UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

An investigation of key issues for improving quality of airport benchmarking : focus on empirical methods Lin, Zhuo 2009

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2009_spring_lin_zhuo.pdf [ 2.83MB ]
Metadata
JSON: 24-1.0068089.json
JSON-LD: 24-1.0068089-ld.json
RDF/XML (Pretty): 24-1.0068089-rdf.xml
RDF/JSON: 24-1.0068089-rdf.json
Turtle: 24-1.0068089-turtle.txt
N-Triples: 24-1.0068089-rdf-ntriples.txt
Original Record: 24-1.0068089-source.json
Full Text
24-1.0068089-fulltext.txt
Citation
24-1.0068089.ris

Full Text

AN INVESTIGATION OF KEY ISSUES FOR IMPROVING QUALITY OF AIRPORT BENCHMARKING: FOCUS ON EMPIRICAL METHODS  by ZHUO UN B.E. in Transportation, Shanghai Jiaotong University, 2006  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE IN BUSINESS ADMINISTRATION  in  THE FACULTY OF GRADUATE STUDIES (Business Administration)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) May 2009  © Zhuo Lin, 2009  ABSTRACT Significant changes have occurred over the last three decades in the aviation industry. Deregulation of the airline industry has led to increasingly competitive airline markets domestically and internationally. Consequently, views about airports began to change: from that of a public utility to a more business/commercial entity, which lead to the worldwide moves towards corporatization, commercialization, and privatization of airports. However, most airports enjoy a quasi-monopolistic position and may abuse such a position. Airport performance measurement and benchmarking, therefore, has become increasingly important for airlines, investors, regulators, and airport managers to ensure efficient operation of airports. The main objectives of this study are to empirically compare the three key methodologies for measuring airport efficiency, namely the productivity index method, Data Envelopment Analysis (DEA) method, and stochastic frontier method; and to examine the effects of regional price variations on efficiency measures and rankings. The study is based on the data for 62 North American airports which was kindly provided in confidence by the ATRS Global Airport Performance Benchmarking project. The main findings are: (a) the efficiency scores and airport rankings measured by the three alternative methods are quite similar to each others among the top and bottom ranking airports, whereas considerable differences are observed among the airports in the middle; and (b) As expected, regional price level adjustments help improve the accuracy of efficiency measurement, suggesting that whenever possible regionally differentiated price data should be used instead of national aggregate price data.  11  TABLE OF CONTENTS  ABSTRACT  ii  TABLE OF CONTENTS  iii  LIST OF TABLES  v  LIST OF FIGURES  vi  ACKNOWLEDGEMENTS  vii  INTRODUCTION  1  1.1 Background  1  1.2 Purpose of the Study  4  1.3 Scope of the Study  4  1.4 Organization  4  LITERATURE REVIEW  6  2.1 Airport Studies with Index Number Method  6  1  2  3  4  2.2 Airport Studies with Data Envelopment Analysis  11  2.3 Airport Studies with Stochastic Frontier Analysis  13  2.4 Conclusion  15  METHODOLOGY  20  3.1 Index Number Method  20  3.2 Data Envelopment Analysis (DEA)  24  3.3 Stochastic Frontier Analysis (SFA)  28  3.4 Comparison of Methodologies  30  DATA CONSTRUCTION AND IMPACT ON AIRPORT EFFICIENCY  .  .35  4.1 Airport Outputs and Inputs  35  4.2 Data Construction  38  4.3 Impact of the Data Improvement on Gross VFP  45  4.4 Conclusion  56 Ill  5  6  ESTIMATION SULTS.58 5.1 Gross Efficiency Results between Methodologies  58  5.2 The Impact of Airport Characteristics on Gross Efficiency Result  64  5.3 Managerial Efficiency Results between Methodologies  68  5.4 Conclusion  72  CONCLUSION  73  6.1 Summary of Key Findings  73  6.2 Suggestions for Further Research  74  BIBLIOGRAPHY  76  APPENDIX  81  iv  LIST OF TABLES  Table 2.1 Summary of Airport Efficiency Studies  .16  Table 3.1 Differences between DEA Rankings  32  Table 3.2 Comparison of Index Number Method, DEA and SFA  33  Table 4.1 Regression Result of Snow Removal Cost  43  Table 4.2 Summary of Data Construction  45  Table 4.3 Significant Impact of Adjusting Non-Aeronautical Revenue  47  Table 4.4 Significant Impact of Adjusting Soft Cost Input  48  Table 4.5 Significant Impact of Adjusting Price Indices  49  Table 4.6 Impact of Adjusting Price Indices on Top/Bottom Airports  50  Table 4.7 Significant Impact of Adjusting Snow Removal Cost  51  Table 4.8 Impact of Adjusting Snow Removal Cost on Top/Bottom Airports  52  Table 4.9 Summary of VFP Scores  53  Table 4.10 Significant Impact of Combined Improvements  55  Table 4.11 Impacts of Combined Improvements on Top/Bottom Airports  55  Table 5.1 Pearson Correlation Matrix of Gross Efficiency Scores  59  Table 5.2 Comparative Gross Rankings between Methodologies  60  Table 5.3 Pearson Correlation Matrix of Gross Rankings  62  Table 5.4 Regression Results on Gross Efficiency Scores  67  Table 5.5 Pearson Correlation Matrix of Residual Efficiencies  68  Table 5.6 Comparative Residual Rankings between Methodologies  69  Table 5.7 Pearson Correlation Matrix of Residual Rankings  70  V  LIST OF FIGURES  Figure 3.1 Illustration of DEA  25  Figure 3.2 Illustration of SFA  29  Figure 3.3a DEA Illustration Figure 3.3b DEA Illustration  —  —  Original  31  Removing an Efficient DMU  32  Figure 4.1 Proportion of Non-Aeronautical Revenue in 2006  37  Figure 4.2 Proportion of Labor Cost in 2006  38  Figure 4.3 Comparison of PPP and COLT Indices  40  Figure 4.4 Output of Non-Aeronautical Revenue by using PPP and COLT  41  Figure 4.5 Input of Soft Cost by using PPP and COLT  42  Figure 4.6 Input of Soft Cost before/after Improvements  44  Figure 4.7 Impact of Adjusting Non-Aeronautical Revenue on Gross VFP  46  Figure 4.8 Impact of Adjusting Soft Cost Input on Gross VFP  48  Figure 4.9 Impact of Regional Price Indices on Gross VFP  49  Figure 4.10 Impact of Adjusting Snow Removal Cost on Gross VFP  51  Figure 4.11 Impact of Combined Improvements on Gross VFP  52  Figure 4.12 Normal Distribution Test of VFP Differences  54  Figure 5.1 Gross Efficiency Measurements of VFP, DEA and SFA  59  Figure 5 .2a Gross Ranking Comparison of Top 15 Airports  62  Figure 5.2b Gross Ranking Comparison of Bottom 15 Airports  63  Figure 5.2c Gross Ranking Comparison of Mid-Ranked Airports  63  Figure 5.3a Residual Ranking Comparison of Top 15 Airports  71  Figure 5.3b Residual Ranking Comparison of Bottom 15 Airports  71  Figure 5.3c Residual Ranking Comparison of Mid-Ranked Airports  72  vi  ACKNOWLEDGEMENTS  I would like to express my sincerest gratitude to my supervisor, Professor Tae H. Oum, for all of his support and encouragement during my master studies and throughout all stages of this thesis. This thesis would not be completed without his guidance and support. I would like to extend my appreciation to Professor Anming Zhang for serving on my committee and for his continuous encouragement and help during my years at UBC. I would like to thank Professor David Gillen for serving on my committee and for his invaluable advice along the way.  Special thanks to Professor Chunyan Yu for serving as my university examiner and her great experience and support for my thesis. I would also like to thank TLOG students  —  Xiaowen Fu, Andrew Yuen, Sarah Wan, Hamed Hasheminia, Kelly Loke, Wei Song, and Qi Zhao  -  for their help and advice during my study. I will be always grateful for their  wiliness to help.  I would like to thank my parents, for their endless encouragement over the years, as I could not have finished this thesis without their support.  Special thanks to my friend, Kenneth Chang, for helping me revise the thesis.  This thesis is dedicated to all of you.  VII  1. INTRODUCTION Air transportation is an important industry in the modern economy. According to Airbus (2007), air transport contributes, directly and indirectly, 8% of world gross domestic product. Even in the current global recession, air passenger travel is expected to grow at 2.5% per annum in the next five years in North America’, and even faster in developing markets, such as China, India, and South America. As an essential component of air transportation, airports play an important role in facilitating movements of aircraft, passengers, and cargo, and providing various other services to both airlines and passengers. In the first chapter of this study, we review the significant changes affecting airports in the last two decades, and describe the importance of measuring efficiencies of airports.  1.1 Background Several major changes affecting airports have occurred during the last two decades. Since the late 1 970s, the continuous deregulation of airline industry has led to increasingly competitive airline markets, domestically and internationally. As a result, air ticket price has been lowered which largely stimulated air travels and boosted passenger throughput at airports. Share of airport costs in airline’s total costs has increased significantly over time. Meanwhile, there is also a trend of commercialization, corporatization, and privatization among airports, which were traditionally owned and operated by governments. In 1987, the seven major public-owned airports in UK were sold to BAA plc, which is a 100% private firm. As a result of airport privatization, regulatory constrains have been lifted, but airports no longer have access to public funding. Governments have largely reduced the subsidies for building new airports and/or expanding capacity of existing airports. Thus, reduced funding to airports expansion and more traveling passengers both have led to serious delay and congestion problems at hub Current Market Outlook (Boeing, 2008)  airports in U.S. and Europe. As a result, monitoring airport performance by profit measures or traffic growth is no longer viable. Airlines, investors, regulators, and airport managers themselves have become increasingly interested in airport efficiency, which measures airport performance using inputs and outputs in both physical and financial terms.  Airline managers use airport efficiency measurement to reduce operation expenses and improve aircraft performance. As stated previously, along with the airport privatization, share of airport costs in airline’s total costs has significantly increased. In the highly competitive airline markets, airlines are seeking every method to reduce cost without compromising service quality. Efficient airports facilitate airlines to accelerate the movements of aircraft and passenger, thus help reduce the turnaround time and increase aircraft utilization. As benchmarking and efficiency measurement identifies these efficient airports, it helps airline managers to choose airports that will enhance their performances.  In addition, regulators and investors are also interested in airport efficiency and benchmarking. First, as airports become more independent of direct government control, regulators require a variety of indicators, including productivity measures, to ensure that airport managers are making the best use of the resources at their disposal and providing required services at a fair price. Airport efficiency measurement provides such indicators for governments to establish and review the regulations they set. Further, as airport privatization has enabled commercial funds to invest airport assets, many investors have been seeking business opportunities in airport sector. As most airport investments affect the relationship between an airport’s inputs and outputs, which in turn influence the overall efficiency, investors and bankers have to rely on benchmarking techniques to assess value of the airports and identify business opportunities.  2  Moreover,  the  evolution  of  airport  ownerships  towards  privatization  and  commercialization naturally leads airport managers to seek ways to gain insights into their operations and improve their performance by benchmarking themselves against others based on efficiencies. As benchmarking identifies the best practice standards for infrastructures and services, it also helps airport managers discover ways to improve performance and deal with delays and congestion.  Besides serving all of these interested parties, airport efficiency measurement is particularly important for the industry considering that airports have substantial market power over the majority of local traffic and airlines. In most cases, airlines, passengers, and other airport users have limited choices when selecting airports. Regulatory, geographical, economic, social, and political constraints all hinder direct competition between airports. Therefore, the competitive pressure in the airport market cannot be relied upon to exert pressures for airports to improve productivity and efficiency. However, by exposing inefficient airports to their stakeholders and the general public, airport benchmarking is helpful in pushing the inefficient airports to be more productive.  In general, airport benchmarking and efficiency measurement is critical to airlines, investors, regulators, as well as airport managers. It is a useful tool to identify deficiencies and excellence in airport performances and can spur competitive forces and shake up conveniently thinking (Kincaid and Tretheway, 2006). In the past, many studies have used various methods to measure and compare the efficiency of airports. However, there are major disagreements on what should be the appropriate approach, how to define airport outputs and inputs, and what form of data to use. Therefore, the empirical results emerged from these studies are not comparable, which leads to the purpose of this study.  3  1.2 Purpose of the Study The principal objective of this study is to review and empirically compare several key methodologies employed in measuring airport efficiency, namely productivity index method, DEA (Data Envelopment Analysis) method, and stochastic frontier method (production function approach). Using an identical airport sample, we estimate and compare the efficiency scores and airport rankings across these three methods. Secondary, we explore some issues in data construction, and examine the impacts of improving quality of the data on measured efficiency. In addition, we examine the appropriate outputs, inputs, and other variables of airports heterogeneity. By elaborating the similarities and differences among airports, we try to provide a standard in identifying airports activities and operations.  1.3 Scope of the Study This study uses the data on a pooi of 55 U.S. airports and 7 Canadian airports in 2006. There are several reasons for choosing North American airports. First, North America is the largest air transport market in the 2 world Second, the ownership and regulatory . framework of North American airports are relatively consistent: they are owned either by government departments or by airport authorities. In addition, airports in North America are more willing to provide extensive and reliable data on traffic and financial performances, which makes it possible to conduct a valid study. Most airport data in this study are kindly provided to us for use on this thesis by the ATRS (Air Transport Research Society) Global Airport Performance Benchmarking project. The detailed data sources are listed in the Appendix.  