Tecnicas Para Una Eficiente Asignacion Para La Manipulacion de Cargas

ARTICLE IN PRESS

Int. J. Production Economics 109 (2007) 149–161 www.elsevier.com/locate/ijpe

Firm and time varying technical and allocative efficiency: An application to port cargo handling firms Ana Rodrı´ guez-A´lvareza,, Beatriz Tovarb, Lourdes Trujillob,1 a

University of Oviedo, Spain Research Group Economics of Infrastructure and Transport, EIT, University of Las Palmas de Gran Canaria, Spain

b

Received 30 December 2005; accepted 27 November 2006 Available online 9 January 2007

Abstract In this paper we present an econometric model to calculate both, the technical and the allocative efficiency in cargo handling firms in the port of Las Palmas (Spain). To achieve these aims, we estimate a system of equations consisting of a translog input distance function and cost shares equations. Using this procedure we can check the hypothesis in which, given technology and prices, terminal port inputs are not optimally allocated in the sense that costs are not minimized. The main contribution of this paper is to apply an empirical model that allows the unbiased estimation of allocative inefficiency of input use in two ways: an error components approach and a parametric approach. We also avoid the Greene Problem and allow allocative inefficiency to be systematic. Both size and traffic mix are first shown to be sufficiently diverse as to allow for a reliable estimation of a representative flexible (translog) function. r 2007 Elsevier B.V. All rights reserved. Keywords: Distance function system; Morishima elasticities; Technical and allocative efficiency; Spanish ports terminals

1. Introduction In the last decade, several models have been proposed to estimate time-varying technical efficiency. These models could be grouped depending Corresponding author. Departamento de Economı´ a, Universidad de Oviedo, Campus del Cristo, 33071 Oviedo, Spain. Tel.: +34 985 10 48 84; fax: +34 985 10 48 71. E-mail address: [email protected] (A. Rodrı´ guez-A´lvarez). 1 A Previous version of this paper has been published like working paper in FUNCAS (Fundacio´n Espan˜ola de las Cajas de Ahorro) n o 201/2005 and were presented at the Permanent Seminar on %Efficiency and Productivity, Oviedo, April 2004 and at the North American Productivity Workshop, Toronto, Canada June 2004. This research was funded with grants from Gobierno Auto´nomo de Canarias, Project PI2003/188.

on the approach chosen to model the inefficiency. On the one hand, there are those who model technical inefficiency through an error component (see, for example, Kumbhakar, 1990; Battese and Coelli, 1992, 1995; Heshmati and Kumbhakar, 1994; Heshmati et al., 1995 or Cuesta, 2000). These models involve the disadvantage of making particular distributional assumptions for the one-sided error term associated with technical efficiency. On the other hand, there are those who model technical inefficiency through the intercept of the function (see, for example, Cornwell et al., 1990; Lee and Schmidt, 1993 or Atkinson and Primont, 2002). In this way, these models avoid making particular distributional assumptions. In this paper we can obtain technical efficiency indices, which may vary

0925-5273/$ - see front matter r 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpe.2006.12.048

ARTICLE IN PRESS 150

A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

through time as well as across firms following this second approach. With regard to allocative efficiency, Atkinson and Cornwell (1994) present two methods that permit the calculation of allocative inefficiency: the parametric approach and the error component approach. Fa¨re and Grosskopf (1990) and Atkinson and Primont (2002) demonstrate that replacing the usual cost frontier with an input distance function, the main drawbacks of the parametric approach can be overcome by obtaining firm and time allocative efficiency indices. With regard to the error component approach, the advantages of the distance function are developed in Rodrı´ guez-A´lvarez et al. (2004) which dealt with a hypothesis in which allocative efficiency is time-invariant and only varies across firms. In this paper we extend the analysis to the case when the efficiency is firm and time-varying in both approaches. In adition, we can calculate firm and time-varying technical efficiency and, separately, a measure of technical change. To do this, we present a distance system that comprises an input distance function and the associated share cost equations. Finally, we also extend the analysis by calculating Morishima elasticities. To illustrate our methodology we apply it to panel data using a sample of cargo handling firms in Spanish ports. The first empirical studies that have attempted to measure port efficiency based on frontier models appear in the 1990s. There are two different techniques to carry out this type of study. The first one, a non-parametric programming method, is called data envelopment analysis (DEA), (i.e. Roll and Hayuth, 1993; Martı´ nez Budrı´ a et al., 1999; Tongzon, 2001; Valentine and Gray, 2001; Bonilla et al., 2002; Martı´ n, 2002; Pestana, 2003; Estache et al., 2004), and the second is through stochastic frontier analysis (SFA), (i.e. Liu, 1995; Ban˜os Pino et al., 1999; Coto Milla´n et al., 2000; Notteboom et al., 2000; Cullinane et al., 2002; Estache et al., 2002; Cullinane and Song, 2003; Dı´ az, 2003; Tongzon and Heng, 2005). Both methods allow the derivation of estimates of relative efficiency levels for all the operators compared.2 Out of these, only four of them have investigated the efficiency of Port Terminals: Notteboom et al. (2000); Cullinane et al. (2002); Cullinane and Song (2003); Tongzon and Heng (2005) and they have several characteristics in common. Firstly, they 2

