Developing Measures of US Ports Productivity and Performance: Using Data Envelopment Analysis (DEA) and Free Disposal Hull (FDH) Approaches

Transcription

1 Developing Measures of US Ports Productivity and Performance: Using Data Envelopment Analysis (DEA) and Free Disposal Hull (FDH) Approaches By: Evangelos I. Kaisar, Ph.D.* Associate Professor Civil Environmental and Geomatics Engineering Florida Atlantic University 777 Glades Road, Bldg. 36, Rm. 214 Boca Raton, FL Phone: Fax: Submitted for Presentation and Publication July 31 st, 2013 Transportation Research Board 93 rd Annual Meeting * Corresponding Author: Evangelos I. Kaisar Word Count: 4047, Figures & Tables: 7*250, Total: 5795

2 ABSTRACT Due to its high capital investment, operating and maintenance cost, a port should be managed as efficiently as possible. Learning from the best practices ports operating under similar conditions is a very useful strategy for maintaining competiveness. Under such a competitive environment, port performance measurement is not only a powerful management tool for port operations, but also constitutes a most informative input for regional and national port planning and operation. Measuring port and container terminal productivity is an interesting issue especially if an automated system across terminals and port is required. Amongst other methods, the efficiency of container port or terminal production can potentially be analyzed by Data Envelopment Analysis (DEA) or by the Free Disposal Hull (FDH) model. Twentyfive ports from US were assessed. The analysis modeled a port as a decision making unit (DMU) utilizing three maor operating inputs, i.e. quay length, gantry cranes, and container yard to produce one maor target output, i.e. number of throughput containers. Three main results are presented. First, an efficient frontier or a set of the best practice ports is identified, where inefficient ports may wish to emulate. Second, the sources and extent of inefficiency on which inefficient ports should focus in order to improve their operations are determined. Third, port efficiency is compared among North American ports. The results can be very useful for planning and managing an airport in a competitive multiple-port network. Key words: Data Envelopment Analysis (DEA), Free Disposal Hull (FDH), Container Terminals, Ports, Efficiency, Production.

3 INTRODUCTION The globalization of the world economy has ever increased the importance of the role for transportation. In the transportation arena container transportation plays a key role in the process because of its technological advantages as compared with the traditional methods of transportation. In order to support the global trade development, US port authorities have increasingly been under pressure to improve port efficiency by ensuring that port services are provided on an internationally competitive basis. The degree of satisfaction that is obtained on the basis of preset standards will indicate the level of port performance achieved. From the foregoing, it is already obvious that port performance levels will be different depending on whether vessels, cargoes or inland transport vehicles are served. Thus, a port/container terminal may offer a very satisfactory services to customers and at the same time be udged inadequate by cargo interests or inland transport operators (or vice versa). It is obviously more likely that poor performance will not be limited to one group of port users, but rather pervade all services offered by the port. Compared with traditional port operation, containerization has improved port production. Port productivity and efficiency is an important contributor to the United State s international competitiveness. Under such a competitive environment port performance measurement is not only a good management practice for port operators, but also an important input for regional and national port planning and operations. Port efficiency has been evaluated by calculating the cargo handling relation at the berth (Tabernacle, 1995, Ashar, 1997), or by comparing actual with optimum throughput over a specific time period (Talley, 1998). Jara-Diaz et. al. (2002) estimate a multi output cost function using a flexible form from a sample of 26 Spanish seaports over an 11-year period. Outputs are containerized general cargo, break bulk general cargo, liquid bulk, dry bulk, and total rent received for leases of port space. The results support the presence of economies of scale and scope. In recent years, significant progress has been made in the measurement of efficiency in relation to production activities. In particular, non-parametric frontier methods such as Data Envelopment Analysis (DEA) and Free Disposal Hull (FDH) have been developed with application across a wide range of sectors including port services. Turner et al. (2003) studied productivity of ports in North America using a panel data between 1984 and To measure the efficiency of ports, they employed the Data Envelopment Analysis (DEA) technique. The study by Kaisar et al. (2006) aimed to analyze port productivity using DEA and two main results were presented. First, an efficient frontier or a set of the best practice ports was identified, which inefficient ports may want to emulate. Second, the sources and extent of inefficiency on which an inefficient port should focus in order to improve their operations are determined. In this study, aims to provide new information on efficiency estimation by applying the two alternatives techniques of DEA and FDH to the same terminal data set was derived from North America s leading container ports. PORT PRODUCTION MEASUREMENT Performance measurements play a significant role in the development of port terminals or other forms of organizational Decision Making Units (DMU). There are several methods that are applicable for evaluating performance of ports including regression analysis (Tongzon, 1995). In recent years, Data Envelopment Analysis (DEA) and Free Disposal Hull (FDH) are two of the many available alternative techniques for estimating the efficient frontier. These two mathematical programming techniques allow the measurement of the relative distance that an individual Decision Making Units lies away from this estimated frontier. The frontier defines the relationship between inputs and outputs by depicting graphically the maximum outputs obtainable from the given inputs. Evaluation among ports was made by comparing indicator values for different given ports over time. Performance indicators suggested by UNCTA (1976), as shown in Table 1, shows productivity and effectiveness measures and can be used as a reference point.

