You are on page 1of 11

Data Envelopment Analysis – An Application

in Airport Efficiency Benchmarking


Tiago Pereira, tbp@lecm.org.mo

1 – Introduction

The objective of the present study is to analyze the relative efficiency of different airports in regard to
their use of land and infrastructure in terms of number of annual aircraft movements. As such, total
surface area, number of runways, and total runway area (thus reflecting runway dimensions) are here
taken as inputs, and annual aircraft movements (i.e., arrivals and departures) are the outputs.

Effective use of infrastructure does not relate solely to number or extent of runways. Runways are the
fundamental infrastructure of airports, but the traffic capacity of airports is a function of the traffic
circulation system as a whole. A runway traffic flow capacity may be maximized through the adequate
definition of taxiways, connecting ways, parallel taxiways, and coupled with effective navigation and ATC
(Air Traffic Control) systems. Moreover, if the condition of such infrastructure is poor, it may impose
restrictions to number of aircraft movements, limit the MTOW (Maximum Take-off Weight) of certain
aircraft, or even become an impediment to certain types of aircraft trafficking such airport pavements.
Poor infrastructure further implies above normal maintenance operations, which might also be a
hindrance to traffic flow.

The factors mentioned above are bundled within the inputs of land area, number of runways and runway
area. An efficient traffic circulation system might be considered to be included in the efficient use of land.
Inefficiencies in this regard might, or might not, be dependent on airport management only, as they might
be related to government-imposed restrictions, geography, fleet mix, climate and so forth. Such
partitioning of the problem is not however a part of the present study.

The number of runways is a decisive factor in the number of movements. Having more than one runway
allows simultaneous aircraft operations and further provides flexibility for inspections and maintenance
operations. How effective runways are might also depend, apart from ATC, on airport traffic circulation
systems, pavement condition, or eventual airspace restrictions. As an example, a single runway, under fair
weather conditions (visual meteorological conditions – VMC) or operating under visual flight rules (VFR),
can accommodate 99 operations per hour of smaller aircraft and about 60 operations per hour of larger
aircraft. Under poor weather conditions (instrument meteorological conditions – IMC), these numbers
reduce to between 42 and 53 operations per hour, depending on the size of aircraft and navigational aids
available. Two parallel runways provide a capacity that is twice the capacity of a single runway, as long as
they are separated by at least 1310 meters, below which the added capacity can be significantly reducedi.

Ultimately, the problem is complex. But by considering these key inputs, it may be possible to understand
key-performance issues in airports: inefficient use of land, underperforming infrastructure, deficient
Data Envelopment Analysis – An Application in Airport Efficiency Benchmarking

airport planning, or underperforming air traffic management. To further pinpoint underlying inefficiencies
however, a more detailed study including more data is required. Inefficient use of land, for example, is
here considered from the view point of aircraft traffic (number of movements). However, an airport has
different commercial activities, and such “inefficiencies” might not be viewed upon as such from a
commercial point of view. Such a situation however, is more characteristic of large airports with a high
volume of traffic. In small to medium airports, with lower volumes, the available land is mostly used for
aviation purposes.

Inefficiencies may further relate to an airport masterplan, as the situation under analysis may refer to a
specific stage in the airport development. For such cases, the prospective future scenarios may also be
analyzed within the context of a DEA study. But even such cases are relevant for this analysis, as it is often
the case that development stages of airports are delayed and perhaps even stalled for an indeterminate
period of time.

The number of runways and runway area are here considered as separate input data. The two are related,
but from the explained above it should be well understood that having more than one runway has a
movements multiplication effect due to simultaneous operations, while runway area reflects the physical
dimensions of runways (length and width). Different aircraft operations require that runways possess
certain physical characteristics (TODA – Take Off Distance Available; LDA - Landing Distance Available). An
Airbus 380, for instance, requires, according to ICAO rules, 60 meters wide runways.

2 – Background

The airport efficiency study here conducted is based on a Data Envelopment Analysis (DEA). The
theoretical background of this technique will only be briefly introduced, as it is extensively described in
the literatureii. Its basic principles will however be outlined so as to facilitate interpretation of the results.