For more details, see Coelli et al. (1999).

estimate a stochastic frontier production function; secondly, they model technical inefficiency through an error component thirdly; they only measure technical efficiency and finally, they consider that technical efficiency is time-invariant. However, Song et al. (2001) estimate the efficiency of the terminal operating companies based on both pooled data set (time-variant) and panel data (timeinvariant). Our paper is the first one dealing simultaneously with firm and time varying technical and allocative port terminals efficiency, and it is also the first one applying a distance function to port terminals. The paper is organized as follows: Section 2 provides an overview of port terminals and their regulation in Spain. In Section 3 the model is presented. Section 4 concerns itself with the econometric model. The data are described and the results are presented in Section 5. The final section contains brief concluding comments. 2. Port terminals and their regulation in Spain Although there are cases where several firms share a port terminal, in our case each cargo handling firm exclusively operates its own terminal in a consessional basis. The terminals analyzed are typical medium size port ones. Terminal prices are subject to price caps, which are seldom binding, but employment is highly regulated. This is not an unusual situation around the world. Economic activities within a port are multiple and heterogeneous. On the one hand, among them cargo handling has been one of the most affected by technological changes and by competition among ports on the other. The importance of this activity is evident when realizing that it means from 70% to 90% a vessel’s disbursement account (De Rus et al., 1994). Cargo handling services are usually performed in port terminals. Technological changes have increased the relative importance of specific terminals within port areas (e.g. multi-purpose,3 containers, liquid and solid bulk). Terminal facilities have now become heavily capital intensive and, depending on port size, more specialized as well, playing a key role in the choice of port by shippers. The role of the port terminals 3

A multiple-purpose (MP) terminal is designed to serve heterogeneous traffic, including containerized and non-containerized cargo. It can be transformed into a specialized one (e.g. containers only) by changing equipment.

ARTICLE IN PRESS A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

within the logistic system makes them key actors of the port industry, playing a central role in the increasing competition within the sector. In addition, the private sector has become increasingly interested in this type of activity. This has shifted the focus of port authorities competitive strategy design from the port as a whole to the terminals, making them the most important elements within the port industry. As pointed out by Heaver (1995), this change of focus is the main factor explaining the increased competition within the sector. The cargo handling service is usually viewed as one that has to be directly provided by the public sector or by private firms through concession contracts. The regulation of the Spanish ports system is based upon a scheme that allows the combination of public ownership of the port infrastructure (docks, land, and so on) with private ownership of the superstructure (warehouses, cranes, etc). The public authority determine the conditions under which the private initiative has to operate by fixing maximum prices, characteristics and length of concessions, and other conditions (for more details see Tovar et al., 2004). In the production of cargo handling services, the following groups of factors are required: basic infrastructure, superstructure, machines and mobile equipment, as well as labor. Traditionally, labor in a port has been strongly regulated, although changes have taken place worldwide during the last decade or so. The stevedores or port workers working in Spanish port terminals are divided into two categories: those who are on the payroll (ordinary employment—LC) and those who are not (special employment—LE). These latter can be recruited on a provisional basis by any company to work 6 h shifts. Regulation drives the level and composition of port labor (including the choice between LE and LC) for each terminal. The possibility of contracting port workers for specific operations (LE) provides the stevedoring companies4 with some flexibility, allowing them to adjust employment to terminal traffic levels. However, this flexibility has its limits, since under the current legislation the operators do not have total

151

freedom to decide how and to what extent each type of worker has to be used.5 The stevedoring companies have:

  

The obligation to perform at least 25% of their activities (in tons) with port workers on their payroll (LC). The obligation to use at least one LE port worker in each shift. A global limit on the number of operations they can perform with LC workers of 22 per month; once they have reached this level, they are required to rely on LE workers. Moreover, as we can check in the empirical application, the regulation imposes certain degrees of complementarity between both types of port workers so as to use them jointly.

3. Modelling firm and time-varying technical and allocative efficiency To calculate firm and time-varying technical and allocative efficiency we propose an empirical model, which consists of an input distance function and the associated input cost share equations: ln 1 ¼ ln Dðypt ; xpt ; K pt ; DTÞ þ vpt þ upt ,

(1)

xipt wipt q ln Dðypt ; xpt ; K pt ; DTÞ ¼ þ vipt þ Aipt , C pt q ln xipt

(2)

where D(ypt,xpt,Kpt, DT) is the short-run input distance function; y is an output vector, x is a variable input vector, K is a quasi-fixed input, DT is a time year dummy to control for neutral technical change, p denotes port terminal and t time. The vpt and vipt error components represent statistical noise, and are assumed to be distributed as multivariate normal with zero mean. 3.1. Measuring technical efficiency The error component uptX0 in (1) represents the magnitude of technical efficiency (TE). We follow Cornwell et al. (1990) who specifies a model that allows us to estimate time-varying technical efficiency levels for individual firms, without making strong distributional assumptions for technical inefficiency or random noise. Thus, if the constant

4

These are firms that are allowed to handle cargo. In our case, the three terminals are also stevedoring companies; conversely they could not do handling cargo by their own, but by contracting a stevedoring company.