4 Table 1. Performance Indicators Suggested by UNCTAD Financial Indicator Tonnage worked Berth Occupancy Revenue per ton of cargo Cargo handling revenue per ton of cargo Labor Expenditure Capital equipment expenditure per ton of cargo Contribution per ton of cargo Total contribution Source: UNCTAD ( 1976, pp7-8) Operational Indicators Arrival Late Waiting Time Service Time Turn-around Time Tonnage per ship Fraction of time berthed per ship per shift Tons per ship-hour in port Tons per ship-hour at berth Tons per gang hours Fraction of time gangs idle DEA is a non-parametric method of measuring the efficiency of decision making unit (DMU) with multiple outputs and inputs. DEA is an appealing technique due to at least two reasons: First, its ability to analyze several outputs simultaneously and to derive efficiency rating within a set of analyzing units, are particular suitable for measuring port efficiency. Second and perhaps most attractive, DEA does not require prior knowledge of production functions or weights associated with inputs and outputs to assess a DMU. DEA has been adopted widely in previous port efficiency studies (Turner, 2004 & Kaisar et al. 2006). In addition, DEA will provide a piece-wise linear function to represent an empirical maximum possible frontier. In Figure 1, it plots the empirical relationship between an input and an output from a set of 7 hypothetical ports, A, B, C, D, E, F, and G. Using DEA will determine that ABCD form an efficient frontier. FDH as a deterministic non-parametric method, assuming it does not have a particular functional form for the boundary and ignore measurement error. FDH uses mathematical programming techniques to envelop the data as tightly as possible, subect to certain production assumptions, which are maintained within the mathematical programming context. Using FDH for any given level of outputs remains feasible if any of the inputs is increased, whereas the latter means that with given inputs it is always possible to reduce or maximize outputs. Notice that the FDH methodology is particularly suited to detect the most obvious cases of port inefficiency as this technique is very assertive regarding the measurement of port inefficiency. To each port declared FDH inefficient, it is possible to find at least one port in the sample that presents a superior performance relative to the first (dominated) municipality (Lovell et al., (1993). Port output can be multi-dimensional depending on the obective that ports want to achieve. Roil and Hayuth (1993) have advocated the use of this approach in measurement of port efficiency and demonstrated, based on the port data, how the relative efficiency ratings of ports could be obtained.

5 DEA-CCR Output Y DEA-BCC H (λ 1 x 1 + λ 2 x 2, λ 1 y 1 + λ 2 y 2 ) C (x 2, y 2 ) D FDH B (x 1, y 1 ) B y o E (x o, y o ) F A G O x o FIGURE 1: Non-parametric Deterministic Frontiers Input X Since its introduction by Charnes et al., (1978), DEA technique has been applied in many different contexts. In transportation, there are applications in airports [Martin and Roman (2001), Bazargan and Vasigh (2003)] and ports [Roll and Hayuth (1993), Cullianane et al., (2004), Turner et al., (2004), Wang et al. 2003/2005, Jim Wu and Goh, 2010)]. The concept of DEA is developed around the basic idea that the efficiency of a DMU is determined by its ability to transform inputs into desired outputs. The concept of DEA is developed around the basic idea that the efficiency of a DMU is determined by its ability to transform inputs into desired outputs. Suppose we have n DMUs (ports), where each DMU ( 1,2,..., n) produces s output y r ( r 1,..., s) by utilizing m inputs x i ( i 1,..., m). DEA, use the following measure of performance for DMU. h s r 1 m i1 u r v x i y r i