1.4 Organization The study is organized as follows: Chapter 1 reviews the importance of airport efficiency 2  In 2007, Intra-North America market achieved passenger traffic of 977.4 billion RPKs (Revenue Passenger Kilometers), which accounted for 23.1% of world total. 4  measurement and introduces the purpose and the outlines of this study. Chapter 2 reviews and summaries previous literatures in measuring airport efficiency. The alternative methodologies of measuring efficiency (Index Number Method, DEA, and Stochastic Frontier Analysis) are reviewed and developed in Chapter 3. Chapter 4 reviews outputs and inputs of the sample airports, and discusses data construction. This chapter further explores a number of data improvements, and examines their impacts on measured efficiency. Chapter 5 estimates and compares efficiencies scores and airport rankings across the three alternative methods. Chapter 6 presents a summary of key findings and suggestions for further research.  5  2. LITERATURE REVIEW This chapter reviews and summarizes airport efficiency literatures, and describes limitations of previous studies. Following the discussion in Chapter 1, although a large number of different methods have been developed to measure productivity and efficiency, the following three methods have been more widely used in measuring airport efficiency than other methods: (1) Index Number Method, (2) DEA (Data Envelopment Analysis), and (3) SFA (Stochastic Frontier Analysis). Forsyth (2000) provided a brief overview of above three methods focus on methodology and assumptions. Oum, Tretheway and Waters (1992) generalized the concepts and methods used for productivity measurement in the transportation industry. However, in the airport sector, significant differences exist among airports in terms of ownership, regulatory framework, and other factors beyond management control. Researchers usually have difficulty in composing an ‘apples to apples’ comparison. So far, there has been no standard approach in measuring airport efficiency. Thus, this chapter organizes the literatures according to the methods they used.  2.1 Airport Studies with Index Number Method In this section, we discuss the studies that use Index Number Method to measure and compare airport productivity and efficiency. In general, Index Number Method defines productivity as a ratio of output on input. Based on whether to incorporate all outputs and inputs, the method distinguishes between Partial Factor Productivity (PFP) and Total Factor Productivity (TFP). Further, since measuring capital input consistently across airports in different countries remains controversial in TFP analysis, Variable Factor Productivity (VFP) with fixed level of capital input has become another widely used index number approach in measuring airport efficiency. We then elaborate the literatures of each index number respectively as following.  Most early studies in airport benchmarking incline to measure partial factor productivities 6  as output per unit of single input. Doganis (1992) summarizes a number of partial productivity indicators for airport performance, including cost performance, labor productivity, capital productivity, revenue per unit of input, etc. His work is one of the early examples of ranking airports in term of cost efficiency. More recently, Air Transport Research Society (ATRS) and Transport Research Laboratory (TRL) both include a number of partial factor measures in their annual airport benchmarking reports, such as productivity per unit of labor or per unit of capital.  Partial productivity measures are easy to understand and compute with less data required. They are useful in comparing efficiency of using a particular input across firms operating in similar operating environments, or over time within a firm when the operating environment and input combinations remain relatively stable. Managers and regulators could use this approach to examine airport performance on using a particular input. However as most industries, including airport, involve multiple outputs and inputs, PFP is not an ideal approach for comparing the overall productivity performance. It is possible that PFP of one input is dependent on the level of other inputs. For example, investment in capital can be used to improve labor productivity, but this does not necessarily mean that the firm is more efficient overall. Raising the productivity of labor input might be at the expense of reducing a productivity of another. Therefore, exclusive reliance on partial productivity measures for overall performance is quite tenuous if not misleading. It is preferred to use some forms of total productivity index that examines the relationship between all outputs and inputs.  The TFP index, which involves all outputs and inputs, is defined as a ratio of aggregate outputs over aggregate inputs. One of the most dominant TFP model is the multilateral TFP method proposed by Caves, Christensen and Diewert (1982). Diewert (1992) further generalized a number of conceptual index number approaches for measuring TFP,  7  including both the non-parametric approaches and parametric approaches. Since then many studies have used the TFP method to measure efficiency of airports.  Hooper and Hensher (1997) applied the TFP method to investigate the performances of six Australian airports. Price deflated aeronautical revenue and non-aeronautical revenue were defined as outputs, while labor, capital, and other costs were distinguished as inputs. They further examined the relationship between airport efficiency and output scale, and calculated the residual TFPs by removing the scale effects. Their study followed the multilateral TFP model by Caves, Christensen, and Diewert (1982). However, because of data restrictions, only a small number of Australian airports were examined within a four-year period. Therefore, it is hard to derive long-term trend of airport performance from their result.  Nyshadham and Rao (2002) estimated productivity performance for 25 European airports in 2005 via a TFP model. Further, their study examined the relationships between TFP index and various partial productivity measures. The result showed that total revenue per workload unit and workload units per employee had the highest impact on TFP index. However, the definition of airport’s outputs and inputs underlying this study is not clearly discussed, the results are thus difficult to compare.  In practice, the multilateral TFP model requires extensive information on output and input, which is not always obtainable for airports. Therefore, some researchers have explored alternative index methods to conduct TFP analysis.  Yoshida (2004) proposed an Endogenous-weight TFP model to measure airport efficiency, which only requires quantitative data for outputs and inputs. In his study, airport outputs were defined as cargo volume, passenger throughput, and aircraft movement (ATM),  8  while total runway length and terminal area size were identified as airport inputs. By using a linear programming model, the author estimated total factor productivities of 30 Japanese airports in 2000. However, in order to solve the TFP model, a set of arbitrary parameters has to be pre-determined, which impedes the further application of this method.  Transport Research Laboratory (TRL) estimates and compares total factor productivities of world’s major airports in its annual benchmarking report. Aimed to solve airport heterogeneity, TRL defines a uniform set of airport activities, covering six aspects in financial and traffic operation, which are EBITDA (Earning Before Interest, Taxes, Depreciation and Amortization), Concession revenue per passenger, Aeronautical revenue per ATM (Air Transport Movement), ATU (Airport Throughput Unit) per unit of asset value, ATU per employee, and Operating plus staff cost per passenger. In the next stage, a multi-attribute assessment is used to derive TFP index by aggregating scores of the above six indicators with equal-weights. TRL report is contentious in the ways they define airport core activities and regard them as equally important. Whether those attributes cover all aspects of airport operations is questionable. The efficiency results from TRL remain uncertain.  In conclusion, TFP model requires detailed specifications on airport outputs and inputs. Previous TFP studies, which tried alternative methods to avoid calculating the prices and quantities, end up with unreliable results. The inherent difficulty in measuring capital input makes it nearly impossible to conduct total factor analysis for multiple airports over time and across countries (The discussion of measuring capital input is incorporated in Chapter 3.2.2). Given the long lead-time and government concern, airport managers usually do not have full control over capital input .Thus, measuring variable factor 3 Even for fully privatized airports, for example the airports under British Airport Authority, major infrastructure decisions are still subject to government approval. 9  productivity (VFP) with a fixed level of capital input is reasonable in the short to medium term.  Air Transport Research Society (ATRS) incorporate a VFP model to measure airport operating efficiencies in its annual airport benchmarking report. ATRS defines three types of outputs as passenger throughput, ATM, and non-aeronautical revenue; while inputs are distinguished between labor cost, and the purchased service and materials. The VFP result implies that on average airports in North America are more efficient than airports in Europe and Asia Pacific. Further, ATRS examines the relationship between productivity scores and a set of airport characteristics using second stage regressions. The result shows that airports with high proportion of non-aeronautical revenue, high percentage of connecting passenger, and dominated by a few airlines achieve high VFP scores.  Oum, Adler and Yu (2006) measured variable productivities of 116 world’s major airports and examined the effects of ownership forms on efficiency result. Their findings confirm that airports owned and operated by U.S. government branches or airport authorities in North America, or airports elsewhere operated by 100% government corporations do not have statistically lower operating efficiency than airports with a private majority ownership. In addition, airports with government majority ownership and those owned by multi-level of government are significantly less efficient than airports with a private majority ownership.  In summary, a number of studies have used productivity index method to measure efficiency of airports. The multilateral TFP method is a preferred approach given sufficient data. In practice, because of the inherent controversy and difficulty in measuring airport capital input, many studies have tried to identify airport operating efficiency by the VFP method.  10  2.2 Airport Studies with Data Envelopment Analysis Besides productivity index number, Data Envelopment Analysis (DEA) is another widely used efficiency measurement in airport sector. DEA was first proposed by Charnes, Cooper and Rhodes (1978) as a new technique in operations research for measuring relative efficiencies. As a non-parametric method, DEA uses linear programming to construct a piecewise linear “efficient frontier” that envelops the Decision Making Units (DMUs) based on outputs and inputs quantities. Efficiency indices are then calculated relatively to this frontier.  Since the introduction, numerous studies have applied DEA method for measuring efficiencies in a wide variety of industries and nonprofit agencies: universities, hospital, retail stores, and transportation. The lower data requirements of DEA have enabled performance comparison in public service industries which is hard to conduct before. Seiford (1994) cited more than 400 articles with DEA method in a comprehensive bibliography. Similarly, in airport sector, a large number of studies have been published exploring both the application and interpretation of DEA ever since.  Gillen and Lall (1997) is among the first studies which used DEA method to benchmarking airports. Their study separated airport activities into airside and terminal , and applied DEA to each independently of the other. Under the assumption 4 operations of constant return to scale and variable return to scale, efficiencies for both sides were derived based on a pooled data of 21 U.S. airports for the  1989-1993 period. Further, in  order to examine how managerial decision-making and initiatives could be attributed to airport performance, they conducted a Tobit regression of DEA efficiency scores on a number of airport characteristics. The results showed that having hub airlines and Airside operations involve the movements of air carrier and commuter; terminal operations indicate the service to handle passenger of cargo. 11  expanding gate capacity improved airside efficiency, while terminal efficiency could be raised by expanding the number of gates and adopting more effective utilization system. However, the issue of whether or not the airside operations are completely separable from terminal operations was not address in the paper.  Martin and Roman (2001) chose DEA model to evaluate the technical efficiency of 37 Spanish airports in 1997. Airport outputs were measured by ATM, passenger throughput, and cargo volume; while inputs were presented by monetary cost of labor, capital, and materials. An output-oriented DEA model was used to derive efficiency scores under both increasing and decreasing return to scale assumptions. The results suggested that prior to privatization, Spanish airports suffered some inefficiency due to the public ownership.  In addition, Sarkis (2000) examined the performance of 44 U.S. airports during the 1990-1994 period using six different DEA models and reached a tentative conclusion that major hub airports are more efficient than spoke airports. Parker (1999) developed DEA model for BAA (British Airports Authority) airports between 1988/89 and 1995/96, and found no evidence for efficiency improvement because of the privatization. Abbott and Wu (2002), in order to identify the influence of price regulation, used input-oriented DEA model to measure Malmquist-TFP indices of 12 Australian airports during 1989-2000 period.  In summary, DEA method has a major advantage of lower data requirement, which makes it very attractive to researchers. While DEA is capable of providing a ranking for airport performance, the ranking is highly dependent on the chosen observations, and is very sensitive to outliers. It is widely argued that traditional DEA model is unable to distinguish among relatively efficient DMUs without additional adjustments. Adler, Friedman and Sinuany-Stern (2002) reviewed six improved ranking methods in DEA,  12  and described how they could be utilized to form a complete DMU ranking. Fried, Lovell, Schmidt and Yaisawarng (2002) introduced a three-stage DEA approach in order to net out the impact of the environmental effects and statistical noise.  2.3 Airport Studies with Stochastic Frontier Analysis Following the literatures of productivity index number and DEA, this section reviews previous studies using frontier method to measure airport efficiency. The conventional approach is to construct cost or production functions by ordinal least square (OLS) model, which implicitly assumes that all firms are successful in reaching the efficient frontier. Essentially, firms are not equally efficient and their efficiencies cannot be fully explained by measurable variables. Stochastic Frontier Analysis (SFA) solves this problem by adding an inefficiency term directly to the cost or production function. Thus, SFA method is believed to be more viable than OLS method in efficiency measurement.  