5 A collective agreement is a three-way agreement among the State Stevedoring Association, the port workers, and the stevedoring companies.

ARTICLE IN PRESS 152

A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

in (1) is B0, then, it is possible to write: bpt ¼ B0 þ upt ¼ bpa Dp þ bpb Dp t þ bpc Dp t2 ,

(3)

where Dp is a dummy variable for the port terminal p, bpa; bpb; bpc are parameters to be estimated for this port terminal, and t is a time trend. However, there is an interpretation problem because there is no easy way to empirically distinguish in (3) between technical inefficiency or technical change (for example, there are contradictory interpretations in Cornwell et al., 1990 or Good et al., 1995). To solve this problem, and following Heshmati and Kumbhakar (1995) and Heshmati et al. (1995) we have included in (1), the dummy time (DT). In this way, by including a time variable among the regressors, it would be possible to resolve the interpretation problem in the Cornwell et al. (1990) specification, since the time variable is associated with technical change and the error component (3) is associated with technical efficiency, which is allowed to vary across producers and through time (for details see, for example, Lovell, 1996). In this sense, the bpa captures time-invariant TE whereas bpb and bpc capture time-varying TE. Thus, each producer has its own intercept (bpt), which is allowed to vary quadratically through time at producer-specific rates. The TE of a port terminal in a time period is obtained from the estimated intercepts as TEpt ¼ exp(upt) where upt ¼ bpt min(bpt). The efficiency index constructed in this way, ranges from 0 to 1. Thus, in each period at least one port terminal is estimated to be 100% technically efficient (with value 1), although the identity of the most technically efficient port terminal may vary through time. 3.2. Measuring allocative efficiency To calculate allocative efficiency two econometric approaches are available (Atkinson and Cornwell, 1994): (a) an error components approach, and (b) a parametric approach (a) The error components approach In this approach we model allocative inefficiency through an error component. Thus in (2) the error components Aipt4 ¼ o0, i ¼ 1,y,n, represent allocative inefficiency, here represented by the difference between actual and stochastic shadow input cost shares form (for more details see Rodrı´ guez-A´lvarez and Lovell, 2004). Moreover, if Aipt is positive, the

input i is being over-utilized with regard to other inputs and vice versa. In our case it is possible to specify allocative inefficiency for the input i as Aipt ¼ aipa Dp þ aipb Dp t þ aipc Dp t2 ,

(4)

where Dp is a dummy variable for the port terminal p and aipa;aipb; aipc are parameters to be estimated for this port terminal and for the input i and t is a time trend. The aipa captures time-invariant allocative inefficiency, whereas aipb and aipc capture timevarying allocative inefficiency. (b) The parametric approach: After estimation of systems (1)–(2), shadow price ratios are determined from the dual of Shephard’s Lemma as qDðy; x; K; DTÞ=qxi wsi ¼ , qDðy; x; K; DT Þ=qxj wsj

(5)

where ws is the shadow prices vector. If the allocative efficiency assumption is satisfied, these shadow price ratios coincide with market price ratios. However, if there is allocative inefficiency, the two price ratios differ. To study such deviations, a relationship between the shadow prices (obtained through the distance function) and the market input prices is introduced by means of a parametric price correction: wsi ¼ ki wi .

(6)

Dividing (6) by the corresponding expression for input j we obtain wsi wi ¼ kij , wsj wj

(7)

where kij ¼ ki/kj. From (7) the degree to which shadow price ratios differ from market price ratios is calculated. If kij ¼ 1, there is allocative efficiency; while if kij4(o)1, input i is under-utilized (overutilized) relative to input j. If we have panel data, it is possible to obtain, for each pair of inputs, producer and time-specific allocative efficiency indices kij. Thus, with both proposed approaches (the error component and the parametric approach) we can get time and firm varying indices of allocative efficiency. Moreover, it is important to distinguish between these two measures of allocative inefficiency (the parametric approach and the error components approach). In the error components approach, the aipts represent the systematic allocative inefficiency for each input. In the parametric approach, kij indicates the allocative inefficiency for each pair of inputs.

ARTICLE IN PRESS A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

Moreover, in the error components approach, behind the aipt lies the assumption of an additive relationship between wi and wis. In the parametric approach a multiplicative relationship between wi and wis is specified, yielding the coefficients kij as indices of allocative inefficiency. However, it is also important to emphasize the existing relationship between the two approaches. When persistent inefficiency exists (that is, if the aipt have mean values significantly different from zero), their inclusion is necessary if the estimated parameters, and consequently the estimated kij, are unbiased. This implies that the inclusion of the parameter ai in the empirical model is indispensable, in order to obtain unbiased estimates of the kij. 4. Econometric specification We now consider how to estimate systems (1)–(2). To do this, we have chosen a flexible functional form, a translog short run multiproduct input distance function which is specified as ln 1 ¼ B0 þ

m X

ar ln yrpt þ

r¼1

þ

n X

bi ln xipt þ

i¼1

þ xf ln K pt þ xff þ

m X

m X m 1X ars ln yrpt ln yspt 2 r¼1 s¼1

n X n 1X b ln xipt ln xjpt 2 i¼1 j¼1 ij n X 1 xfi ln K pt ln xipt ln K 2pt þ 2 i¼1