DEA is a nonparametric linear programming technique used to measure the relative efficiency of
homogeneous decision making units (DMUs). DMUs, in the present case, are the airports being analyzediii.

There are different variations of DEA models, but in all of them, the DMU(s) with the best efficiency in
converting inputs (X1, X2,…, Xn) into outputs (Y1, Y2, …, Ym) is identified and subsequently all other DMUs
are ranked relative to the most efficient.

In the basic DEA model – the CCR model, named after its authors Charnes, Cooper, and Rhodes – the
efficiency score for DMUQ is given by:
∑ 𝑤𝑖 ∙𝑦𝑖𝑄
𝐻𝑄 = ∑𝑖 (1)
𝑗 𝑣𝑗 ∙𝑥𝑗𝑄

Subject to, for each DMU:


∑𝑖 𝑤𝑖 ∙𝑦𝑖𝑠
∑𝑗 𝑣𝑗 ∙𝑥𝑗𝑠
≤1 (2)

And:

Page 2
Data Envelopment Analysis – An Application in Airport Efficiency Benchmarking

wi, vj ≥ 0; s = number of DMUs.

wi and vj are weights applied to outputs yis and inputs xjs, being chosen to maximize the efficiency score
HQ for DMUQ.

The efficiency is calculated for each DMU with a value between 0 and 1, with 1 being the optimal score. A
convex efficiency frontier is computed, which covers all the data points. DMUs sitting in the frontier
operate at full efficiency (1.0), while those inside the frontier operate at less than full efficiency.

The number of DMUs to be included in a model should respect the following condition:

𝑆 ≥ 𝑀𝑎𝑥{𝑁 × 𝑀; 3(𝑁 + 𝑀)} (3)

With N being the number of inputs, and M the number of outputs. In the present case, N = 3 and M = 1.
Therefore: S ≥ 12.

3 – Data and Methodology

Data was collected for a total of 15 airports (all international).


Table 1 – DMUs, Input and Output Data

Input Output
Airport IATA Code Aircraft Movements Notes
Airport Surface Area (km2) Number of Runways Runway Area [m2]
(2016)
Macau MFM 1.92 1 151200 56,000
Fukuoka FUK 3.53085 1 168000 173,680 (2015)
Luton LTN 2.45 1 99360 131,356
Urumqi URC 4.84 1 162000 162,265
Gatwick LGW 6.74 1 149220 275,633
Porto OPO 3.539 1 156600 77,361
San Diego SAN 2.67497 1 174765 196,935
Singapore Changi SIN 13 2 480000 360,490
Lisbon LIS 6.404 2 279225 178,639
Hong Kong HKG 12.55 2 456000 411,530
Shenzhen SZX 10.8 2 381000 318,600
Beijing Capital PEK 14.8 3 616000 606,086
Guangzhou CAN 15 3 618000 435,231
New York JFK 19.951 4 790773 458,707
Las Vegas LAS 11.3 4 614054 535,740

Page 3
Data Envelopment Analysis – An Application in Airport Efficiency Benchmarking

Airport Data (Number of Movements refer to the year 2016, except FUK - 2015)
900000 25

800000

20
700000
m2; Number of Movements

600000

15
500000

km2
400000
10

300000

200000
4 4 5
3 3
100000 2 2 2 2
1 1 1 1 1 1 1

0 0
MFM FUK LTN URC LGW OPO SAN SIN LIS HKG SZX PEK CAN JFK LAS

Runway Area [m2] Airport Surface Area (km2) Number of Runways Aircraft Movements (2016)

Figure 1 – Airport Data

Seven of the airports have a single runway, four of them have two runways and there are two airports
with three runways and two more with four runways. This is considered appropriate to capture properly
the effect of the inputs number of runways and runway area.

The DEA model chosen was the CCR Model in output oriented mode. This means that the efficiency scores
were determined by evaluating the level of output (movements) that a DMU (airport) should be
generating given its inputs (land area, number and area of runways)iv.