Aigner, Amemiya and Poirier (1976), Aigner, Lovell and Schmidt (1977) proposed the original SFA model in efficiency measurement. Stevenson (1980) generalized previous works on SFA method and extended the distribution of inefficiency term to half-normal, gamma, and exponential statistical distributions. Battese and Coelli (1992) developed stochastic frontier model via log-linear function and applied the approach to examine the technical efficiency of paddy farmers in India. As a parametric frontier method, SFA enables researchers to conduct statistical test and distinguish between sources of the inefficiency. As a result, many researchers have used SFA method to measure the efficiency of airports.  Pels, Nijkamp and Rietveld (2001, 2003) utilized SFA model to evaluate efficiency performance of 33 European airports during 1995-1997. Probably because it is not possible to evaluate multiple outputs simultaneously, two separate translog production  13  functions were estimated: one for the output of ATM (Air Transport Movement), and the other for APM (Airport Passenger Movement), respectively. The question of whether or not these two outputs are separable was not address in this paper. Both models defined inputs as a number of physical representatives, and the stochastic inefficiency term was assumed to follow a half normal distribution, truncated at zero. In the ATM model, airports with a time restriction or slot coordination were more efficient, while time restriction and airline load factors explained the inefficiency in the APM model. A problem of this study is the omission of labor input, which makes it difficult to interpreter the results for total factors.  Oum, Yan and Yu (2008) estimated a stochastic cost frontier in translog function to measure the efficiency of 109 airports around the globe and identify the effects of ownership differences on airports’ efficiency. Assuming that airports maintain a fixed level of capital infrastructure in short term, their study defined airport output as passenger volume, ATMs, and non-aeronautical revenue; and input as labor, non-labor variable input, number of runways, and total terminal size. They further incorporated variables to the cost function which accounts for airport characteristics and time trend. A half-normal distributed inefficiency term and a statistical noise term were distinguished separately from the random deviation. The key findings of the study include that airports with mixed ownership of government majority were less efficient than 100% government owed firms, and privatization of one or more airports in cities with multiple airports would improve the efficiency of all airports.  In addition, Barros (2008) estimated technical efficiency for 27 UK airports during 2000-2005 via a stochastic cost frontier model. The author defined outputs as passenger volume and aircraft movements, while inputs as operation cost, labor and capital. The stochastic inefficiency term is assumed to follow a half-normal distribution truncated at  14  zero. Different from other SFA works, this paper introduced a specific random variable to captures the inefficiency due to airports’ heterogeneity, as airports with different sizes share little in operation and managerial strategies. The results for the two models displayed differences to some extent, but on average the homogenous model produced lower efficiency score than heterogeneous one. However, this paper measured airport capital input in an unreliable way (see Chapter 3.2.2), which reduces the credibility of the result.  2.4 Conclusion During the last two decades, airport efficiency has attracted more attention from both industry and academic perspectives. Numerous studies have measured airport efficiency by using three prevailing methods, namely Index Number Method, DEA, and SFA. Among them, DEA has become the most popular approach mainly because it requires less data.  As implied from table 2.1, most studies in airport bencbmarking only utilize a single method to measure airport efficiency. So far, only few studies have measured efficiencies by using different methodologies. Hjalmarsson, Kumbhakar and Heshmati (1996) compare the efficiency results derived from the DEA, SFA and DFA (Deterministic Frontier Analysis). Based on a pooled data of 15 Colombian cement plants observed during 1968-1988, their results showed a substantial variation across models.  15  Table 2.1 Summary of Airport Efficiency Studies Paper  Method  Hooper and  TFP  Hensher (1997) Nyshadam  TFP  and Rao  Study Scope  Outputs  Inputs  6 Australian  1) Aeronautical revenue  1) Labor input  airport during  2) Non-aeronautical  2) Capital input  1988/89-1991/92  revenue  3) Other inputs  25 European  n.a.  n.a.  1) No. of passengers  I) Runway length  2) ATMs  2) Terminal size  airports in 1995  (2002) Yoshida  EW-TFP  (2004)  30 Japanese airports in 2000  3) Tonnes of cargo TRL (2004)  Multi-attribute Assessment  47 airports in  1 )EBITDA  2004  2)Concession revenue per passenger 3)Aeronautical revenue per ATM 4)ATU per unit of asset value 5)ATU per employee 6)Operating plus staff cost per passenger  Oum, Adler  VFP  and Yu (2006)  115 airports  1) No. of passenger  1) Labor input  during  2) ATMs  2) non-labor variable input  200 1-2003  3) Non-aeronautical revenue  ATRS(2008)  VFP  142 airports in 2008  1) No. of passenger  1) Labor input  2) ATMs  2) non-labor variable input  3) Non-aeronautical revenue Gillen and Lall (1997)  DEA  21 US airports  Terminal side:  Terminal side:  during  1) No. of passenger  1) No. of runways  1989-1993  2) Pounds of cargo  2) No. of gates 3) Terminal area  Air side:  4) No. of baggage collection  1) ATMs  belts  2)Commuter movements  5)No. of public parking sports Air side: 1) Airport area 2) No. of runways 3) Runway area 4) No. of employees  16  Paper  Method  Parker (1999)  DEA  Study Scope  Outputs  Inputs  22 UK airports  1)Aircraft turnover  1) No. of employees  during  2) No. of passenger  2) Capital input  1988/89-1996/97  3) cargo and mail  3) Other inputs  during Paper  Method  Martin and  DEA  Roman  Study Scope  DEA  Inputs  37 Spanish  1) No. of passengers  1) Labor cost  airports in 1997  2) ATMs  2) Capital cost  3) Tones of cargo  3) Material cost  (2001) Sarkis (2000)  Outputs  44 US airports  1 )Operating revenue  1) Operating cost  during  2)Commercial ATM  2) No. of employees  1990-1994  3)General aviation  3) No. of gates  movements  4) No. of runways  4) No. of passengers 5) Total freight Abbott and  DEA  Wu (2002)  12 Australian  1) No. of passengers  1) No. of employees  airports during  2) Tonnes of cargo  2) Capital stock  1989-2000 Pels, Nijkamp  DEA  and Rietveld  33 EU airports during  3) Total runway length Terminal side: No. of passengers  1995-1997  (2001, 2003)  Terminal side: 1) Terminal size 2) No. of A/C parking  Airside: ATMs  positions at terminal 3) No. of remote A/C parking positions 4) No. of check-in desks 5) No. of baggage claims Airside: 1)Total airport area 2) Total runway length 3) No. of A/C parking positions at terminal 4) No. of remote parking positions.  Martin-Cejas  Deterministic  40 Spanish  (2002)  cost function  airports during  WLU  1) Labor 2) Capital  1996-1997 Barros (2008)  SFA  27 UK airports  1) No. of passengers  1) Operational cost  during  2) ATMs  2) Labor cost  2000-2005  3) Capital cost 17  Paper  Method  Pels, Nijkamp  SFA  and  33 EU airports  Terminal side: No. of  Terminal side:  during  passengers  1) No. of baggage claims 2) No. of parking positions at  1995-1997  Rietveld  Inputs  Outputs  Study Scope  Air side: ATMs  (2001, 2003)  terminal 3) No. of remote parking positions Air side: 1) Total runway length 2) No. of parking positions at terminal 3) No. of remote parking positions  Oum, Yan and  SFA  Yu (2008)  109 airports  1) No. of passengers  1) Labor input  during  2) ATMs  2) non-labor variable input  200 1-2004  3) Non-aeronautical  3) No. of runways  revenue  4) Terminal size  Coelli and Perelman (1999) compared three alternative methodologies employed in efficiency measurement: (1) parametric frontier using linear programming approach; (2) parametric frontier using corrected OLS method (including SFA); and (3) non-parametric piece-wise linear frontier using DEA. When applying them to a pooi data of 17 European railways from 1988 to 1993, the technical efficiencies emerged from the three methods displayed no substantial differences, with positive and significant correlations between each other. The authors then claimed that a researcher could safely select one of these methods without too much concern for their choice having a large influence upon results.  Pels, Nijkamp and Rietveldet (2001) compared the efficiency results derived respectively from DEA and SFA. Based on the dataset of European airports, the results emerged from the two methods were reasonably consistent, despite the fact that SFA produced less dispersed efficiency scores.  In conclusion, most previous studies in airport benchmarking used a single approach to 18  measure efficiencies. They vary further in defining outputs, inputs, and research scopes. As stated in Oum, Tretheway and Waters (1992), productivity studies of transportation industry, using different measures of outputs and inputs, cannot be directly compared with each other. So far, few effects have been directed towards the comparison between different methodologies and their empirical results. Therefore, this study aims to offer the first step toward filling this gap.  19  3. METHODOLOGY As discussed in the literature review, three prevailing methodologies have been widely used to measure airport efficiency during the past two decades: (1) Index Number Method, (2) Data Envelopment Analysis, and (3) Stochastic Frontier Analysis. However, as originated from different backgrounds, these three methods vary significantly in many aspects. Therefore, this chapter reviews and develops model for each of the above method, and compares some of their notable characteristics.  Before elaborating methodologies, it is necessary to clarify the concept of productivity and efficiency. In transportation economics, productivity of a firm is usually defined as the ratio of the output(s) that it produced to he input(s) that it used (Coelli et al. 2005). In addition, efficiency or technical efficiency 5 refers to the ability of a firm to use its inputs efficiently. The efficiency of a firm could be defined as either maximizing outputs with a constant level of input, or minimizing inputs under a fixed level of output. To avoid confusion, this study uses the terms of productivity and efficiency interchangeably. The remaining part of this chapter describes the three methodologies and attempts to compare them in some aspects.  3.1 Index Number Method As a non-parametric approach, Index Number Method directly defines the productivity as output index over input index. The method is easy to conduct for single output and input firms. However, airports utilize multiple inputs such as labor, capital, and other resources to produce various services for both airlines and passengers. Since the mid of last century, a number of productivity index numbers have been developed for multiple output and input firms. This study chooses the multilateral index number method proposed by Caves, Battese and Coelli (1992) defines the concept of technical efficiency of a given firm as the ratio of its mean production to the corresponding production if the firm utilized its levels of inputs most efficiently; therefore it is a number between zero and one. 20  Christensen and Diewert (1982), as it is the dominant method with strong underlying economic support.  Based on whether to incorporate capital input, multilateral productivity index distinguishes between Total Factor Productivity (TFP) and Variable Factor Productivity (VFP). In this section, we first develop the TFP model, then discuss the measurement of airport capital input, which leads to the use of the VFP model.  3.1.1 TFP Model A typical multilateral TFP model, which involves all factors of outputs and inputs, is defined as: in TFPk  —  in TFPJ.  =  (in Y,  —  ln Y.)  —  (in Xk  —  in X)  k  2  =  R+R. y ‘in 2 ln+  R+R ‘ 2  Y ‘in —u i’;  + X 1 . 2  j;, (3.1)  where TFPk is the productivity of kth firm Y k and X,k represent the 1 ,th  firm, respectively; R,k and W,k are the weights for the  th 1  output and input of  output and input of kthl firm  respectively; A bar over weights presents sample arithmetic mean, while a tilde demonstrates geometric mean.  As implied from equation (3.1), the TFP index is formed by a series of binary comparisons between each observations and the sample mean. The productivity of a firm is defined as the ratio of aggregate outputs over aggregate inputs. In order to utilize the model, output and input quantities have to be obtained, as well as the aggretation weight for each output and input. Ideally, revenue and cost elasticities should be used as weights for output and input, respectively. However, as those numbers are usually not obtainable for most industries, including airports, Diewert (1992) suggests using revenue and cost  21  shares as approximations. This adjustment comes with further assumptions on constant return to scale across all outputs and revenue maximization (or cost minimization) behavior of sample firms. In practice, these assumptions are reasonable for relatively large firms under commercial management, which represents most of our sample airports in North America. Further, the TFP model is capable of applying with both time series and cross-sectional data, and it allows for comparing the productivity of the same firm in different periods, or different firms across time. As there are large controversies in measuring airport capital input, we discuss the issue as following.  3.1.2 The Measurement of Capital Input Capital input is important for airports as it involves the expenditures on physical infrastructures, such as the expansion of terminals, runways, and the use of land. In order to incorporate it into the multilateral TFP model, capital input has to be divided into price and quantity. The ideal approach in measuring capital input is through the perpetual inventory method, which is introduced by Christensen and Jorgenson (1969). Specifically, capital quantity is measured by perpetual inventory estimates of capital stock, while capital price is captured by capital service price. Although the method is theoretically correct, it is extremely data intensive  —  it requires data on investment expenditure, price  index, retirement patterns for different assets, etc. The method is nearly impossible to conduct across firms in different countries because of the different accounting and taxation systems.  