xfr ln K pt ln yrpt þ

r¼1

þ

T X

n X m X

rri ln yrpt ln xipt

i¼1 r¼1

gT DT þ upt þ vpt ,

ð8Þ

T¼1

n m X X xipt wipt ¼ bi þ bij ln xjpt þ rri ln yrpt C pt j¼1 r¼1

þ xfi ln K pt þ Aipt þ vipt ,

ð9Þ

where y ¼ ðy1 ; . . . ; ym Þ is an output vector, x ¼ ðx1 ; . . . ; xn Þ is a variable input vector, K is a quasifixed input, DT is a time dummy for year T, p denotes port terminal; t denotes months and vpt, vipt, upt and Aipt are disturbance terms which have been already defined. a, b, x, r, g are parameters related to outputs, inputs, quasi-fixed input, cross term between outputs and inputs and time, respectively. Homogeneity of degree +1 of the input distance function in variable inputs is enforced by imposing

the

n P

restrictions

i¼1

0 ð8 r ¼ 1; . . . ; mÞ;

n P

bi ¼ 1;

153 n P j¼1

bij ¼ 0;

n P

rri ¼

i¼1

xfi ¼ 0: We also impose the

i¼1

symmetry conditions ars ¼ asr ; bij ¼ bji ; xfi ¼ xif ; xfr ¼ xrf ; rri ¼ rir . The estimation of this equation system provides the suitable empirical context to calculate the technical and allocative efficiency by means of the above proposed methodology. 5. Data and results 5.1. Data description The sample consists of three multipurpose port terminals operating within the area of Las Palmas Port, located in the Canary Islands (Spain). The fact that all the port terminals considered are in the same port is a key advantage in the coherence and uniformity of the accounting data used. Thus, the regulation is the same, and environmental factors are equal or very similar for all of them. We have monthly data from 1992 until 1997 for Terminal 1 (T.1), from 1991 until 1999 for Terminal 2 (T.2), and from 1992 till 1998 for Terminal 3 (T.3). The final panel data set consists of 264 observations. Although these terminals mainly deal with containers, they also operate roll-on/roll-off cargo (roro) as well as general break-bulk cargo, so we distinguish three outputs measured in total tons: containers (CONT), ro-ro cargo (ROD) and general break-bulk cargo (MG). The input variables are: port workers: ordinary (LC) and special6 (LE), capital (GK), intermediate consumption (GI). We also consider a quasi-fixed input: total area (K). The information available regarding the amount of workers7 used is expressed in number of shifts per month. A shift is a 6 h work schedule. The total monthly labor expense for the terminals is calculated as the sum of the cost of each type of worker. Capital covers all the components of tangible assets of the company —i.e. buildings, machines, etc. The monthly cost results from the addition of 6 This can be recruited on a provisional basis by any company to work 6 h shifts. 7 Information about port labor is very difficult to obtain. Out of the previous papers dealing with port terminal efficiency only Cullinane and Song (2003) has it.

ARTICLE IN PRESS A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

154

Table 1 Monthly average input, output and expense values get for the entire sample and for each terminal Variable

Unit

Terminals Sample Mean

Outputs CONT MG ROD INPUTS LC LE GI GK K Expenditure GLC GLE

1000 Ton 1000 Ton 1000 Ton

59.2 5.6 2.1

T.1 S.E.

T.2

Mean

41.57 6.35 2.36

S.E.

T.3

Mean

S.E.

Mean

S.E.

53.1 0.6 1.0

9.72 0.78 0.71

33.5 9.9 0.8

7.45 7.39 0.86

97.4 4.4 4.7

54.36 3.12 2.49

Number of shifts per month Number of shifts per month 1000 PTAS deflated 1000 PTAS deflated M2

336.4 339.4 24,534.2 12,985.4 61,484.4

206.13 344.0 161.40 207.5 8445.04 21,961.4 7728.52 6,063.2 11758.16 63,971.8

140.28 93.11 5,485.22 429.87 7,892.55

251.0 400.4 20,573.2 11,043.0 57,530.6

49.90 193.44 3,556.72 3,939.85 2,597.86

439.8 374.0 31,832.1 21,416.1 64,435.8

306.94 75.70 10192.23 7119.51 18481.79

1000 PTAS deflated 1000 PTAS deflated

17,964.1 21,447.9

8563.95 13,113.6 12515.12 18,759.9

6,826.70 6,911.54

14,463.6 20,738.9

3,592.70 9,453.66

26,622.4 24,663.6

7979.02 17967.74

the accounting depreciation for the period plus the return on the active capital of the period.8 With regard to area, the terminals under analysis can make use of a specific area that has been granted under concession, which may be increased by provisionally renting—upon prior request— additional area from the Port Authority. The addition of both types of areas is called total area, which is monthly measured in square meters. Lastly, the rest of the productive factors used by the company that have not been included in any of the three preceding categories, such as office supplies, water, electricity, and the like, have been denominated as intermediate consumption. The monthly expense results from the aggregation of the rest of the current expenses other than depreciation, personnel expenses and payment for area, after the pertinent corrections, in such a manner that, the resulting monthly expense truly reflects consumption and not accountancy. The total monthly production expenses for the terminals result from the aggregation of expenses of all the productive factors defined above. Table 1 8 This rate of return evidences the compensation earned by riskfree capital, which is made up of bank interest plus a risk premium. It has been considered that, for the period under analysis, the return for both concepts amounts to 8% per annum. The price of capital is the quotient of the cost of capital divided by the active capital of the period (net fixed assets under exploitation for a given period t.)