The CCR model was chosen because for this specific analysis considering constant returns to scale appears
to be realistic. The addition of a runway provides a considerable increase in aircraft movement capacity.
As mentioned above, when adding a runway to a single runway airport, capacity may well duplicate. And
in terms of land area, while, as also mentioned previously, in large airports there is a greater percentage
of land dedicated to non-aviation related activities, the data does show an appreciable correlation
between land area and number of movements. This option is thus considered justified when analyzing
traffic volume. The same would not be true if analyzing airport revenue where clearly variable returns to
scale would have to be considered.

Page 4
Data Envelopment Analysis – An Application in Airport Efficiency Benchmarking

700,000

600,000 R² = 0.770

Number of Movements
500,000

400,000

300,000

200,000

100,000

0
0 5 10 15 20 25
Airport Surface Area

Figure 2 – Relationship between Airport Surface Area (km2) and Number of Aircraft Movements

An alternative analysis was performed excluding the two 4-runway airports (JFK; LAS) and the efficiency
scores for all other airports remained the same as when considering all 15 airports.

700,000

600,000 R² = 0.8029
Number of Movements

500,000

400,000

300,000

200,000

100,000

0
0 200000 400000 600000 800000 1000000
Runway Area

Figure 3 – Correlation between Runway Area and Number of Movements

4 – Analysis Results

The DEA-Solver softwarev was used in the calculations here presented.

Page 5
Data Envelopment Analysis – An Application in Airport Efficiency Benchmarking

Airport

OPO 0.3792

MFM 0.3962

JFK 0.4744

LIS 0.488

CAN 0.5996

SZX 0.638

SIN 0.6651
DMU

URC 0.6794

LAS 0.7261

HKG 0.7717

FUK 0.8175

PEK 0.8395

LTN 0.9775

SAN 1

LGW 1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Efficiency

Figure 4 – Efficiency Scores (DEA CCR Model – Output Oriented)

Figure 4 presents the efficiency scores obtained. The following observations can be readily made:

• The most efficient airports are London Gatwick (LGW) and San Diego International Airport (SAN);
they are both taken as working at maximum efficiency and lie in the frontier;
• Gatwick’s score is hardly surprising given its infamous title as the World’s busiest single runway
airport;
• The San Diego (USA) airport is also evaluated at maximum efficiency; while its traffic volume is
lower when compared to Gatwick’s, it has a much smaller land area;
• Beijing Capital Airport (PRC) shows the highest efficiency score between the multi-runway
airports;
• JKF Airport (New York, USA), in contrast to Beijing, despite having one more runway and a much
larger land area, has considerably less traffic movement; it is therefore no surprise that it ranks
as the third less efficient airport between the 15 analyzed;
• Guangzhou Baiyun Airport (PRC), with approximately the same land area, number of runways and
runway area as Beijing Capital Airport has a considerably lower traffic volume; it’s efficiency score
is thus not surprisingly low;
• Also with a low score is Lisbon’s Humberto Delgado Airport; This is, in part, due to the fact that it
has 2 runways; these are intersecting runways, and it is known that one of them (RWY 17/35) has
very little usage;
• Porto’s Francisco Sá Carneiro Airport (Portugal) and the Macau International Airport have the
lowest efficiency scores, both of which, given their infrastructure, should be receiving a much
higher traffic volume.

Page 6
Data Envelopment Analysis – An Application in Airport Efficiency Benchmarking

Table 2 – DMU Efficiency Scores and Benchmarks (Reference DMUs)

No. DMU Score Rank Reference(Lambda)


1 MFM 0.3962 14 SAN 0.718
2 FUK 0.8175 5 LGW 0.215 SAN 0.777
3 LTN 0.9775 3 LGW 0.209 SAN 0.39
4 URC 0.6794 8 LGW 0.533 SAN 0.467
5 PEK 0.8395 4 LGW 1.667 SAN 1.333
6 SIN 0.6651 9 LGW 1.882 SAN 0.118
7 LGW 1 1 LGW 1
8 LIS 0.488 12 LGW 0.478 SAN 1.19
9 OPO 0.3792 15 LGW 0.256 SAN 0.677
10 HKG 0.7717 6 LGW 1.771 SAN 0.229
11 SZX 0.638 10 LGW 1.341 SAN 0.659
12 CAN 0.5996 11 LGW 1.716 SAN 1.284
13 JFK 0.4744 13 LGW 2.276 SAN 1.724
14 SAN 1 1 SAN 1
15 LAS 0.7261 7 LGW 0.427 SAN 3.149

The analysis was also carried out without considering the two airports with four runways (JFK and Las
Vegas). The results are illustrated in Figure 5.