For airport sector, the measurement of capital input becomes even more complicated. Since airports are usually operated under different ownerships and regulatory frameworks, they differ radically in measuring and depreciating asset value. Some publicly owned airports adopt government accounting system and consider land as their own asset. In addition, even for depreciating the same asset, it is still far from reaching a common  22  standard: airport runways are deprecated for up to 100 years in U.K., while the duration is 30 to 49 years in Amsterdam, and 20 years in Paris. Moreover, we are unable to estimate the replacement value for airport assets, because there is no second market exists. Although airports release their accounting information embodied with depreciation rules, it has little relation to actual use of resources. Capital cost can be dominated by interest cost and, in turn, can be more a function of the way the airport has structured its finances (Hooper and Hensher, 1997). In general, although the ideal approach of measuring capital input has been introduced, it is very difficult, if not impossible to conduct it for airports.  Because of the difficulty of employing the perpetual inventory method, many researchers have tried alternative ways to estimate capital input of airports. Barros (2008) defined price of capital-premises as amortizations divided by total assets and capital-investment as cost of long-term investment divided by long-term debt. Martin and Voltes-Dorta (2007) used the ratio of capital cost over ATM to present the capital price. Craig, Airola and Tipu (2005) measured capital price as a weighted index of runways and terminal price index, which are approximated by Federal Highway construction price and office building construction price, respectively. Martin-Cejas (2002) derived capital price as the amortization (depreciation cost) divided by WLU. However, all of these approaches have two major shortcomings. First, amortization and depreciation cost do not capture the capital flow of airports, as public owned airports may not have to depreciate certain assets. Second, neither ATM nor WLU is a well-defmed indicator for capital quantity. Capital input measured by these alternative approaches is thus not reliable.  In summary, the ideal method of measuring capital input is too complicated to implement, especially across airports in different countries. Other approaches in estimating capital input are also proved to be unreliable. Hence, without an international standard or an immense researching budget and time, incorporating capital input into airport TFP  23  analysis is nearly impossible.  3.1.3 VFP Model Following the discussion in the previous section, this study chooses to measure the variable factor productivities of airports while keeping capital input as constant. As stated previously, the VFP method is viable for measuring airport efficiency in short-to-medium term, and has been utilized by a number of studies in airport benchmarking, including ATRS (2008) and Oum, Adler and Yu (2006).  The multilateral VFP model is similar to the TFP model defined in equation (3.1). The only difference is not to incorporate capital as a separate input. The VFP model used in this study is defined as follow: )=R 1 InVFPk —1nVFP =(lnY -lnY)-(lnXk _lnX I  R±]  k 1  ln  I  Xiki k X 1 Y  2  2 (3.2)  The notations and other properties of the VFP model are the same as those described in TFP model.  3.2 Data Envelopment Analysis (DEA) The second methodology of efficiency measurement examined in this study is the Data Envelopment Analysis (DEA). Similar to index number method, DEA is also a non-parametric approach. It uses linear programming to construct a piece-wise production frontier, and relative efficiency of each Decision Making Unites (DMUs) is calculated according to the frontier. Observations that lie on the production frontier are defined as Pareto efficient DMUs, while those not on the frontier are inefficient observations. Therefore, the frontier, which has been achieved by at least two DMUs, is  24  considered as a sign of relative efficiency. As originated from the study of operations research, DEA was first proposed by Charnes, Cooper and Rhodes (1978), in which they defined DEA as: A mathematical programming model applied to observational data provides a new way of obtaining empirical estimates of external relations such as the production functions and/or efficient production —  possibility surfaces that are a cornerstone of modern economics.  Figure 3.1 Illustration of flEA  Input x2 Implied inefficiency ofDIvlUD  DEA frontier  Y (Fixed output level)  0  Input x 1  Figure 3.1 illustrates the DEA efficiency in a fixed output and two input situation. In this case, firms A, B and C forms the DEA frontier since no other firms could produce more output with less input. They are thus regarded as 100% efficient. Firm D is inefficient because we could proportionally reduce its inputs without affecting the output by moving point D along the radical line to D’ locating on the frontier. The relative efficiency of D is OD’ defined as-=-. OD  Similar to productivity index numbers, DEA is also capable of measuring efficiencies for  25  multi-output/input firms. Under the assumption of convex production set, the original DEA model is defined as: Max V X 1 SI.  u,y.  (3.3)  V X U, V  0  where y, and x 1 are the vectors of output and input for  th  observation, respectively; u and v  are vectors of weights to be determined by the solution of the problem. DEA efficiency of a DMU is actually a peered efficiency. In the case of airport benchmarking, the efficiency of an airport is evaluated relatively to the performances of airports with similar size.  Based on the target variable, DEA method distinguishes between input-oriented and output-oriented models. Specifically, output-oriented DEA should be utilized if the objective is to examine the efficiency by increasing output under a constant level of input. Conversely, if the objective is to reduce the input level as much as possible without changing the output level, the input-oriented DEA is more suitable. In this study, we choose the input-oriented model because airports have more control on their input side rather than output side. First, air travel demand is a derived demand that is tied directly to other economic activities. Thus airports have little influence in generating air travels. Second, airport demand is believed as price inelastic, thus adjusting airport charges may not affect the output substantially. As a result, the volume of airline traffic is exogenous to the control of airport managers. On the other hand, airports have some level of control on their inputs. For example, they could decide how many employees to hire and how much services/goods to purchase. Therefore, an input-oriented DEA model is more appropriate in measuring airport efficiency, and has been chosen by most previous studies, including Gillen and Lall (1997), Abbott and Wu (2002), Pels, Nijkamp, and Rietveld  26  (2001, 2003), etc.  Besides different orientations, DEA could further be classified between constant return to scale (CRS) and variable return to scale (VRS). After two steps of dual transforms from equation (3.3), the input-oriented CRS model is defined as:  t9 82 Min s.t. —y + 1 Y20 1 x  —  X2  (3.4) 0  where 6 is the efficiency score; Y is M X I output matrix; X is N X I input matrix; Y , X is 3 the vector of outputs and inputs for the JIh firm, respectively; and ) is a I X 1 vector of constants. Further, by adding a convex hull constraint on  ,  the CRS model transfers to  VRS model in (3.5). Because of the existence of scale inefficiency, the DEA efficiency under CRS is always no larger than that under VRS. For the airport sector, most previous studies have manifested that the output scale only exists in airports with less than 3 million WLU, for example Doganis and Thompson (1973, 1974), Main et a! (2003), and Jeong (2005). As all of our sample airports exceed that limit, we choose to measure airport DEA efficiencies by CRS model. 9 MinOA s.t. -j’  +Y%0  xi— X2  0 (3.5)  2O  27  In addition, DEA method demands data of output and input quantities, and is capable for analysis across firms and/or time periods. Nevertheless, DEA has a disability in distinguishing between Pareto efficient firms that lie on the production frontier. Besides, the method is very sensitive with data outliers, adding or removing one observation might considerably change the entire ranking results.  3.3 Stochastic Frontier Analysis (SFA) The third method this study utilizes to measure airport efficiency is stochastic frontier analysis (SFA). Different from productivity index number and DEA, SFA specifies the form of a production or cost function, and identifies the inefficiency as a stochastic disturbance. First proposed by Aigner, Amemiya and Poirier (1976), the general form of stochastic frontier model is as follows: =  f(x,;fl)exp(V  —  U,)  (3.6)  where 1’, represents the output of the ith firm; f(x ;/3) is the deterministic core function of a 1 input vector x and a unknown parametric vector  fi;  V is a normally distributed random  variable that represents the effects of non-observable explanatory variables and random shocks; U 1 is a non-negative random variable represents inefficiency, and is assumed to follow one of half-normal, exponential, or gamma distribution.  As implied from equation (3.6), SFA explains output by a vector of inputs and a stochastic disturbance, which consists of two parts: a stochastic inefficiency and a traditional ‘noise’ term. A single output and input SFA is shown in Figure 3.2. Different from traditional OLS frontier method, SFA breaks down the variation between each data point and the fitted value into two components: V 1 and U. While V, could be either positive or negative, U, is always positive. In this case, the technical efficiencies of firm A and B are identified as TEA=qA!qA* and TEBqBIqB*, respectively.  28  Figure 3.2 Illustration of SFA  determinisLic frontier Yi=f(i .X)  noi::cffccsc effect :ect  With specification of the deterministic function, a Translog SFA-Production function form used in this study is developed as: in Yi =  /3 in X  + j=1  +  fl/k  in X, in X  +  (V,  —  6 u,)  (3.7)  j=l k=1  where Y, is the aggregated output index for airport 1; X is the jth input; V 1 is assumed to follow distribution N (0, 2 1 is assumed to follow N (pt, au) where,u?:0. The technical cr v ); U efficiency of airport i is then derived by the ratio of its mean output over the output if it uses inputs most efficiently. TE I  =  E(IIU,,X,) E’IU = 1 O,X,)  =exp(—U)  (3.8)  The estimation of production function in (3.7) uses index number theory to aggregate multiple outputs into a single output index. As a result, SFA-Production function requires The efficiency scores measured in function (3.7) are gross measurements. In order to estimate airport managerial efficiency, addition variables of airport characteristics should be added to the function. 29  data on output/input quantities and output revenue shares, and the method is capable of dealing with both panel and cross-sectional data.  3.4 Comparison of Methodologies After elaborating the three methodologies, it is necessary to compare some of their notable characteristics before estimating empirical results. This section focuses on comparing assumptions, data requirement, strength and weakness between each method primarily on empirical perspectives.  First, the three methodologies are based on different sets of assumptions. The index number method assumes that firms are allocative efficient and under constant return to scale. In order to use revenue/cost shares as aggregating weights, it further assumes an entity behavior of revenue maximization (or cost minimization). On the other hand, DEA assumes the continuity and convexity of production set. Moreover, the SFA-Production function form uses an aggregated output index, which assumes the existence of a continuous and convex production function, as well as existence of a consistent output index. The SFA method further assumes a particular form of inefficiency distribution: usually one of half-normal, exponential, and gamma distribution.  In addition, the required data for each method are different, which have been described individually along the previous discussion. In general, productivity index number method requires the highest amount of data, followed by SFA (production function), while DEA requires the least amount of data (see Table 3.2).  Furthermore, as originated from different theories and backgrounds, these three methods have developed their unique pros and cons. Index number method and SFA-Production function both have support of underlying economic theories, while DEA is an adopted  30  approach from operations research and is rather a measurement technique than an explanation of inefficiencies. A major disadvantage of index number is the difficulty in measuring capital input. While DEA is able to incorporate physical measure of capital input such as terminal size, number and/or length of runways and thus, is easy to perform with less demanding data. However, as stated before, DEA is very sensitive to data quality and outliers. Adding or removing one observation might considerably change the ranking results of all other DMUs. As illustrated in Figures 3.3a and 3.3b, after dropping an efficient DMU (firm A), the efficient frontier moves accordingly, and thus, the ranking of the remaining observations changes. Table 3.1 listed the substantial differences of DMU ranking before and after the removal. Firm F, which is among the least efficient observations in the former case, becomes more efficient than firm D and G after the removal.  Figure 3.3a DEA Illustration Original -  1nput F  D  DEA fiier  Inp*xL  31  Figure 3.3b DEA Illustration  —  Removing an Efficient DMU  1npQ  DEA fmrd,er  Jnp xi  Table 3.1 Differences between DEA Rankings DMU A B C D E F G  Pre-Ranking Post-Ranking 1 1 1 1 1 4 4 7 1 6 3 5 5 -  As DEA efficiency index lacks “transitivity”, DEA airport efficiency rankings can change substantially as one adds or drops airports from the sample. On the other hand, index number methods preserve the relative index values or rankings, even when one adds or drops one or more airports from the sample (DMUs).  In addition, as the only parametric method, SFA-Production function involves a specification of frontier function, which enables it to conduct hypotheses test and distinguishing the sources of efficiency growth. Further, as SFA does not assume that all firms are efficient, it allows existence of systemic inefficiency in the error terms, and 32  does not restrict the combined error term (which includes inefficiency distribution) to be assumed independently and identically distributed (i.i.d.). However, the specification of SFA frontier requires extremely complicated computation, which makes the method difficult to communicate with industry audience. At the same time, the SFA-Production function form cannot be applied to firms with more than a single output unless one aggregates multiple outputs into an output index, which leads the researcher to accept all of the neoclassical assumptions for computing index numbers. This is because readily available econometric techniques do not allow us to identify more than one dependent variable in one equation (i.e., multiple outputs in one production . 