shows the monthly values obtained for the entire sample and for each of the three terminals, both in terms of the defined inputs and outputs, as well as the total expense incurred during service provision.9 It is worth stressing that data was gathered directly from the firms files and that all the details were discussed with executives when necessary, particularly for the monthly assignment of expenses. Data is described in detail in Tovar (2002). On the one hand, out of the three products, general break-bulk cargo (‘‘general cargo’’) represents an average of 9.9% of the monthly total tons moved, containers represent a 87.4% and ro-ro 2.7%. On the other hand, labor costs account for an average of 53%10 of the monthly expenses for the entire sample. Total area represents 13%, capital amounts to 8% and intermediate consumption reaches 26%. Within port workers, ordinary ones account for 46% while special workers represent 54%. The figures reveal similar patterns per company. Moreover, the analysis of the information contained in Table 1 leads to a first approximation of the size of companies. Thus, taking into consideration the aggregated product volume, the largest company is T.3., followed by T.1. and by T.2. in last 9

All monetary variables have been deflacted. Note this is the same percentage that Cullinane and Song (2003) report as a typical expenditure of a port. 10

ARTICLE IN PRESS A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

position. It is to be noted that, where the variable used as the size indicator is the total monthly production expense (mean value), even though T.3. is still found to be in the first place, the other two companies, T.1 and T.2. swap positions. This result is due to monthly expenses and do not vary monotonically with total production. This makes the different composition of outputs a likely explanation for cost differentials when factor prices are similar for the three companies. For example, the only explanation for the expenses of T2, being larger than those of T1 would be the difference in the traffic mix, particularly, the larger volume of general cargo. This already suggests higher marginal costs for general cargo, which

155

reinforces the need for a multioutput analysis Jara-Dı´ az et al. (2005). 5.2. Results We have estimated systems (8)–(9) by means of iterative seemingly unrelated regressions (ITSUR), which is invariant to the omitted share equation. In Tables 2–4 we present the estimated values from the input distance system. The variables have been divided by the geometric mean. Therefore, the first-order coefficients can be interpreted as elasticities. At the sample mean, the regularity conditions are satisfied: it is non-decreasing and quasi-concave in inputs and decreasing in outputs.

Table 2 Distance system estimated Variable

Coefficient

Standard error t-Statistic

L(CONT) L(MG) L(ROD) L(LC) L(LE) L(GI) L(K) L(GK) L(CONT).L(CONT) L(CONT).L(MG) L(CONT).L(ROD) L(CONT).L(LC) L(CONT).L(LE) L(CONT).L(GI) L(CONT).L(K) L(CONT).L(GK) L(MG).L(MG) L(MG).L(ROD) L(MG).L(LC) L(MG).L(LE) L(MG).L(GI) L(MG).L(K) L(MG).L(GK) L(ROD).L(ROD) L(ROD).L(LC) L(ROD).L(LE)

0.32822 0.35813 0.01642 0.14139 0.25114 0.27117 0.18176 0.33629 0.13236 0.02318 0.01045 0.02655 0.03048 0.00436 0.33217 0.00830 0.00856 0.00155 0.00184 0.00071 0.00092 0.03581 0.00021 0.00322 0.00789 0.00561

0.3947 0.0057 0.0095 0.0369 0.0279 0.0458 0.1415 0.0435 0.0747 0.0086 0.0111 0.0132 0.0155 0.0074 0.1905 0.0046 0.0016 0.0014 0.0017 0.0017 0.0008 0.0317 0.0005 0.0027 0.0022 0.0022

8.3160 6.3068 1.7225 3.8291 8.9895 5.9193 1.2847 7.7245 1.7714 2.6832 0.9447 2.0089 1.9698 0.5858 1.7438 1.8100 5.6796 1.1364 1.1019 0.41710 1.0878 1.1298 0.4520 1.2079 3.6192 2.5637

Variable

Coefficient

Standard error t-Statistic

L(ROD).L(GI) L(ROD).L(K) L(ROD).L(GK) L(LC).L(LC) L(LC).L(LE) L(LC).L(GI) L(LC).L(K) L(LC).L(GK) L(LE).L(LE) L(LE).L(GK) L(LE).L(GI) L(LE).L(K) L(GI).L(GI) L(GI).L(GK) L(GI).L(K) L(K).L(K) L(K).L(GK) L(GK).L(GK) DT92 DT93 DT94 DT95 DT96 DT97 DT98 DT99

0.00130 0.08451 0.00099 0.02116 0.05233 0.04950 0.01692 0.02399 0.05444 0.03723 0.06949 0.10786 0.17636 0.05736 0.0433 0.31434 0.04762 0.11858 0.12548 0.09536 0.13487 0.15500 0.02817 0.13772 0.18434 0.13160