Airport

OPO 0.3792

MFM 0.3962

LIS 0.488

CAN 0.5996

SZX 0.638

SIN 0.6651
DMU

URC 0.6794

HKG 0.7717

FUK 0.8175

PEK 0.8395

LTN 0.9775

SAN 1

LGW 1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Efficiency

Figure 5 - Efficiency Scores (DEA CCR Model – Output Oriented) without considering JFK and LAS

As expected, the efficiency scores for the remaining airports remain unaffected.

Table 2 presents the efficiency scores of each DMU and the benchmarks (reference DMUs), with
corresponding intensities (lambda values). A review of this Table in conjunction with Figure 1 provides an
intuitive interpretation:

Page 7
Data Envelopment Analysis – An Application in Airport Efficiency Benchmarking

• The benchmark for the Macau International Airport (MFM) is clearly the San Diego International
Airport (SAN), given its similarities in terms of relative proportions of input;
• Through the same logic, the Urumqi International Airport (URC) is something of a mix bag
between these two reference airports, with hence its ideal model being somewhere in between
these two cases;
• Such logic can be applied to the other airports considered.

As a comparison example, an alternative analysis was performed considering the following virtual
scenario:

• Lisbon Airport considered with only one runway (its main runway);
• Porto’s Airport with a total number of annual movements of 195000.

This virtual scenario yields the following efficiency scores (Figure 6).

Airport

MFM 0.3962

JFK 0.4744

CAN 0.5996

SZX 0.638

LIS 0.6638

SIN 0.6651

URC 0.6794
DMU

LAS 0.7261

HKG 0.7717

FUK 0.8175

PEK 0.8395

OPO 0.9558

LTN 0.9775

SAN 1

LGW 1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1


Efficiency

Figure 6 – Efficiency Scores for a virtual scenario (Lisbon with 1 Runway only; Porto with 195000 annual departures)

In this virtual scenario, Lisbon Airport’s score improves by approximately 18%, while Porto Airport jumps
from least efficient airport with 38% efficiency to 4th in the ranking, with 96% efficiency.

Page 8
Data Envelopment Analysis – An Application in Airport Efficiency Benchmarking

5 – Conclusions

Data Envelopment Analysis was used to analyze the relative efficiency of 15 airports from vastly different
regions and sizes. The study focused on analyzing how well, in relative terms, the airports managed
available land and infrastructure from the point of view of airport traffic volume, providing results that
are intuitive.

The CCR model was used, thus considering constant returns to scale, which is found appropriate for the
problem at hand.

The study can be refined with more data and an improved definition of the problem. Airport data however
is difficult to obtain, the problem being augmented by the fact that the higher the number of inputs and
outputs, the more DMUs will be required for the analysis to be truly significant.

In this regard, it should be mentioned that there is doubt concerning the airport surface area data, as in
some cases the information was not obtained from the airport authorities, but from outside sources. The
interpretation of the results here presented should therefore be made bearing this factor in mind.

For the problem at hand, it would have been more useful to consider airport surface area for aviation
purposes, instead of overall airport surface area.

Terminal capacity, cargo storage capacity and total aviation related staff employed are further inputs that
could be added in a more refined analysis. Likewise, annual number of passengers and cargo throughput
could be added as outputs.

It might be suggested that such a study might best be conducted by considering separate categories of
airports based on size. However, in this regard, it is noteworthy that the Beijing Capital Airport is ranked
fourth in the analysis, with 84% efficiency.

Nevertheless, the analysis provides meaningful results that serve the purpose of a general evaluation of
the relative efficiencies of airports in terms of traffic volume, irrespective of their size.