6 where: v i ( i 1,..., m) and u r ( r 1,..., s) are weights assigned to output r and input i, for DMU, respectively. The above equation is determined for each programming problem (Charnes et al., 1978). DMU by the following mathematical * ho Maxh o u r v i s.t. h 1 1,2,..., n v, u r 0 where: h o s r1 m i1 u r v x i y ro io represents the ratio of aggregated outputs to aggregated inputs for one of the n DMUs denoted th th DMU i ( i 1,2,..., n) and y ro are respectively the i input and r output of DMU i. For this fractional programming problem with potential infinite number of optimal solutions, Charnes et al. (1978) were able to specify an equivalent linear programming problem. This requires the introduction of a scalar quantity ( ) to adust the input and output weights: 1 T T T, u, u u T X 0 In above DEA model, the Linear programming (LP) is known as the CCR ratio model, as Charnes et al. (1978) proposed. Later on there are many variations to deal with certain type of data and assumptions. One form that was used in this study is an output-oriented and variable return-to-scale (VRS) DEA model. The idea may be explained by using a graphical illustration of hypothetical single input/output ports, as shown in Figure 1. The mathematical form is shown in the following LP: max Z Subect to y x i0 n ro n 0 1 n 1 y x i r r 1,2,... s i 1,2,... m (1)

7 To determine whether a port is on the frontier, we simply solve (1). The LP problem above is a generalized formulation for multiple input and output cases. In this formulation, n, s and m represent number of samples, number of outputs and number of inputs respectively. is a scalar vector associated with each airport and has n elements. x i, yr are quantity of input i and output r of port respectively. x io, yro are quantities of input i and output r of an port o for which the productivity is to be measured. or the performance score is a number by which the current output level has to be multiplied in order to reach the frontier. If a port is on the frontier, solving this LP will result in an optimal obective function Z 0 1. In other words, it is sufficiently productive and does not need to increase output. The performance score can be used as an index to measure total productivity of an port in terms of how far it is from the efficient production frontier. The LP needs to be solved n times, each time for an individual port. This process can be automated in many off-the-shelf optimization packages including Microsoft Excel s Solver that is used in this study. One form that was used in this study was an output-oriented and variable return-to-scale (VRS). The DEA-BBC model Banker et at., 1984, on the other hand, allows for variable returns to scale and it is graphically represented by the piece-wise linear convex frontier (Figure 1). FDH model DEA-BBC, and DEA-CCR models define different production possibility wets and efficiency results. The idea may be explained by using a graphical illustration of hypothetical single input/output ports, as shown in Figure 1. An input oriented efficiency measurement problem can be written as a series of linear programming enveloping problems, with constraints differentiating between the DEA-BBC, DEA-CCR and the FDH models as shown in the following mathematical form: min Subect to: x r X O r 1,2, S Y y r 0 (DEA-CCR) e 1m (DEA-BCC) 0,1 (FDH) r This study examined operational efficiency of ports with respect to containerized cargoes across regions in North America. One of the results is an index or a performance score, which can be used for performance comparison across ports. The score can also be used for benchmarking purpose. In the subsequent study, a plan was developed to have a causal relationship between this performance score and port factors. Such relationship would be useful for port managers and policy makers to understand factor affecting productivity of ports and ultimately derive a strategy to enhance the operational efficiency. ANALYSIS Container port infrastructure productivity is a key performance measure and is influenced by industry structure, conduct and demand. For this reason, this study has two distinct obectives: measurement of the trend in infrastructure productivity during the study period, and examination of the factors that determines infrastructure productivity through DEA and FDH methodologies. Based on the study issues and literature, this study followed Wang et at., (2003) and suggested data envelopment analysis and free