0.0009 0.0929 0.0005 0.0122 0.0114 0.0055 0.0340 0.0035 0.0138 0.0030 0.0054 0.0311 0.0053 0.0031 0.0158 0.6175 0.0102 0.0029 0.0396 0.0790 0.0945 0.1052 0.1173 0.1303 0.1393 0.1511

1.4431 0.9100 2.1150 1.7274 4.6108 8.9437 0.4972 6.8374 3.9455 12.4542 12.7574 3.4794 33.0481 18.4742 2.7515 0.5091 4.6748 41.0775 3.1696 1.2067 1.4279 1.4727 0.2402 1.0566 1.3232 0.8710

Equation

Mean

R2

Std. error of regression

Input distance function Ordinary worker share equation Special worker share equation Intermediate consumption share equation Capital share equation

— 0.236319 0.278165 0.323562 0.161953

— 0.7541 0.8406 0.8505 0.9602

0.079379 0.044199 0.042865 0.020191 0.012590

 Statistically significant at 10%. Statistically significant at 5%.

ARTICLE IN PRESS A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

Table 3 bpt Components (from Eq. 3) Parameters

Coefficients

Standard error

t-Statistic

b1A b1B b1C b2A b2B b2C b3A b3B b3C

0.0425450 0.0000061 0.0000782 0.1322430 0.0049411 0.0000437 0.2295630 0.0251980 0.0002957

0.1105030 0.0057844 0.0000493 0.0458950 0.0035596 0.0000294 0.1167930 0.0061530 0.0000526

0.3850170 0.0010705 1.5836800 2.8814200** 1.3880800 1.4861700 1.9655600** 4.0951200** 5.6142100**

Table 4 Aipt Components (from Eq. 4) Parameters

Coefficients

Standard error

t-Statistic

aLC1A aLC1B aLC1C aLC2A aLC2B aLC2C aLC3A aLC3B aLC3C aLE1A aLE1B aLE1C aLE2A aLE2B aLE2C aLE3A aLE3B aLE3C aGK1A aGK1B aGK1C aGK2A aGK2B aGK2C aKI3A aGK3B aGK3C

0.3719030 0.0091169 0.0000545 0.2436080 0.0061470 0.0000365 0.1496150 0.0006403 0.0000066 0.3003560 0.0131560 0.0001023 0.0951980 0.0056841 0.0000380 0.0117950 0.0011804 0.0000091 0.1602610 0.014876 0.0000181 0.1904440 0.0000873 0.0000018 0.1849410 0.0006198 0.0000031

0.0456560 0.0020863 0.0000187 0.0470980 0.0007202 0.0000053 0.0527010 0.0014934 0.0000141 0.0479150 0.0020900 0.1849460 0.0342090 0.0008852 0.0000068 0.0440560 0.0011371 0.0000103 0.0437710 0.0004296 0.0000042 0.0437210 0.0002108 0.0000017 0.0456000 0.0005529 0.0000051

8.145760** 4.369740** 2.911360** 5.172390** 8.534300** 6.786440** 2.838960 0.428774 0.470838 6.268500** 6.294670** 5.534250** 2.782840** 6.420770 5.569630 0.267731 1.038090 0.887793 3.661320** 3.462580** 4.257970** 4.355870** 0.414233 1.057590 4.055760** 1.120990 0.622118

In Table 2, the estimated parameters are shown. It can be seen that all first-order parameters are statistically significant and have the correct sign, except the quasi-fixed input total area (K). As the quasi-fixed input total area (K) is not statistically significant this means that the marginal productivity of total area could be zero. This could be due to the fact that port terminal infrastructure and superstructure must be built with determined minimum

dimensions, so it is not possible to enlarge a terminal port in a continuous way. So, given that terminals keep growing, they are having more total area than previously needed. 5.2.1. Technical efficiency In Table 3, and following Section 3.1 we present the estimated coefficients from which it is possible to obtain the technical efficiency index (upt). These indices have been represented in Fig. 1 (following Cornwell et al. (1990) approach). The temporal pattern of technical efficiency for the three terminals shows that T.3 is the most efficient for almost the whole period, when its technical efficiency score begins getting worse. This occurs, more or less, when T.3 moves from a smaller to another bigger place inside the port area. This change of size requires a huge investment which probably drives this result. Remember that T3 is the largest company, so it seems to exhibit a relationship between size and efficiency (this relationship has also been reported in Cullinane et al. (2005). Terminal T.2 presents a declining technical efficiency score in the first years and an increasing one in the latter ones, being finally substituted by T.3 at the end of the period. Lastly, T.1 shows a general decline in technical efficiency. Other technical efficiency scores are shown in Fig. 2. These are calculated considering the best observation as the reference technology. Fig. 2 tells us a similar story. In terms of the trend in technical efficiency, T.3 shows a general decline (with the exception of the few first periods), T.2 and T.1 present a similar, but softer, pattern than in Fig. 1.

1.2 Technical Efficiency

156

1 0.8 0.6 0.4 0.2 0

1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 Time (CSS-1990)

T.1

T.2

T.3

Fig. 1. Temporal pattern of technical efficiency for the three terminals.

