zhang (2011)) – the possibility of ghost flights during a negative demand shock should be taken into account as well.

One major limitation of our exploratory study is the underlying data. While this study is the first to investigate the evolution of load factors and abnormal flights during the years 2020 and 2021 in the European aviation system, there is always an opportunity to extend the study under availability of additional data. It should be noted that single low demand flights are possibly not covered by our study since Sabre only provides monthly figures. Hence, routes with few ghost flights may not have been observed. In addition, we would like to particularly highlight the opportunities for a counter-factual scenario estimating the potential effects of not waiving the slot policy in March 2020 and the following months. Similarly, with more economic airline and operational airport data available, it would be interesting to further dissect the responses of airlines. However, the amount of required data for such studies is likely not available at a larger scale. Finally, it would be interesting to formally assess the causality between the made observations and slot policies during the COVID-19 pandemic. Sabre does non-trivial data preprocessing after merging information from various systems; these preprocessing steps are mostly proprietary and cannot be verified or validated. Once identifying slight changes in the flight data, we cannot ensure causality with the slot regime changes. It should also be noted that the ‘chaotic’ process at airlines might lead to stronger data signals than the slot regime change. Here, results can only be as good as the input data.

CRediT authorship contribution statement

Xiaoqian Sun: Conceptualization, Methodology, Writing – original draft. Sebastian Wandelt: Conceptualization, Software, Writing – reviewing & editing. Anming Zhang: Conceptualization, Validation, Writing – reviewing and editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.