References

Airport Data:

• Runway Information: www.airportguide.com


• Number of movements for Lisbon and Porto Airports: ANA Aeroportos de Portugal – Relatório
de Gestão e Contas 2016
• Lisbon Airport Land Area: www.publico.pt/2005/12/19/economia/noticia/area-do-aeroporto-
de-lisboa-vale-965-milhoes-de-euros-1242292
- Gatwick Airport Information:
- www.gatwickairport.com : Gatwick Facts & Stats; Gatwick by Numbers
- Gatwick Airport Interim Master Plan – Executive Summary – BAA Gatwick
• Luton Airport:

Page 9
Data Envelopment Analysis – An Application in Airport Efficiency Benchmarking

- Surface Area:
https://www.iema.net/assets/nts/Terence_O%27Rourke/London_Luton_Airport_NTS_
November_2012.pdf
- Statistics: https://www.london-luton.co.uk/corporate/lla-publications/statistics
• Guangzhou Airport:
- Surface Area:
https://www.travelchinaguide.com/cityguides/guangdong/guangzhou/airport.htm
- Statistics: http://news.carnoc.com/list/393/393555.html
• Shenzhen Airport:
- Surface Area: Tai Wai Ying, Selina; Airports in Hong Kong and Shenzhen: Competition
and Cooperation. Master of Arts Dissertation. The University of Hong Kong;
- Statistics: http://i.carnoc.com/detail/385649
• Beijing Capital Airport:
- Surface Area: https://www.finavia.fi/zh/node/4528
- Statistics: http://www.caac.gov.cn/XXGK/XXGK/TJSJ/201702/t20170224_42760.html
• Urumqi Airport:
- Surface Area: https://www.travelchinaguide.com/cityguides/xinjiang/urumqi/airport/
- Statistics: http://www.caac.gov.cn/XXGK/XXGK/TJSJ/201702/t20170224_42760.html
• Singapore Changi Airport:
- Surface Area:
https://en.wikipedia.org/wiki/Infrastructure_of_Singapore_Changi_Airport#cite_note-
AIP-6
- Statistics: http://www.changiairport.com/corporate/media-
centre/newsroom.html#/documents/annex-a-passenger-airfreight-and-aircraft-
movement-statistics-for-2016-63855
• Fukuoka Airport:
- Surface Area: http://www.fuk-ab.co.jp/english/sp/gaiyou.html
- Statistics:
https://web.archive.org/web/20161021205147/http://www.mlit.go.jp/common/00114
1840.pdf
• Macau International Airport:
- Surface Area: http://www.chec.bj.cn/zg/tabid/896/InfoID/6289/Default.aspx
- Statistics: http://www.macau-airport.com
• Hong Kong International Airport:
- Surface Area: http://www.hongkongairport.com/eng/media/facts-figures/facts-
sheets.html
- Statistics: http://www.hongkongairport.com/eng/media/facts-figures/air-traffic-
statistics.html
• New York JFK:
- Surface Area: https://www.panynj.gov/airports/jfk-facts-info.html
- Statistics: http://www.panynj.gov/airports/pdf-traffic/ATR2016.pdf
• San Diego International Airport:
- Surface Area: https://skyvector.com/airport/SAN/San-Diego-International-Airport
- Statistics: http://www.san.org/News/Air-Traffic-Reports
Page 10
Data Envelopment Analysis – An Application in Airport Efficiency Benchmarking

• Las Vegas McCarran International Airport


- Surface Area: https://skyvector.com/airport/LAS/Mc-Carran-International-Airport
- Statistics:
https://www.mccarran.com/FSWeb/file/396253/2016%20FAA%20Tower%20Operations
.pdf

i
Alexander T. Wells, Ed.D. & Seth Young, Ph.D. “Airport Planning & Management”, 5th Edition. McGraw-Hill
Companies, The. 2003
ii
Wade D. Cook, Larry M. Seiford. “Data Envelopment Analysis (DEA) – Thirty Years On”. European Journal of
Operational Research 192 )2009= 1-17
iii
Sérgio P. Santos. “Organizational Performance”. DBA Advanced Research Topics Notes. 2017
iv
David Schaar and Lance Sherry, Ph.D. “Comparison of Data Envelopment Analysis Methods Used in Airport
Benchmarking”. Third International Conference on Research in Air Transportation. 2008
v
http://www.saitech-inc.com/products/prod-dsp.asp

Page 11

You might also like