ARTICLE IN PRESS

Technical Efficiency

5.2.2. Allocative efficiency: error components approach In Table 4, following the error component approach, estimated coefficients and asymptotic standard errors for AE parameters (Aipt components) are shown. Remember that Aipt represent the systematic allocative inefficiency for each input. The results of the estimation of the firm-specific temporal pattern of allocative inefficiencies following the error components approach are reported in Figs. 3 and 4 for the three terminals with respect to ordinary and special port workers (LC and LE), respectively. It is interesting to note that the two graphs represent the inverse image of each other. At the very beginning T.1 and T.2 over-utilized LC and underutilized LE, but as time passes these two terminals achieve allocative efficiency in LC but not in LE, which goes on being over-utilized. However, T.3 over-utilized LC and underutilized LE during the whole period. The joint analysis of Figs. 3 and 4 shows that each terminal has a preferred adjustment factor. It is LC for T.1 and T.2, while it is LE for T.3. This would suggest that firms are operating with allocative efficiency with respect to one factor

1.2 1 0.8 0.6 0.4 0.2 0 1

9

17 25 33 41 49 57 65 73 81 89 97 105

Time

T.1

T.2

T.3

Allocative Inefficiency

Fig. 2. Technical efficiency score considering the best observation as the reference technology.

Allocative Inefficiency

A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

157

0.15 0.1 0.05 0 -0.05 -0.1 -0.15 -0.2 1

10

19

28

37

46

55

64

73

82

91 100

Time LET.1

LET.2

LET.3

Fig. 4. Firm-specific temporal pattern of allocative inefficiencies for LE (error components approach).

and with inefficiency with respect to the other factor. This behavior is best understood by considering the relevance of regulation. Indeed, it is useful to remember that regulation drives the level and composition of port labor (including the choice between LE and LC) for each terminal. Regulation impedes the labor adjustments for both types of workers jointly. This is why they end up trying to maintain one degree of freedom and may tend to focus on adjustments in one of the workers’ categories at a time. Figs. 3 and 4, indeed, clearly show that the employment policies of T.1 and T.2 differ from those followed by T.3. Finally, as we can see from Figs. 5 and 6 the three terminals are allocatively inefficient in the use of the two other factors considered: intermediate consumption (GI) is over-utilized and capital (GK) is under-utilized. The orders of magnitude are quite small in GI. In the case of capital this probably simply reflects the difficulty of adjusting capital perfectly in a situation of growth as was the case for these terminals during the period of analysis. The area comes first and then, the rest of capital (buildings, cranes, and so on).

0.3 0.25 0.2 0.15 0.1 0.05 0 -0.05 1

9

17 25 33 41 49 57 65 73 81 89 97 105

Time LCT.1

LCT.2

LCT.3

Fig. 3. Firm-specific temporal pattern of allocative inefficiencies for LC (error components approach).

5.2.3. Allocative efficiency: parametric approach In the parametric approach we get firm and timespecific measures of AE for each pair of inputs. However, in order to conserve space we have only reported their values evaluated at the sample mean in Table 511 which have been estimated from Eqs. (5) and (7). Remember that kijo1 means that the ratio of the shadow prices of input i to that input j is lower than 11

Firm and time-specific results are available from the authors on request.

ARTICLE IN PRESS A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

Allocative Inefficiency

158 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0

1

9

17 25 33 41 49 57 65 73

8

1

89 97 105

Time GIT.1

GIT.2

GIT.3

Allocative Inefficiency

Fig. 5. Firm-specific temporal pattern of allocative inefficiencies for GI (error components approach).

0 -0.05

1

9 17 25 33 41 49 57 65 73 81 89 97 105

-0.1 -0.15 -0.2 -0.25 Time GKT.1

GKT.2

GKT.3

Fig. 6. Firm-specific temporal pattern of allocative inefficiencies for GK. (error components approach).

Table 5 Average values for coefficients kij Meana

Coefficients kOrdinary Worker, Special Worker k LC,LE K Ordinary Worker, Intermediate

t-Statistic

0.327803

1.726430

0.287873

1.547900

0.696075

6.500820

the corresponding ratio of actual prices. The analysis of the average estimated values for the kij reinforced the conclusions of the previous paragraphs and shows that LC is over-utilized relative to LE and GK. Moreover, GK is under-utilized relative to all the other production factors. Finally, the coefficient kij which links GI with the other productivity factor is always not statistically significant, except in the case of GK. With respect to labor, the result may reflect the labor-specific regulatory environment which impedes the adjustments needed by the operators. As for capital, it may be useful to point out that the levels of inefficiency are probably due to the impossibility of full adjustment as a result of indivisibility as previously mentioned; terminal or port infrastructure and superstructure must be built with determined minimum dimensions.12 The figures per company reveal similar patterns as we can see from Fig. 7. In addition, it is possible to analyze the degree of substitutability between inputs (or the isoquant curve) by means of Morishima elasticities of substitution. Formally, Morishima elasticities inform us about the variations of shadow prices given a variation in the input quantity. Positive values of the elasticities indicate complementarity between inputs and negative values imply substitutability. The Morishima elasticities have been divided into two components: M ij ¼ E ij  E ii . The term Eij is the cross-elasticity shadow price, and tells us whether the input pairs are net substitutes or net complementaries, while Eii represents the elasticity price characteristic of each input. Formally: M ij ðy; xÞ ¼  d ln ½ðDi ðy; x=xi Þ=Dj ðy; x=xj Þ=d ln ½xi =xj 

¼ xi Dij ðy; xÞ=Dj ðy; xÞ  xi Dii ðy; xÞ=Di ðy; xÞ

Consumption

K LC,GI K Ordinary Worker, Capital K LC,GK k Special Worker, Intermediate

¼ E ij ðy; xÞ  E ii ðy; xÞ.