8 disposal hull methodology be employed to evaluate container port infrastructure productivity during the study period. The input and output variables should reflect actual obectives and process of container terminal productivity as accurately as possible. As discussed previously, the DEA and FDH empirical analysis uses one output measures: TEUs handled (the number of twenty foot container equivalent units). Container throughput is the most appropriate and analytically tractable indicator of the effectiveness of port production. The input measures used are: berths (the number of container berths), length (the length of the berths), total area (the total area of the berths), storage (ports many times are used as a storage area because the container yards act as buffer between sea and inland transportation or transshipment), and Shipshore Cranes (Quay Cranes are very critical for port operation). In addition, front-end handlers, yard tractor, and yard chassis, are important yard equipment. Measures of these variables were inputs into the model. The measurement of terminal production, therefore, was a mean of quantifying efficiency in the utilization of these three variables. Given the characteristics of container port production, the total quay length and the terminal area were the most suitable proxies for the land factor and the quay gantry cranes is the most suitable proxy for the equipment factor input. A summary of the maor characteristics of the input and output variables is presented in Table 2. Table 2. Descriptive Statistics for Input and Output Variables Number Berths Length Total Area Storage Shipshore Crane Front-End Handlers Yard Tractors Yard Chassis Size Min ,000 1, ,609 Max 70 7,806 5,259, , ,900,000 Mean ,430 1,389,851 24, ,787 Median 7 1, ,000 13, ,526 S.D ,742 1,274,279 30, ,024,811 The DEA and FDH models can be distinguished according to whether they are input or output oriented. Marlow and Paixao (2002) argue that the development of agile ports required the implementation of a two-stage integration process, the internal and the external one. Also, they advocate that DEA should be used for port performance measurement and its suitability has been examined by Wang et al. (2003). Qualitative performance indicators are at the heart of lean ports and consequently of port networking. A port usually is able to know or to predict its container throughput for the ensuing year. This happens because a container port has a stable customer base of shipping lines. In addition, a container terminal can also attempt to predict its future throughput by styling historical data. For these reasons an input oriented model is most appropriate on the analysis of container production given the output. The sampling frame for this study consists of the most important US and Canadian container ports from the 1996 to Although 25 ports of all regions (West-Gulf-East) were included in this study, a couple of ports had to be removed for some years due to incomplete input data, therefore the sample size for the analysis comprises a total of 174 observations. The secondary data was collected from the Containerization International Yearbook and the US Department of Transportation Maritime Administration. Table 3 lists all ports along with their TEU outputs during 1996 to Among them, Los Angeles (LA) Port served the most TEUs whereas Saint John (St. John) Port served the least with only 49,000 TEUs in Because the data source often reports container throughput data for the entire port, this study investigates individual ports not individual terminals.

9 Table 3. List of Ports with the Annual TEU* Port Halifax Montreal St. John Vancouver Baltimore Boston Charleston Galveston Gulfport Hampton Rds Houston Jacksonville Long Beach Los Angeles Miami Mobile New Orleans NY/NJ Oakland Philadelphia Port Everg Portland San Francisco Savannah Seattle Tacoma Wil. DE Wil. NC * TEU is the abbreviation for Twenty foot Equivalent Unit, referring to the most common standard size of 20 ft. in length. RESULTS AND ANALYSIS Given 29 seaports studied over a six year period, the maximum sample size would have been 174 observations. However, data availability affected the actual sample size. For the model, any variations or missing data within an observation resulted in the exclusion of that observation from the sample. For example, in the analysis, Freeport (TX), was not included for the years 1996 to 2001 as they either had no OSG cranes or lacked reliable information during such period. For purposes of this study then, they were not considered as a container port during this period.