7 functio n)  In conclusion, Table 3.2 summarizes the notable characteristics of Index Number Method, DEA, and SFA-Production function form. With different assumptions and data requirements, the three methodologies have developed their unique strength and weakness. Table 3.2 Comparison of Index Number Method, DEA and SFA Index Number Method • •  Assumption  1) a a Requirement  •  • • •  • • •  Strength • •  CRS Allocative efficiency Entity revenue maximization Quantity Price Revenue/cost Shares Economic support Allow to compare two firms in different periods No frontier specification Easy to communicate  DEA •  •  SFA-Production Function  Continuous & convex production set  •  Quantity  •  •  •  •  •  •  Low data requirement Able to include capital input No frontier specification  •  •  •  Inefficiency distribution Continuous & convex production set Quantity Output revenue share Account for statistical noise Able to conduct hypotheses test Able to distinguish sources of productivity growth  It is possible for one to estimate stochastic frontier cost ftinction in order to deal with multiple output production situation. However, this is beyond the scope of this thesis research. 33  Index Number Method • •  Weakness  High data requirement Very difficult to incorporate capital input  DEA •  •  •  Less solid background, but a measurement technique Results are sensitive to outliers and to the set of DMUs included in the study Incomplete ranking  SFA-Production Function •  • •  •  Single output assumption or use of an aggregate output index Frontier specification High computational requirements Difficult to communicate with industry  34  4.  DATA  CONSTRUCTION  AND  IMPACT  ON  AIRPORT  EFFICIENCY During the previous discussions, three methodologies of efficiency measurement have been described in general situations. In order to apply them to measure efficiency of airports, this chapter identifies outputs and inputs of sample airports, and constructs data accordingly. Further, we explore some issues in data construction aims to improve the accuracy of efficiency measurement. Finally, by examining the impact of these improvements on measured efficiency, the importance of data construction on benchmarking studies is revealed.  4.1 Airport Outputs and Inputs One fundamental issue in measuring airport efficiency is to distinguish airport activities, and define outputs and inputs appropriately. In the past, the traditional function of an airport is to assist airlines to transport passengers and cargo, and charge them by using of facility and service. Along with the airport privatization, commercial activities, such as retailing and car parking, have become another important part of airport activities. However, as airports around the world are under different ownership and regulatory frameworks, the scope of airport activities varies widely across different airports. As stated in the literature review, previous studies in measuring airport efficiency have not reached an agreement in defining the outputs and inputs.  Nevertheless, our sample airports in U.S. and Canada are under similar ownership and regulatory framework, which makes it possible to define a common set of airport activities. U.S. airports are either owned by government departments or airport authorities (Oum, Adler and Yu, 2006; Beaudoin, 2006). The ownership of Canadian airports is similar to airport authorities in U.S. In addition, although our sample airports are publicly involved, they adopt many business strategies, and most of them are self-financing. U.S. 35  FAA (Federal Aviation Administration) requires airports under government department to be financially independent and self-sustained in order to improve efficiency. Authority-run airports are also responsible for their own financial and operation status, as they operate in a manner of corporation enterprises and reinvest their profit into airport assets. Therefore, all of our sample airports have incentives to breakeven revenue and cost by adopting commercial strategies, and thus we are able to define a uniform set of outputs and inputs.  First, as airports typically charged separately from handling aircrafts and passengers, numbers of aircraft movements (ATMs) and passenger volume are two major outputs of an airport. Some studies argue that ATMs might be correlated to passenger volume  —  thus  is not an independent output. In fact, airlines are able to change the number of flights by adjusting load factor, seating arrangement, and the size of aircraft for their business purposes. Therefore, ATMs is not necessarily an endogenous output. Further, air freight is another output for airport. However, as most cargo services are handed directly by airlines or third-party logistics companies, airports only receive small amount of usage fees for leasing space and terminal. In addition, as cargo service only accounts for a small percentage in total airport revenue, most airports do riot report it separately. As a result, air cargo is not interpreted as an individual output in this study.  The above aeronautical outputs are, in most cases, under government regulation. Airports further rely on a number of non-aeronautical and commercial activities to bring in extra revenues, such as duty free shops, beverage, car parking, concession, etc. Such leasing and outsourcing activities offer flexibility to managers and allow them to respond efficiently to market forces. Although non-aeronautical activities are different from the traditional airport services, they have become increasingly important and accounts for somewhere between 30% and 80% for most of our sample airports in 2006 (see Figure  36  4.1). As discussed in Oum, Adler and Yu (2006), aeronautical and non-aeronautical activities among most airports are not separable, and any productivity computed without including the non-aviation service output would lead to severely biased results. Therefore, we define non-aeronautical revenue as the third output.  Figure 4.1 Proportion of Non-aeronautical Revenue in 2006 % Non-aeronautical Revenue  Q  -  —,  n.N  F  x  00  Z  -  F-  -,  ut X <  Source: ATRS 2008  On the input side, airports have to exploit certain resources to generate outputs. First, labor is one of the most important inputs. In 2006, personnel expenses accounts for somewhere between 15% and 70% of total operating cost for our sample airports (see Figure 4.2). Since most airports contract out part of their services, some employees may be hired by outsourcing companies rather than airport operators. To avoid double accounting, we define labor input as the employees directly hired by airport operators. In addition, airport inputs also include the purchased goods and materials, and purchased services. Ideally, these two inputs are separable; however few airports actually provide information for each cost. Therefore, we combine the two inputs and term it as ‘soft cost input’. Specifically, soft cost input is a catch-up input of non-labor, non-capital operation cost. The concept of soft cost input has been widely used in previous studies, i.e. Oum, 37  Adler and Yu (2006), ATRS (2008), etc. Finally, since this study examines how efficiently airports produce outputs at a given level of capital infrastructure, capital resource is not incorporated as a separate input.  Figure 4.2 Proportion of Labor Cost in 2006 % of labour cost 80% 70% 60% 50% 40% 30% 20% 10%  -  Source: ATRS (2008)  To sum up, this study defines airport outputs as: •  Passenger  •  ATMs  •  Non-aeronautical Revenue  and inputs as: •  Labor  •  Soft cost input  In the next section, we construct data for the above outputs and inputs respectively, and try to explore some issues in data construction.  4.2 Data Construction In order to measure airport efficiency by all three methodologies (Index Number Method, 38  DEA, and SFA-Production function), we have to acquire quantity, price, and revenue/cost shares for each output and input. The relationship between these variables is showed in equation (4.1). The remaining part of the section constructs data for each outputs and inputs respectively. We further explore some data issues in order to improve the accuracy of efficiency measurement. Quantity X Price  =  Revenue (Cost)  (4.1)  4.2.1 Passenger The annual passenger throughput handled by an airport is used to represent the output quantity. The revenue of passenger output is measured by airport passenger charges, while the price is derived accordingly by equation (4.1).  4.2.2 ATMs The number of aircraft movements and the airside revenue (airport landing fees) are collected to represent this output. The output price is then defined as the ratio of airport landing fees over number ofATMs.  4.2.3 Non-Aeronautical Revenue Non-aeronautical revenue is the sum of revenues from all non-aviation activities. In order to derive the output quantity, most previous studies use Purchase Power Parity 8 (PPP) as a price index to deflate total non-aeronautical revenue. Using PPP index assumes that the price of purchasing the same amount of non-aeronautical service is constant across different cities inside one country. However, it is not always true. Suppose Kansas City airport-MCI and New York airport-JFK generate the same amount of parking revenue. Then the quantities of this output will be same if deflated by PPP. However, as the hourly parking fee at MCI is usually lower than at JFK, JFK should be producing less parking The Purchasing Power Parity (PPP) uses the long-term equilibrium exchange rate of two currencies to equalize their purchasing power. PPP equalizes the purchasing power of different currencies in their home countries for a given basket of goods. 39  output than MCI. This output difference will in turn influence productivities between the two airports. In general, using PPP to deflate non-aeronautical revenue favors airports in relatively expensive cities.  In order to address the issue, we use a city-based Cost of Living Index 9 (COLI) to capture the price differences between each airport in U.S. For Canadian airports, such living index does not exist; therefore city-based CPI is used to adjust Canadian airports’°. As illustrated in Figure 4.3, there are significant differences between PPP and COLI indices. The two airports in New York City, JFK and LGA, have the highest level of living expenses, while the airports in Houston and Detroit are among the lowest. As a result, when measuring the output quantities of non-aeronautical revenue by each index, considerable differences still exist (see Figure 4.4).  Figure 4.3 Comparison of PPP and COLT Indices 2.0  —.—  National Level PPP City Level COLI  1.8 1.6 1.4  IL  1.2 1.0 0.8 0.6  The Cost of Living Index (COLI) is a composition index to measure relative price level for consumer goods and services in areas for a mid-management standard of living. The overall index (100%) is composed by Grocery items (13%), Housing (29%), Utilities (10%), Transportation (10%), Health care (4%), and Mics. goods and services (35%). 10 The COLT and CPI indices are linked with US-Canada PPP exchange rate in 2006: 1US$1.245CA$ 40  Figure 4.4 Output of Non-Aeronautical Revenue by using PPP and COLI Non Aero Revenue in ppp  350,000,000  LNonAreo Revenue  mCOLj  300,000,000 250,000,000 200,000,000 150,000,000 100,000,000  ILh  øu  50,000,000 0  4.2.4 Labor Input The total personnel expenses and number of directly hired employees are collected for each airport, respectively. The price of labor input is then derived according to equation (4.1).  4.2.5 Soft Cost Input Soft cost input is a catch-up cost of all non-capital, non-labor operation cost. The conventional approach in measuring soft cost input suffers from two major issues: (1) it uses PPP as a price deflator; and (2) the composition of soft cost is not consistent across airports. This study addresses these issues by switching to a regional price index and ducting snow removal cost.  (1) Adjustment of price index Similar to non-aeronautical revenue, PPP has also been used to deflate soft cost input in most previous studies. The problem of using this national aggregate price index still exists. One could expect that airports in relatively expensive cities like New York and San Francisco have to pay more to purchase the same amount of service (or goods) than 41  airports in less expensive cities, such as Salt Lake City and Austin. Using PPP as a price index raises the actual use of this input for airports in expensive cities, thus lowers their efficiencies. Therefore, we use regional based COLT to capture the price differences across cities. The quantities of soft cost input by PPP and COLI are compared in Figure 4.5. As expected, after using COLT, JFK and LGA significantly reduced their amounts of this input by nearly 50%, while the input raised up for IAH, DTW, and other less expensive airports.  Figure 4.5 Input of Soft Cost by using PPP and COLI soo,ooo,ooo  —.—  Soft Cost Input inPPP 1 Soft Cost Input in COLI  a  400,000,000  c  -  <‘  i- C) <  z  - -  0 Z  o><  “  x  < 00  <  >  -0 — N .-  )  0 Z F—  <0 F.) F.)  <)  N  -  F.)  (2) Deduction of snow removal cost Because of different geographic locations, airports in cold climate may have an additional snow removal cost. These airports suffer extra expenses on hiring additional staff, purchasing special equipments and supplies to remove the snow on airport field. In fact, this cost could be quite significant for some airports in our sample. For example, in 2006 snow removal cost accumulated to US$9.8million for JFK and over US$l0million for Denver-DEN. Deducting this cost is necessary in order to create a fairer comparison.  42  In practice, a number of airports do not report snow removal cost separately. Therefore, we use regression analysis to estimate the missing data in our sample. Three independent variables are defined as follow: (1) total runway length is used to represent the area where airport need to shovel snow; (2) the amount of snowfall is the direct reason for such cost; and (3) the cost of living index (COLI) captures the regional price differences in terms of hiring labor and purchasing utility. Equation (4.2) exhibits the detailed regression model, all variables are in logarithm forms.  Snowfee,  +  RunwayLength /3 1  +  SnowAmount, 2 /J  +  C0L1, 3 /3  +  (4.2)  ‘  In order to obtain a better result, we include historical observations as many as possible. The final regression result is showed in Table 4.1:  Table 4.1 Regression Result of Snow Removal Cost OLS (log-log) -  Coefficient  Standard Error  t-Stat  Runway Length  1.381  0.377  3.66*  Snow Amount  0.786  0.143  5.48*  COLI  3.815  0.601  6.35*  Intercept  -0.954  3.3 12  -0.29  2 R Observations: *  0.68 61  The coefficient is significant at the 95% level.  As expected, the coefficients of all three independent variables are positive and statistically significant at 95% confidence. The regression has a R-square of 0.68, meaning most variations of the dependent variable (Snow Cost) have been explained by these exploratory variables. Based on the result, we estimate the missing data in our sample. 43  To sum up the discussion of soft cost input, two adjustments have been implemented to improve data credibility: apply regional price index, and deduct snow removal cost. The combined effect of these two adjustments is showed in Figure 4.6. The differences are not negligible: airports like JFK, SF0, and LAS significantly reduce soft cost input after the adjustments. Their efficiencies are expected to increase compared to conventional approach.  Figure 4.6 Input of Soft Cost before/after Improvements -  —  —.—  Soft Cost Input beibre Improvements  -  -  Soft Cost ILoPP 400,000,000  ::::J’\  -  -  --  ..  .  .  300,000,000  AVAAAJJLc  We conclude this section by summarizing the data construction in Table 4.2. For all three outputs and two inputs, we construct data on quantity, price and revenue (cost) shares. In order to improve the quality of data, regional price indices have been used instead of national aggregate prices, and snow removal cost has been deducted from total cost. The effects of these data improvements on measured efficiencies will be examined in the next section.  44  Table 4.2 Summary of Data Construction Quantity (1)  Price (2)  Revenue (Cost) (3)  Passenger No. of passengers  (3)1(1)  Terminal charge  ATM No. of ATMs  (3)1(1)  Airside charge  Output  Non- aeronautical Revenue (3)/(2)  COLI-CPI Non-aeronautical revenue  Input Labor No. of employee  (3)1(1)  Personnel expenses Non-labor, non-capital  Soft Cost Input (3)1(2)  COLI-CPI operating cost  4.3 Impact of the Data Improvement on Gross VFP Following the discussion in the previous section, using regional price indices and deducting snow removal cost improve the quality of data. In this section, we examine the effects of these data improvements on measured efficiency. For this purpose, different sets of gross VFPs with conventional dataset and improved dataset are estimated and compared, respectively. ’ 1  4.3.1 Impact of Regional Price Indices We examine the impacts of using regional price indices on airport efficiencies by first using them to deflate non-aeronautical revenue, then soft cost input, and lastly both of them together.  4.3.1.1 Adjust Non-Aeronautical Revenue by Regional Price Indices Keeping everything else unchanged, we develop two sets of gross VFPs: one uses PPP as price index to deflate non-aeronautical revenue, the other use COLI. As illustrated in  It is also possible to examine the data impact on DEA or SFA scores; however that is beyond the scope of this study. 45  Figure 4.7, there are differences between the two sets of VFP scores. Airports in less expensive cities moved up their efficiency scores after the adjustment, while most airports in expensive cities suffered efficiency scores downgrade after the adjustment.  Figure 4.7 Impact of Adjusting Non-Aeronautical Revenue on Gross VFP r  25  Revenue) •VFP Scores (COLI as Price of Non aero Re.enue)  -_____________  II 1.5  I  —  Iii.  —  I  ••  •  1.0 .—-—  I  II  0.5  0.0 OQ  For a number of airports, the influences of this adjustment are significant (see Table 4.3). Dallas-DFW raises its ranking by 13 places, while Cleveland-CLE, Kansas City-MCI, and Austin-AUS also advanced considerably. On the other hand, airports in New York (LGA), Honolulu, and Orange County experienced larges ranking decline under the adjustment. We find that airports with significant variations are mostly middle-ranked airports, the rankings of top and bottom ranged airports are similar across the three methods  46  Table 4.3 Significant Impact of Adjusting Non-Aeronautical Revenue  DFW CLE MCI AUS PIT SAN SEA SNA HNL LGA  VFP Score  VFP Score  with PPP  with COLT  0.843 0.900 0.839 0.854 0.746 0.903 1.063 1.085 0.912 1.025  0.941 0.970 0.933 0.928 0.797 0.778 0.939 0.943 0.755 0.768  Pre-Ranking  Post-Ranking  46 40 47 44 52 39 26 23 38 28  33 29 36 37 45 48 35 32 51 50  4.3.1.2 Adjusting Soft Cost Input by Regional Price Indices Similar to non-aeronautical revenue, soft cost input is also adjusted by regional price index. In order to examine the effect of this adjustment on measured efficiency, two sets of gross VFPs are estimated, using PPP and COLT respectively. Charlotte-CLT is the most efficient airport in the former case, while Atlanta-ATL overtakes its place after the adjustment. As implied from Figure 4.8, there are considerable differences between the two sets of VFP scores. As expected, airports in expensive cities moved up in either VFPs or rankings after the adjustment. The two airports in New York, JFK and LGA, advanced their rankings by 17 and 22 places, respectively (see Table 4.4). Meanwhile, airports in less expensive cites, for example San Antonio, Houston, and Cincinnati, had significant ranking decline after the adjustment. In general, the variations are again less significant for airports on the top and lower ends than in the medium set.  47  Figure 4.8 Impact of Adjusting Soft Cost Input on Gross VFP VFP Scores (PPP as Price of Soft Cost Input)  2.0  I VFP Scores (COLT as Price of Soft Cost Input)  1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0  Table 4.4 Significant Impact of Adjusting Soft Cost Input VFP Score  VFP Score  with PPP  with COLI  0.912 1.025 0.903 0.554 1.063 1.070 1.035 1.214 1.163 0.985  1.152 1.556 1.083 0.865 1.238 0.978 0.970 1.073 1.003 0.895  Pre-Ranking Post-Ranking HNL LGA SAN JFK SEA [ND DEN CVG JAR SAT  38 28 39 61 26 24 27 12 16 29  15 6 20 44 10 34 37 23 30 43  4.3.1.3 Combined Effects of Regional Price Indices After examining the effects of adjustment of non-aeronautical revenue and soft cost input separately, we further use the regional price indices to deflate these two output and input together and  examine  the  combined  effects.  As  stated previously,  deflating 48  non-aeronautical revenue inclines to raise the VFP scores for less expensive airports, while deflating soft cost input favors expensive airports. Thus, when adjusting them together, the effects of using regional price indices on overall efficiency scores canceled out to some extent. As illustrated in Figure 4.9, the two sets of VFP scores are quite consistent. Figure 4.9 Impact of Regional Price Indices on Gross VFP 2.0  L  ..  1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0  Nevertheless, the combined effects still have significant impacts on certain airports. We find that airports in New York, Oakland, and San Diego, all increased their VFP scores and rankings after the adjustment. On the other hand, airports in less expensive cities, such as Phoenix, Houston, and Dallas experienced considerable declines in efficiency and ranking. These changes possibly show that the effect of adjusting soft cost input (increase the efficiencies of expensive airports) outweighs the effect of adjusting non-aeronautical revenue (increase the efficiencies of inexpensive airports) for our sample airports.  Table 4.5 Significant Impact of Adjusting Price Indices  LGA OAK JFK  Conventional VFP 1.025 0.799 0.554  Price Adjusted VFP 1.166 0.837 0.668  Pre-Ranking Post-Ranking 28 50 61  16 46 57 49  SNA SEA STL ALB DFW IAH PHX  Conventional VFP 1.085 1.063 0.730 0.655 0.843 1.163 1.092  Price Adjusted VFP 1.120 1.094 0.712 0.650 0.823 1.108 1.082  Pre-Ranking  Post-Ranking  23 26 53 57 46 16 20  20 24 55 59 49 21 25  Furthermore, as the original purpose of airport benchmarking is to distinguish “winners and losers” in the industry, we further examine separately the price impact of the top and bottom 10% airports. As implied from Table 4.6, for the top airports, except for the ranking exchange between CLT and ATL, airports remain the same rankings after the adjustment. For bottom airports, the differences are also minor: the only change is the advance of JFK.  Table 4.6 Impact of Adjusting Price Indices on Top/Bottom Airports  Top 10% Airports  Bottom 10% Airports  Before Adjustment CLT 1.887 ATL 1.862 RDU 1.694 MSP 1.692 YVR 1.550 RIC 1.455  After Adjustment ATL 1.856 CLT 1.849 RDU 1.703 MSP 1.692 YVR 1.565 RIC 1.465  ALB SJC BWI MSY JFK MIA  JFK SJC ALB BWI MSY MIA  0.655 0.646 0.62 1 0.559 0.554 0.523  0.668 0.658 0.650 0.626 0.534 0.522  50  of gross VFPs with and without snow removal cost. Figure 4.9 compares the efficiency changes because of the deduction. Except for a number of airports with large amount of snowfalls, the impact of snow cost is not significant. As listed in Table 4.7, four Canadian airports advanced their rankings after the adjustment, and these relatively large changes all occur for the mid-ranked airports. The deduction barely affects the airports on the top or bottom ranges.  Figure 4.10 Impact of Adjusting Snow Cost on Gross VFP  2.0  DConventiona1VFP  Snow Cost Adjusted VFP  1.8 1.6  I_.  1.4 1.2 1.0 0.8 0.6  I.  0.4 0.2 0.0  Table 4.7 Significant Impact of Adjusting Snow Removal Cost  Conventional  Snow Cost  VFP  Adjusted VFP  0.949 1.381 0.938 1.057 0.617  1.049 1.517 1.018 1.147 0.651  Pre-Ranking Post-Ranking YOW YYC YEG YWG BWI  35 7 37 25 59  29 6 30 19 58  51  Table 4.8 Impact of Adjusting Snow Removal Cost on Top/Bottom Airports  Top 10% Airports  Bottom 10% Airports  Before Adjustment CLT 1.869 ATL 1.848 MSP 1.680 RDU 1.679 YVR 1.539 RIC 1.44  After Adjustment CLT 1.871 ATL 1.848 MSP 1.712 RDU 1.679 YVR 1.574 YYC 1.517  ALB SJC BWI MSY JFK MIA  ALB BWI SJC JFK MSY MIA  0.649 0.641 0.617 0.554 0.554 0.519  0.658 0.651 0.641 0.574 0.554 0.519  4.3.3 Combined Effect When we utilize regional price indices and deduct snow removal cost together, the gross VFPs do not vary significantly after the adjustment. A summary of VFP scores is described in Table 4.9. In general, the improved VFPs have a slightly larger mean value and variance, while on average the two sets of VFPs are highly correlated.  Figure 4.11 Impact of Combined Improvements on Gross VFP 2.0 I 1.8  D Conventional VFP Scores A11  1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0  52  Table 4.9 Summary of VFP Scores VFP 1 (Conventional)  VFP2 (Improved)  Most Efficient Airport  CLT1 .887  ATL= 1.828  Least Efficient Airport  MIA0.523  M1A0.5 15  VFP Mean  1.024  1.029  VFP Variance  0.091  0.088  Correlation  0.98  In order to statistically examine the differences between the two sets of VFPs, we define -VFP j=],2,...n 1 D=VFP , 21  (4.2)  where D, reflects only the differential effects of the combined improvements, and n=62 in our case.  Assume D 1 is an i.i.d. random variable draw from N(u, a 2) (The normality of D 3 is tested in Figure 4.12). Then  5D D= 1  is  D  =  —  has a t-distribution with n-i degree of freedom, where  Sd / the  mean  value  of  , 1 D  and  the  standard  fl j=1  2 1 —D) deviationSd =J(D  53  Figure 4.12 Normal Distribution Test of VFP Differences  c’J  0  -.1  We then test the hypothesis as  0  .1 Inverse Normal  .2  .3  . ,uO 0 H Hj:,uO  In our case, t-value equals to 0.937, which fails to reject the null hypothesis at w=0.05 . This implies that the differences between the two sets of VFP scores are not 2 level’ significant at means.  Nevertheless, the combined adjustments affect a number of airports with significant changes in efficiency scores and ranking. As listed in Table 4.10, the two airports in New York (LGA and JFK) and three Canadian Airports (YOW, YEG, and YWG) significantly raised their VFP scores and rankings after the adjustment. On the other hand, the combined effects also brought down certain airports. Phoenix-PHX falls from 20 to 27, while Houston-IAH and Palm Bench-PBI decline by 6 and 5 places, respectively.  12  The critical value of one tail t-distribution at 95% confidence interval equals to 1.67. 54  Table 4.10 Significant Impact of Combined Improvements Conventional  VFP with Combined  VFP  Improvements  1.025 0.956 0.945 1.065 0.554 1.092 0.983 0.966 1.163 1.092  1.177 1.062 1.030 1.141 0.666 1.081 0.970 0.95 1 1.092 1.065  Pre-Ranking Post-Ranking LGA YOW YEG YWG JFK JAX FLL PBI JAR PHX  28 35 37 25 61 21 30 32 16 20  15 28 30 19 56 25 34 37 22 27  Moreover, for the top and bottom 10% airports, the impact of the combined adjustments is quite significant. Except for YVR, the six top ranked airports have entirely changed their places. The differences are also significant for bottom six airports: besides the changes in sequence, ONT takes place of JFK and becomes one of the bottom ranked airports.  In conclusion, when applying the three improvements together, the two sets of VFPs are highly correlated to each other. However, there are still non-negligible differences in both VFP scores and airport rankings even for the top and bottom 10% airports.  Table 4.11 Impacts of Combined Improvements on Top/Bottom Airports Before Adjustment  Top 10% Airports  CLT ATE RDU MSP YVR RIC  1.887 1.862 1.694 1.692 1.550 1.455  After Adjustment ATL CLT MSP RDU YVR YYC  1.828 1.821 1.704 1.679 1.574 1.476 55  Before Adjustment ALB SJC BWI MSY JFK MIA  Bottom 10% Airports  After Adjustment  0.655 0.646 0.621 0.559 0.554 0.523  ONT ALB SJC BWI MSY MIA  0.661 0.655 0.648 0.643 0.525 0.515  4.4 Conclusion Airports utilize a number of resources to produce various services for both passenger and airlines. Besides aviation charges, commercial activities have become another essential revenue source for most airports. In order to define a uniformed set of airport activities, we distinguish three outputs as passenger, ATM, and non-aeronautical revenue; and two inputs as labor and soft cost input. Soft cost input includes purchased service and purchased goods and material.  For the purpose of applying all three methodologies of efficiency measurement, quantity, price, and revenue/cost share are identified for each output/input. We further explore several data improvements in order to increase the accuracy of measuring airport efficiency.  Specifically,  regional  price indices  have  replaced PPP to  deflate  non-aeronautical revenue and soft cost input, and snow removal cost has been deducted from soft cost input.  Furthermore, gross VFPs are developed to examine the effects of these data improvements on measured efficiency scores. We find that airports in less expensive cities increased VFPs after adjusting non-aeronautical revenue by regional price indices; while airports in expensive cities advanced VFP scores after adjusting soft cost input. When applying regional price indices on both soft cost input and non-aeronautical revenue, the differences become less significant. Moreover, deducting snow removal cost  56  help raise the VFP scores of airports in cold climate. In general, these data improvements help address a number of issues in conventional data construction, and improve the accuracy of airport efficiency measurement. Their impacts are more significant for mid-ranked airports than airports on the top and bottom ranges.  57  5. ESTIMATION RESULTS Having elaborated methodologies and constructed data, in this chapter we measure airport efficiencies by VFP, DEA, and SFA-production function, respectively. Based on an identical airport sample, efficiency scores and airport rankings are estimated and compared across the three methods. Further, gross efficiency measurements are affected by a number of airport characteristics, thus may not reflect airports’ managerial efficiencies. We therefore estimate and compare airport managerial efficiencies and rankings as well.  5.1 Gross Efficiency Results between Methodologies This  section estimates  gross  efficiency measurements using VFP,  DEA and  SFA-production function. By comparing gross efficiencies scores 13 and airport rankings across the three methods, we examine the impact of methodology on gross efficiency measurements. In this section, SFA production function is estimated only with the input quantity indices (labor input and soft cost input) while using the output quantity index (aggregated using the index number approach described in Chapter 4) as the single dependent variable.  