10 Table 4 shows performance scores from solving DEA-CCR, DEA-BCC and FDH models respectively for 1996 to The average values of , and for the 1996, , and for the year 1997, and , 07948, and for the Table 4. Port Efficiency of CCR-BCC and FDH Models for 1996 to 1998 Port CCR-I BCC-I FHD-I CCR-I BCC-I FHD-I CCR-I BCC-I FHD-I Halifax Montreal St. John Vancouver Baltimore Boston Charleston Galveston Gulfport Ham. Roads Houston Jacksonville Long Beach Los Angeles Miami Mobile New Orleans NY/NJ Oakland Philly Port Everg Portland San Franci Savannah Seattle Tacoma Wil. DE Wil. NC In Table 4 the index value of represents a perfect efficiency score. For 1996, 11 out of 29 ports identified to be efficient when the DEA-CCR input oriented model was applied. Using DEA-BCC

11 input-oriented model, 16 out of the 28 ports are efficient compared with the 24 efficient when the FDH model is applied. Table 5 shows performance scores from solving DEA-CCR, DEA-BCC and FDH models respectively for 1999 to The average values of , and for 1999, , and for the year 2000, and , 08595, and for the year Table 5. Port Efficiency of CCR-BCC and FDH Models for 1999 to 2001 Port CCR-I BCC-I FHD-I CCR-I BCC-I FHD-I CCR-I BCC-I FHD-I Halifax Montreal St. John Vancouver Baltimore Boston Charleston Galveston Gulfport Ham. Roads Houston Jacksonville Long Beach Los Angeles Miami Mobile New Orleans NY/NJ Oakland Philly Port Everg Portland San Franci Savannah Seattle Tacoma Wil. DE Wil. NC Also, in Table 5 the index value of represents a perfect efficiency score. In 2001, 13 out of 29 ports identified to be efficient when the DEA-CCR input oriented model is applied, compared with nineteen and twenty-four efficient ports when the DEA-BCC input-oriented model and the FDH model are respectively applied.

12 These results are not surprising. Spearman s rank order correlation coefficient between the efficiency derived by DEA-CCR, and DEA-BCC, DEA-CCR and FDH, and DEABCC and FDH methods are , , and respectively. The positive and high Spearman s rank order correlation coefficients indicate that the rank of each DMU derived by the three methodologies is similar. Also, the small absolute value of the spearman s rank suggests that the efficiency of ports is not a significant influence by its size. Table 6. A Proection of Inefficient DMU to be Efficient Baltimore Actual 2001 Proection Value by Type of DEA Model CCR-I BCC-I FDH-I # of Berths Length 3, Total Area 3,690, Storage 12, Shipshore Cranes Front-End Handlers Yard Tractors Yard Chassis Size (TEU) 499, , , NY/NJ Actual 2001 Proection Value by Type of DEA Model CCR-I BCC-I FDH-I # of Berths Length Total Area Storage Shipshore Cranes Front-End Handlers Yard Tractors Yard Chassis Size (TEU) One of the most important issues in DEA and FDH methodologies is diagnosis. In this analysis, Baltimore and the New York / New Jersey ports were studied. Table 6 shows how these ports could improve their efficiency under performance measurement models. For instance, for the New York / New Jersey port, for DEA-CCR and DEA-BCC methodologies one possible case might be that the Shipshore Cranes should be reduced to 21 cranes. Based on the FDH results the Shipshore Cranes should be reduced to 19 cranes. A similar analysis could also be made for the other inputs variables as wells. In the case that one of these methodologies has given efficiency scores the proection value has to be similar as the actual value.