All three methods are applied with the improved dataset generated in Chapter 4. As gross efficiency scores of DEA and SFA are both bounded between zero and one, we normalize VFP scores in order to constrain them in the same range. Figure 5.1 compares the gross efficiency scores of the three methods. As can be seen, there are differences across the three sets of efficiency scores. VFP and DEA methods generate the highest scores as 1, while there is no airport with 100% gross efficiency score in SFA result. In general, the three sets of efficiency scores are highly correlated with each other, as implied in Table 13  The gross efficiency scores cannot be interpreted as how efficient airports are, because they are affected by a number of airport characteristics which are beyond manager’s control. We will measure airport efficiencies in the later section by conducting second stage regressions. 58  5.1 that their correlation coefficients are significantly high.  Figure 5.1 Gross Efficiency Measurements of VFP, DEA and SFA {iVFP Score  1.0  a DEA  0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0  Table 5.1 Pearson Correlation Matrix of Gross Efficiency Scores VFP  *  DEA  SFA  -  -  VFP  1.000  DEA  0.811  1.000  SFA  0.843  0.806  -  1.000  All correlations are significant at the 0.01 level (2-tailed).  In addition, we compare airport rankings across these three methods based on their gross efficiency scores. Table 5.2 reports the relative gross rankings according to each method, together with the mean ranking, mean efficiency, and standard deviations. As shown, there are certain airports whose gross rankings are consistent between methodologies, for example, ATL, CLT, RDU, STL, MIA, and MSY. We find that these robust airports are on the top and bottom rages of gross efficiency scores. Meanwhile, rankings of some other  59  airports, especially those mid-ranked airports, are more sensitive to methodologies. For instance, SAT and RNO are ranked between 20 and 30 places in gross VFP and SFA, while they are surprisingly with 100% gross efficiency scores airports in DEA result. The reason for this considerable difference might be explained by the weak distinguishing ability of DEA method. On average, the rankings between VFP and SFA are more consistent and correlated with each other, as showed in Table 5.3.  Table 5.2 Comparative Gross Rankings between Methodologies Airport  VFP  DEA  SFA  Mean Ranking  St. Dev.  Mean Efficiency  St. Dev.  ATE  1  1  1  1.0  0.0  0.978  0.039  CLT  2  1  3  2.0  1.0  0.974  0.041  MSP  3  1  2  2.0  1.0  0.953  0.041  RDU  4  1  4  3.0  1.7  0.945  0.047  YVR  5  1  5  3.7  2.3  0.926  0.070  YYC  6  1  6  4.3  2.9  0.904  0.096  RIC  7  1  9  5.7  4.2  0.897  0.102  ABQ  9  1  13  7.7  6.1  0.863  0.149  LGA  15  1  11  9.0  7.2  0.845  0.182  TPA  10  18  8  12.0  5.3  0.807  0.106  SDF  8  14  17  13.0  4.6  0.848  0.106  MCO  20  13  10  14.3  5.1  0.818  0.182  RNO  23  1  24  16.0  13.0  0.823  0.206  LAS  16  26  7  16.3  9.5  0.759  0.133  MEM  12  22  21  18.3  5.5  0.770  0.110  MKE  11  19  25  18.3  7.0  0.781  0.104  SLC  17  23  16  18.7  3.8  0.764  0.126  BNA  14  21  22  19.0  4.4  0.767  0.113  SAT  31  1  26  19.3  16.1  0.804  0.236  EWR  39  1  19  19.7  19.0  0.796  0.257  CVG  13  29  18  20.0  8.2  0.752  0.117  SNA  21  15  28  21.3  6.5  0.798  0.172  YWG  19  16  35  23.3  10.2  0.794  0.149  DEN  29  27  15  23.7  7.6  0.729  0.156  PHX  27  31  14  24.0  8.9  0.728  0.152  PDX  18  32  23  24.3  7.1  0.738  0.125  IAH  22  40  12  24.7  14.2  0.698  0.163 60  Afrport  VFP  DEA  SFA  Mean Ranking  St. Dev.  Mean Efficiency  St. Dev.  IND  26  28  27  27.0  1.0  0.728  0.139  SEA  24  38  20  27.3  9.5  0.698  0.158  JAX  25  25  34  28.0  5.2  0.733  0.132  FLL  34  24  31  29.7  5.1  0.717  0.170  lAD  32  33  32  32.3  0.6  0.704  0.163  YUL  33  34  33  33.3  0.6  0.696  0.160  PBI  37  20  44  33.7  12.3  0.717  0.171  YEG  30  30  43  34.3  7.5  0.704  0.134  DTW  35  42  30  35.7  6.0  0.665  0.177  YOW  28  41  45  38.0  8.9  0.673  0.135  BOS  42  36  38  38.7  3.1  0.653  0.186  DCA  36  45  36  39.0  5.2  0.639  0.184  SAN  40  37  40  39.0  1.7  0.655  0.171  JFK  56  17  49  40.7  20.8  0.675  0.269  ORD  43  52  29  41.3  11.6  0.615  0.217  1-INL  38  48  39  41.7  5.5  0.628  0.186  DFW  49  44  37  43.3  6.0  0.622  0.208  OAK  46  35  50  43.7  7.8  0.639  0.187  MDW  48  39  48  45.0  5.2  0.628  0.188  CLE  41  49  47  45.7  4.2  0.612  0.186  SF0  51  47  41  46.3  5.0  0.599  0.211  MCI  47  43  51  47.0  4.0  0.614  0.183  YHZ  44  46  56  48.7  6.4  0.592  0.163  LAX  55  50  42  49.0  6.6  0.579  0.230  AUS  45  51  52  49.3  3.8  0.589  0.185  PHL  50  60  46  52.0  7.2  0.561  0.231  PIT  53  54  54  53.7  0.6  0.545  0.204  STL  52  56  53  53.7  2.1  0.541  0.217  SMF  54  53  55  54.0  1.0  0.548  0.203  ONT  57  58  58  57.7  0.6  0.509  0.209  ALB  58  55  61  58.0  3.0  0.506  0.188  SJC  59  57  59  58.3  1.2  0.508  0.208  BWI  60  59  57  58.7  1.5  0.510  0.223  MIA  62  61  60  61.0  1.0  0.446  0.244  MSY  61  62  62  61.7  0.6  0.427  0.208  61  Table 5.3 Pearson Correlation Matrix of Gross Rankings VFP  *  DEA  SFA  -  -  VFP  1.000  DEA  0.826  1.000  SFA  0.912  0.811  -  1.000  All correlations are significant at the 0.01 level (2-tailed).  Furthermore, in order to examine whether certain groups of airports are more sensitive to benchmarking methodology, we divide our sample into three groups based on their gross efficiency scales, and examine ranking differences for top 15, bottom 15, and the mid-ranked airports, respectively (see Figure 5.2a, Figure 5.2b, and Figure 5.2c). The result implies that for top 15 airports, rankings between VFP and SFA are largely consistent, while 10 out of these 15 airports are with 100% gross efficiency scores in DEA result. On the other hand, the rankings of bottom 15 airports are also similar across the three methods, especially the bottom nine. Finally, for the mid-ranked airports, there are significant differences across the three sets of airport rankings.  Figure 5.2a Gross Ranking Comparison of Top 15 Airports 70 60 50 40 30 20 10 0  -  ._1—  -  1-  F  --• .  c)  -—  (_)  -  cY<  <  ‘-  00  62  Figure 5.2b Gross Ranking Comparison of Bottom 15 Airports 70  ZZE  60 50 C  40  —4E-—SFA Ranking  t 0  :i:  i  20 10 0 cIDCIDQ<C)  >-_<  C)  Figure 5.2c Gross Ranking Comparison of Mid-Ranked Airports  <C)  Z  ><<  <X  -‘  <  Z  -  In summary, the gross efficiency scores and airport rankings between VFP, DEA, and SFA-production function are highly correlated. The three set of rankings are quite similar to each others for the top and bottom ranked airports, while there are significant differences in the medium set, i.e. the mid-ranked airports are more sensitive to the benchmarking method.  63  5.2 The Impact of Airport Characteristics on Gross Efficiency Result  The gross efficiency scores derived in the previous section are affected by a number of airport characteristics, for example airport output size, capacity constraint, level of commercial services, etc. As some of these factors are beyond airport manager’s control, the gross measured efficiency scores are not necessarily good estimators for airport managerial performances. Therefore, in this section, we use regression analysis to decompose gross efficiency scores and examine the impacts of these airport characteristics on measured efficiency. The results are further utilized to estimate airport managerial efficiencies in the next section.  For gross VFP and DEA scores, we use regression analysis to identify the potential effects of airport characteristics. The Log-Linear OLS model is used to decompose gross VFPs. Further, as gross DEA efficiency scores have an upper bound of one, there might be a truncated bias if we use OLS model. Thus, as done in many previous studies utilizing the second stage regression analysis we apply the Tobit regression model (Tobin, 1958) to DEA scores. The Tobit model is defined as: 0* =fl X 1 0 +/3 0=0  +8  f&*>0  0=0* f6*<0  (5.1)  where 0 is the efficiency scores; Xi is a set of airport characteristics which have potential influence on airport efficiency. It is noticeable that, as all the variables are in logarithm forms, the upper bound becomes zero. In addition, as discussed in Chapter 3, SFA-production function is a parametric frontier method. Instead of decomposing efficiency scores by regression analysis, we incorporate airport characteristics as independent variables to the original production function to directly estimate managerial efficiency. On the basis of various previous studies including the ATRS Global Airport 64  Performance benchmarking studies in the past seven years (2002-2008) and our own analysis led us to incorporate the following variables in the VFP and DEA regressions as these may influence the gross efficiency measures reported in previous section •  Congestion Delay: Many of our sample airports suffer from runway and terminal congestions. The congestion may affect airport productivity as it causes delay for passengers and airlines. In this study, we use the percentage of non-weather delay as an indicator for congestion delay.  •  Airport Output Size: Airports handling more passengers are expected to achieve higher efficiencies, because the continuous flow of passengers keeps airport employees and facilities busy and more productive.  •  Average Aircraft Size: Large aircrafts carry more passengers and cargo at one time, which require a larger number of operators and other facilities to provide land services. Thus, airports have to provide sufficient landside capacity for “peak” hours; however it leads to a lower utilization and productivity in “off-peak” hours. On the other hand, airports that mostly handled large aircrafts tend to have higher utilization of airs ide facilities.  •  Percentage of International Traffic: International traffic requires more airport services than domestic traffic. On the other hand, airports collect more revenues from international passengers. As a result, the impact of international traffic on airport efficiency depends on the counter-balance of these two factors.  •  Percentage of Air Cargo: Since our productivity measurements do not include cargo as a separate output, the efficiencies of airports with a higher percentage of air cargo in total traffic may be underestimated.  •  Percentage of Non-Aeronautical Revenue: This indicator is used to present the business strategy of airport. Commercial activities expand airport revenue; however they also require additional inputs. Therefore, it is necessary to examine the impact of non-aeronautical activities on airport efficiency.  65  •  Percentage of Connecting Passenger: Hub airports usually have a significant amount of connecting passengers. Connecting passengers requires less service than direct passengers. Therefore, airports with high proportion of connecting passengers are expected to have a high productivity.  •  Percentage of Hub Carrier Market Share: The dominance by a hub carrier at an airport may allow better coordination and cooperation between the carrier and the airport. Therefore, airports that are dominated by a hub carrier are expected to have higher efficiencies than airports with a large number of competing airlines.  The regression results are listed in Table 5.4, we try to include all our sample airports in the regression. However, the information on non-weather delay is not available for Canadian airports. Therefore, the results only illustrate the situation of U.S. airports.  The VFP regression shows that airport efficiency does not have a statistically significant relationship with airport congestion delay, percentage of cargo service, or hub carrier’s market share. Among the other factors, airport efficiency is negatively correlated to percentage of international traffic, and the coefficient is statistically significant. It is expected that increasing the percentage of international traffic by 10% reduces airport efficiency by 0.2%. Furthermore, the coefficient of average aircraft size is also negative and statistically significant, indicating that airports in North America tend to have landside congestions. At the same time, the efficiency is positively correlated with level of non-aeronautical revenue, connecting traffic, and output size, and the coefficients are all statistically significant. A 10% increase in percentage of non-aeronautical revenue is expected to increase the airport’s efficiency by 5.7%. While a 10% increase in passenger volume is expected to raise the efficiency by 2.3%. In addition, the regression result of DEA is largely consistent with that of VFP. In order to make our SFA efficiency measures comparable to the ‘residual VFP and residual DEA efficiency indices”, this time the SFA  66  production function was estimated by adding these airport characteristic variables in addition to the input quantity variables.  Table 5.4 Regression Results on Gross Efficiency Scores VFP OLS (log-log)  DEA Tobit (log-log)  Coefficient  t-Stat  Coefficient  t-Stat  Congestion Delay  0.012  0.07  0.128  0.50  Output Size  0.23 1  3.46*  0.207  2.26*  Ave. Aircraft Size  -0.390  3.23*  -0.197  -1.18  % International  -0.021  1.71*  -0.037  2.O8*  % Cargo  -0.037  -1.19  -0.010  -0.23  % Non-Aeronautical Revenue  0.572  4.22*  0.615  3.23 *  % Connecting Passenger  0.027  2.01*  0.033  1.72*  % Hub Carrier  0.003  0.05  -0.026  -0.29  Intercept  1.283  2.07  0.972  1.12  2 R  0.55  Log-likelihood value Observations *  -  55  -19.15 55  The coefficient is significant at the 95% level.  **The coefficient is significant at the 90% level.  5.3 Managerial Efficiency Results between Methodologies By removing the impacts of airport characteristics that are beyond manager’s control, we estimate residual (managerial) efficiencies of the three methodologies, respectively. As implied in Table 5.5, in general airport managerial efficiencies across the three methods are highly correlated to each other.  67  Table 5.5 Pearson Correlation Matrix of Residual Efficiencies VFP  *  DEA  SFA  -  -  VFP  1.000  DEA  0.83 1  1.000  SFA  0.913  0.871  -  1.000  All correlations are significant at the 0.01 level (2-tailed).  Furthermore, we rank our sample airports according to their managerial performances. Table 5.6 summaries the residual rankings between the three alternative methods, together with their mean ranking and standard deviation. We find that the rankings for airports like ATL, RDU, MIA, and PIT are consistent across all methods. Among them, ATL and RDU are the among the most efficient airports, while MIA and PIT is two most inefficient airports. On the other hand, a number of mid-ranked airports vary significantly in their rankings across the three methods, for example SEA, EWR, and JFK. In general, the three sets of airport rankings are highly correlated to each other. Again, the ranking results between VFP and SFA are more consistent then between DEA (see Table 5.7).  