13 CONCLUSION This paper contributes to the extant research in that the two non-parametric approaches of DEA and FDH have been studied comparatively within the North America container terminal industry. Analysis of the efficiency yielded by two DEA models (CCR and BCC) and the FDH model confirms that the DEA and FDH mathematical programming methodologies tend to give significantly different results. Thus, the choice of methodology matters if one is interested in ranking DMUs in terms of efficiency and seeking to identify the potential source for improvements in the production of inefficient producers. It is clear that a combination of DEA and FDH analysis can be of great significance and value to the decision makers of ports and terminals. On the one hand, the results from applying the FDH model identify the most obvious efficient counterpart(s) for the inefficient DMUs to learn from in terms of realistically comparable industry best practice. This result is convincing because these efficient counterparts are real. However, the FDH model is more likely to identify as efficient DMUs that are not performing well. In this respect, DEA has greater potential to provide efficient goals for the DMUs to work towards, although these goals should be subect to further study in terms of their feasibility in practice. This paper compares cost efficient estimates of North American ports using a variety of econometric and mathematical programming methodologies. The principal obective is to provide new information on the effects of the choice of methodology on cost efficiency estimates. The secondary obective is to analyze some of the results. The findings indicate that the choice of efficiency estimation methodology makes a significant difference in terms of the estimated cost efficiency values. The efficiency rankings are well-preserved within the two econometric methodologies. Cross-sectional data were unitized in this study within the North American ports. One of the surprising results was that some maor ports were suffering from inefficient production. On the basis of using cross-sectional data, however, this inefficiency could very likely be caused by a investment in future production. In order to overcome this potential source of bias in the efficiency estimates derived, an approach based upon panel data is more suited to an analysis that attempts to reduce the long-term efficiency trends of container terminals. REFERENCES 1. Ashar A., (1997) Counting the Moves, Port Development International, Volume 13, pp Bazargan, M. and Vasigh, B. (2003) Size versus Efficiency: a Case Study of US Commercial Airports, Journal of Air Transport Management, Volume 9, Issue 3, pp Cullinane, K. Song, W. Ji, P. and Wang, T. (2004) An application of DEA Windows Analysis to Container Port production Efficiency, Network Economics, Vol 3, Issue 2, pp Charnes, A. Cooper, W. Rhodes, E. (1978) Measuring the Inefficiency of Decision making Units, European Journal of Operation Research 2, pp Jara-Diaz S. Martines-Budria, E. Cortes, C. and Basso, L. (1992) A Multi-output Cost Function for the Services of Spanish Ports Infrastructure, Transportation 29, pp Jim Wu YC., and Goh M., (2010) Container Port Efficiency in Emerging and More Advanced Markets, Transportation Research Part E, Volume 46, pp Kaisar E., Pathomsiri S., and Haghani, A., (2006) Efficiency Measurement of US Ports Using Data Envelopment Analysis (DEA),National Freight Conference, Long Beach, CA.

14 8. Lovell C., and Vanden Eeckaut P., (1993) Frontier Tales: DEA and FDH, in mathematical Modeling in Economics: Essays in Honor of Wolfgang Eichhorn, Springer-Verlag, Berlin, New York, pp MARAD, 2004 Public Port Financing in the US. US Marine Administration Washington, USDOT. 10. Marlow, P. and Paixão, A. C. (2002), Measuring Lean Ports Performance, Proceedings of the International Association of Maritime Economists Conference, November, Panama, pp Martin, J. C. Roman, C. (2001) An Application of DEA to Measure the Efficiency of Spanish Airports Prior to Privatization, Journal of Air Transport Management, Volume 7, Issue 3, pp Roll, Y. and Hayuth, Y. (1993) Port Performance Comparison Applying Data Envelopment Analysis (DEA), Maritime Policy and Management 20, pp Tabernacle, J.B. (1995) A Study of Charges in Performance of Quayside Container Cranes, Maritime Policy and Management, Vol. 22, No. 2, pp Talley, W. K. (1998) Optimum Throughput and Performance Evaluation of Marine Terminals, Maritime Policy and Management, Vol. 15, No. 4, pp Tongzon, J.L. (1995) Determinants of Port Performance and Efficiency Transportation Research 29A, pp Turner, H. Windle, R. and Dresner, M. (2004) North American Containerport: , Transportation Research Part E, 40, pp UNCTAD, (1976), Port Performance Indicators, TD/B/C.4/131/Supp.1/Rev1, New 18. Varaprasad, N. (1986) Optimum Port Capacity and Operating Policies, a Simulation Study, Transport Policy and Decision. 19. Wang T., Cullinane k.p.b., and Song D-W., (2003) Container Port Production Efficiency: A Comparative Study of DEA and FDH Approach, Journal of the Eastern Asia Society for Transportation Studies, Vol 5, pp Wang, T., D-W. Song and K. Cullinane. (2005) The Application of Mathematical Programming Approaches to Estimating Container Port Production Efficiency, Journal of Productivity Analysis 24, pp