Table 5.6 Comparative Residual Rankings between Methodologies Airport ATE RDU RNO CLT PBI BNA MSP JAX LGA SAT TPA SNA  VFP 3 2 9 1 7 5 4 6 11 12 8 10  DEA 1 5 2 7 3 12 9 13 4 10 14 16  SFA 3 4 1 6 5 2 7 8 13 9 10 11  Mean Ranking 2.3 3.7 4.0 4.7 5.0 6.3 6.7 9.0 9.3 10.3 10.7 12.3  St. Dev. 1.2 1.5 4.4 3.2 2.0 5.1 2.5 3.6 4.7 1.5 3.1 3.2 68  Airport MCO MKE FLL PDX SAN SEC OAK RIC ABQ SEA HNL EWR lAD LAS MEM MDW DCA IAH JFK END PHX SMF AUS SDF DEN DTW SF0 MCI BOS CVG CLE SJC ALB PHL DFW STL ONT LAX  VFP 13 15 16 14 19 22 25 18 23 17 20 38 27 21 30 32 26 24 45 33 29 31 28 34 37 36 40 39 41 35 43 42 44 46 53 49 48 54  DEA 8 17 15 21 24 20 19 25 18 35 27 6 22 30 23 28 38 37 11 29 33 26 40 31 34 41 39 36 32 43 44 45 42 54 46 52 48 47  SFA 18 12 15 14 16 20 19 21 23 17 22 28 26 25 29 27 24 30 36 31 33 38 32 37 39 35 34 41 43 40 42 45 50 44 47 46 51 48  Mean Ranking 13.0 14.7 15.3 16.3 19.7 20.7 21.0 21.3 21.3 23.0 23.0 24.0 25.0 25.3 27.3 29.0 29.3 30.3 30.7 31.0 31.7 31.7 33.3 34.0 36.7 37.3 37.7 38.7 38.7 39.3 43.0 44.0 45.3 48.0 48.7 49.0 49.0 49.7  St. Dev. 5.0 2.5 0.6 4.0 4.0 1.2 3.5 3.5 2.9 10.4 3.6 16.4 2.6 4.5 3.8 2.6 7.6 6.5 17.6 2.0 2.3 6.0 6.1 3.0 2.5 3.2 3.2 2.5 5.9 4.0 1.0 1.7 4.2 5.3 3.8 3.0 1.7 3.8  69  Airport ORD BWI PIT MSY MIA  VFP 50 51 52 47 55  DEA 51 49 50 53 55  SFA 49 53 52 54 55  Mean Ranking 50.0 51.0 51.3 51.3 55.0  St. Dev. 1.0 2.0 1.2 3.8 0.0  Table 5.7 Pearson Correlation Matrix of Residual Rankings VFP  *  DEA  SFA  -  -  VFP  1.000  DEA  0.847  1.000  SFA  0.924  0.890  -  1.000  All correlations are significant at the 0.01 level (2-tailed).  Moreover, we divide our sample airports, and examine the impact of methodology on certain groups of airports. For the top 15 airports, except for BNA, the rankings are largely consistent across the three methods. Most airports in this group have similar efficiency rankings no matter what methodology we use. In addition, the rankings for the bottom 15 airports are also similar across the three methods. Except for BOS and PHL, there are only minor changes between each method. Finally, there are significant variations in the rankings of medium ranged airports. Substantial differences exist in this group, however the three sets of rankings still have similar trends.  70  Figure 5.3a Residual Ranking Comparison of Top 15 Airports 55  Residual VFP Ranking Residual DEA Ranking -*-ResidualSFARankuig —.--  -  -  50  —  45  40 .  35  30 25 < 20  15 10  5 0  z  x  <(-<<0 -  Cd,  E-  z  1-I.  -  ,  Figure 5.3b Residual Ranking Comparison of Bottom 15 Airports 55 50 45 40 35 30  C%klLkLI\H’l\t1lLI%  25  Residuni SFA Ranking  Residuil I )I_A Rankitii  20 15  10 5 0  71  Figure 5.3c Residual Ranking Comparison of Mid-Ranked Airports —.—  55  —.—  :  5°  ::..  Residual VFP Ranking Residual DEA Ranking  :  3° o 25  20 15 10 5 0 ><  Z  L)  /)C)Q  CJ  cy<  -  E5  -  C—  “  -  Z  C  ‘_c,<Cl)C  5.4 Conclusion This chapter estimates gross and managerial efficiency scores for our sample airports by VFP, DEA and SFA-production function, respectively. We find that the three sets of gross efficiency scores and airport rankings are highly correlated with each others. The top and bottom ranked airports have similar rankings across the three methods, while significant differences exist in the medium set. In addition, by removing the effects of airport characteristics that are beyond manager’s control, we estimate managerial efficiencies  and rankings between the three methods. Similarly, the three sets of efficiencies and airport rankings are highly positively correlated with each others. Rankings of airports on the top and bottom ends are similar across the three methods, while there are considerable differences between the rakings in the medium set of airports.  72  6. CONCLUSION In this chapter, key findings of this thesis are summarized, and some suggestions for further research are described.  6.1 Summary of Key Findings This study reviews and compares airport efficiency indices measured by Index Number Method, Data Envelopment Analysis (DEA), and SFA (production function approach). Based on a sample of 62 North American airports, gross and managerial efficiency scores and airport rankings are estimated by each method, respectively. A series of data improvements have been explored and their effects on measured productivities are examined as well. Furthermore, some airport characteristics are identified to •have potential impacts on overall efficiency scores. The findings of this study may be summarized into the following three aspects.  The impact of data improvements Aims to improve the accuracy of measuring airport efficiencies, conventional data construction has been improved in several aspects: using city-based Cost of Living Index as the price index to deflate non-aeronautical revenue; using Cost of Living Index to deflate soft cost input; and deducting snow removal cost for airports in cold climate. We find that these data improvements have significant impact on efficiency result, and the influences are more substantial for middle ranked airports than for airports on the top and bottom ends. When applying these improvements together, the differences in efficiencies and rankings become less significant. It is possible that the effects of adjusting non-aeronautical output and adjusting soft cost input canceled out some impacts on overall productivities. However, for a number of airports, especially those in the medium set, the changes are still considerable. In general, there is significant evidence that airport efficiencies are directly affected by data construction. Obtaining valid data is one of the 73  most important issues in airport benchmarking.  The impact of airport characteristics A number of airport characteristics have potential effects on measured efficiencies. We find that non-aeronautical revenue, output size, average aircraft size, level of international passenger, and level of connecting passenger are significantly related to airport efficiency. A 10% increase in percentage of non-aeronautical revenue is expected to increase the airport’s efficiency by 5.7%; while a 10% increase in output size is expected to increase efficiency by 2.3%.  The efficiency comparison between methods In the case of gross measurements, VFP, DEA, and SFA-production function produce highly correlated efficiency scores and airport rankings. There are minor changes for the top and bottom ranked airports, while the benchmarking results of the mid-ranked airports are different across the three methods.  For the managerial efficiency scores, we find that the three methods again generate positively correlated results. Airport rankings and productivities for the top and bottom ranges of efficiency scale are quite similar across the three methods, while significant variations exist in the mid-ranked airports.  6.2 Suggestions for Further Research Without doubt, there have been considerable developments within the area of airport benchmarking in recent years, and the sector no longer lags so much behind other industries, including airlines, in the knowledge and practical use of efficiency measurement. The findings of the current study provide impetus for future researchers in the following aspects. First, capital input is an essential part of airport cost. If possible,  74  future studies should consider obtaining reliable data, and incorporate capital input into airport benchmarking study. Second, the use of SFA in a production function brings in additional assumptions, which might influence the outcome of SFA measurement. It is suggested that further studies construct SFA in cost function form to increase the credibility of measured efficiency. Finally, based on the current study, including additional airports and constructing time series comparison would be another area that warrants further research. Because of the extreme difficulty and costliness in terms of research time and budget in obtaining reliable data, researchers tend to use short-cut data collection or some data approximations for conducting studies on airport efficiency measurement and/or benchmarking. In the light of the findings in this thesis, future researchers should be cautious when interpreting their research results based on such data.  75  BIBLIOGRAPHY Abbott, M., Wu, S., (1992), “Total Factor Productivity and Efficiency of Australia Airports,” The Australian Economic Review, Vol. 33, no. 3, pp. 244-260. Adler, N., Friedman, L., and Sinuany-Stern, Z., (2002), “Review of ranking methods in the data envelopment analysis context,” European Journal of Operational Research, 140, pp. 249-265. Aigner, D.J., Amemiya, T., and Poirier, D.J. (1976), “On estimation of production frontiers: Maximum likelihood estimation of the parameters of a discontinuous density function,” International Economics Review, 17, pp. 377-396. Aigner, D.J., Lovell, C.A.K., and Schmidt, P.J. (1977), “Formulation and estimation of stochastic frontier production function models,” Journal of Econometrics, 6, pp. 2 1-37. Airbus (2007), “Global Market Forecast, 2007-2026,” Airbus, 2007. ATRS (Air Transport Research Society) (2008), The ATRS Airport Performance Benchmarking Report: Global Standards for Airport Excellence, Vancouver, British Columbia, Canada. Barros, C.P. (2008), “Technical efficiency of UK airports,” Journal of Air Transport Management, 14.,pp. 175-178. Battese, G.E. and Coelli, T.J. (1992), “Frontier Production Functions, Technical Eefficiency and Panel Data, with Application to Paddy Farmers in India,” The Journal ofProductivity Analysis, 3, pp. 153-169. Beaudoin, Justin (2006), “An Empirical Investigation of the Effects of Managerial and Ownership Structure on the Efficiency of North America’s Airports,” Master of Science Thesis, University of British Columbia, Vancouver, Canada. Boeing (2008) “Current Market Outlook 2008,” Boeing Commercial Aircraft Company, 2008. Caves, D.W., Christensen, L.R., and Diewert, W.E. (1982a), “Multilateral Comparisons 76  of Output, Input, and Productivity Using Index Numbers,” Economic Journal, 92, 73 -86. pp. Charnes, A., Cooper, W.W., and Rhodes, E. (1978) “Measuring the efficiency of decision making units,” European Journal of Operation Research, 2, pp. 426-444. Christensen, L.R. and Jorgenson, D.W. (1969), “The measurement of U.S. real capital input, 1929-1967.”, The Review ofIncome and Wealth, 15(1), pp. 293-320. Coelli, T. and Perelman, S. (1999). “A Comparison of Parametric and Non-Parametric  Distance  Functions:  With  Application  to  European  Railways,” European Journal of Operational Research, 117, pp. 326-39. Coelli, T., Rao, D.S., O’Donnell, C.J., and Battese, G.E. (2005). An Introduction to Efficiency and Productivity Analysis, 2nd Edition. Boston: Kiuwer Academic Publishers. Craig, S., Airola, J., and Tipu, M. (2005), “The effect of institutional fonu on airport governance efficiency,” Unpublished manuscript, Department of Economics, University of Houston, Houston, Texas, U.S. Diewert, W.E. (1992), “The Measurement of Productivity,” Bulletin of Economic Research, 44(3), pp. 163-198 Doganis, R. (1992), The Airport Business, Routledge: London, UK and New York, NY. Doganis, R.S. and Thompson, G.F. (1973), The Economics of British Airports, Transport Studies Group. Doganis, R.S. and Thompson, G.F. (1974), “Establishing Airport Cost and Revenue Functions,” The Aeronautical Journal, Vol. 78, pp. 285-3 04. Forsyth, P. (2000). “Models of Airport Performance,” Handbook of Transport Modeling, D.A. Hensher, K.J. Button, eds. Fried, H.O., Love!!, C.A.K., Schmidt, S.S. and Yaisawamg, S. (2002), “Accounting for Environmental Effects and Statistical Noise in Data Envelopment Analysis,” Journal ofProductivity Analysis, 17:1/2 (January), pp. 157-74.  77  Gillen, D. and La!!, A. (1997), “Developing measures of airport productivity and performance: an application of data envelopment analysis,” Transport Research Part E, Vol. 33, No.4, pp. 26 1-273. Hjalmarsson, L., Kumbhakar, S., Heshmati, A. (1996), “DEA, DFA and SFA: A Comparison,” The Journal ofProductivity Analysis, 7, pp. 303-327. Hooper, P.G. and Hensher, D.A. (1997), “Measuring total factor productivity of airports-an index number approach”, Transportation Research E, Vol. 33, No.4, pp. 249-259. ICAO (International Civil Aviation Organization) (2007), “Outlook for Air Transport to the Year 2025,” International Civil Aviation Organization, Montreal, Canada. Jeong, Jinha (2005), “An investigation of operating costs of airports: focus on the effects of output scale,” Master of Science Thesis, University of British Columbia, Vancouver, Canada. Kincaid, I. and Tretheway, M. (2006), “Guidelines for Benchmarking Airports,” Working Paper, German Aviation Research Society. Main, B.G.M., Lever, B. and Crook, J. (2003), “Central Scotland Airport Study,” Report, The David Hume Institute. Martin-Cejas, R.R. (2002), “An approximation to the productive efficiency of the Spanish airports network through a deterministic cost frontier,” Journal of Air Transport Management, 8 (2002), pp. 233-238. Martin, J.C. and Roman, C. (2001), “An application of DEA to measure the efficiency of Spanish airports prior to privatization,” Journal of Air Transport Management, 7 (2001), pp. 149-157. Martin, J., Voltes-Dorta, A. (2007), “Stochastic frontier efficiency estimation in airports: a Bayesian approach”, A paper presented at the  th 11  ATRS World Conference,  Berkeley, USA. Nyshadham, E.A., and Rao, V.K. (2000), “Assessing efficiency of European airports: a  78  total factor productivity approach,” Public Works Management and Policy, 5, pp. 106- 144. Oum, T.H., Adler, N., and Yu, C. (2006), “Privatization, Corporatization, Ownership Forms and their Effects on the Performance of World’s Major Airports,” Journal of Air Transport Management, 12(2006), pp. 109-121. Oum, T.H., Tretheway, M.W., and Waters, W.G. (1992), “Concepts, methods and purposes of productivity measurement in transportation,” Transportation Research A, Vol. 26A, No.6, pp. 493-505. Oum, T.H., Waters, W.G. and Yu, C. (1999), “Survey of Productivity and Efficiency Measurement in Rail Transport,” Journal of Transportation Economics and Policy, vol. 33, Part I (Apr. 1999), pp. 9-42. Oum T.H., Yan, J. and Yu, C. (2008), “Ownership Forms Matter for Airport Efficiency: A Stochastic Frontier Investigation of Worldwide Airports,” Journal of Urban Economics, 64, pp. 422—435. Pels, E., Nijkamp, P. and Rietveld, P. (2001), “Relative efficiency of European airports,” Transport Policy, Vol. 8,Iss. 3, pp. 183-192. Pels, E., Nijkamp, P. and Rietveld, P. (2003), “Inefficiencies and scale economies of European airport operations,” Transportation Research Part E, 39, pp. 34 1-361. Seiford, L.M. (1994). A DEA bibliography (1978-1992). In: Charnes, A., Cooper, W., Lewin, A.Y.,  Seiford, L.M. (Eds.), Data Envelopment Analysis: Theory,  Methodology and Applications. Kiuwer Academic, Boston. Stevenson, R.E. (1980), “Likelihood functions for generalized frontier estimation,” Journal ofEconometrics, 13, pp. 57-66. Tobin, J. (1958), “Estimating the relationship for limited dependent variable,” Econometrica, 26 (1958), pp. 24-36. Transport Research Laboratory (TRL) (2004), Airport Performance Indicators, Wokingham, TRL.  79  Yoshida, Yuichiro (2004), “Endogenous-weight TFP measurement: methodology and its application to Japanese-airport benchmarking,” Transportation Research Part E, 40 (2004), pp. 151-182.  80  Appendix  Data used in this study are collected from the following sources: •  Air Transport Research Society (ATRS)  •  U.S. Federal Aviation Administration (FAA)  •  U.S. Bureau of Transportation Statistics  •  U.S. Census Bureau  •  U.S. Department of Labor  •  U.S. National Oceanic and Atmospheric Administration (NOAA)  •  Statistic Canada  •  Environment Canada  •  Airport annual reports and websites  81  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0068089/manifest

Comment

Related Items