0.059402

0.504053

Consumption

K LE,GI K Special Worker, Capital K LE,GK K Intermediate Consumption, kGI,GK

0.547863 Capital

0.573215

6.949870 10.37480

Note: We have tested the significance of the indices using the Wald test. a Evaluated at the means of the data using parameter estimates of (8)–(9).  Statistically significant different from one at 10% level.  Statistically significant different from one at 5% level.

ð10Þ

The calculations made of the Morishima substitution elasticities (Mij) and their two components (Eij and Eii) are shown in Tables 6–8. From the results, we can deduce little scope for substitution possibilities among the different pairs of inputs. The estimation of the Eii shows the proper sign. Moreover, our estimation of the cross-elasticities shadow price Eij, behaves quite well. As expected, due to regulation, LC and LE are net soft complementaries, and the degree of complementarity between 12

It is not possible to enlarge a terminal port in a continuous way.

ARTICLE IN PRESS A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

159

1.4 1.2 1 0.8 0.6 0.4 0.2 0

KLCLE

KLCGI

KLCGK Sample

KLEGI T1

T2

KLEGK

KGIGK

T3

Fig. 7. Average firm allocative inefficiencies for each pair of inputs (parametric approach).

Table 6 Morishima substitution elasticities: Mij

M M M M M M

lc,le lc, gi lc, gk le, gi gi, gk le, gk

Meana

t-Statistic

1,0587 0.6678 0.7790 0.5271 0.1790 0.6726

10.2714** 4.2113** 10.4009** 6.7505** 1.7666* 9.9710**

M M M M M M

le, lc gi, lc gk, lc gi, le gk, gi gk, le

Table 8 Time effects Meana

t-Statistic

Period

TCa

t-Statistic

1.1535 0.004 0.4777 0.0729 0.4358 0.4991

6.0648** 0.0035 10.019** 0.7081 10.1179** 9.9215**

1991–92 1992–93 1993–94 1994–95 1995–96 1996–97 1997–98 1998–99

0.125484 0.030120 0.039505 0.020130 0.183165 0.109557 0.046616 0.052731

3.169570** 0.605594 1.465350 0.880385 6.922290** 3.172850** 1.231000 1.272980

Table 7 Cross and own elasticities: Eij and Eii Meana E E E E E E E E

lc, le lc, gi lc,gk le, gi gi, gk le, gk lc, lc gi, gi

0.3497 0.0411 0.7000 0.0051 0.1006 0.1404 0.7098 0.0784

Meana

t-Statistic **

7.5443 0.9777 2.3167** 0.1292 2.6778** 4.7940** 8.3117** 1.1577

E E E E E E E E

le, lc gi, lc gk, lc gi, le gk, gi gk, le le, le gk, gk

t-Statistic

0.6212 4.0677** 0.0789 0.9052 0.1666 3.9478** 0.0055 0.1312 0.1247 3.1301** 0.1880 4.1426** 0.5312 9.7235** 0.3110 33.7855**

Note: We have tested the significance of the indices using the Wald test. * Statistically significant at 10%. ** Statistically significant at 5%. a Evaluated at the means of the data using parameter estimates of (8)–(9).

LE and LC is higher than the one between LC and LE. This could be due to the fact that operators always have to include at least one LE port worker per shift. On the other hand, as we also expected, the cross-elasticities shadow price between port

Note: We have tested the significance of the indices using the Wald test. * Statistically significant at 10%. ** Statistically significant at 5%. a Evaluated at the means of the data using parameter estimates of (8)–(9).

labor (LC or LE) and capital show that these factors are net complementaries. It happens exactly the same way between capital and GI. Finally, our estimation of cross-elasticities shadow prices between labor (LC or LE) and GI are not statistically significant. 5.2.4. Technical change The time year dummy coefficients show the effect on the distance function of unobserved variables which, when evolving over time, affect all firms equally. We can check how these time effects influence the distance function from one year to another through the following expression: TCTþ1;T ¼ gTþ1  gT .

(11)

ARTICLE IN PRESS 160

A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161

A positive (negative) value for TC indicates an upward (downward) shift in the distance function (see Fa¨re and Grosskopf, 1995). This measure is usually associated with technological change. The indices obtained from expression (10) are presented in Table 6. There were only two periods where these indices were statistically significant, and therefore, where time had an influence on firm activity. The first one, from 1991 to 1992 where the indices evolved favorably, and the second one, from 1995 until 1997, where the indices had a negative sign, which indicates that time had a negative influence on firm activity. However, it can be observed that in the last period (1998–1999) the coefficient returns to being positive, but not statistically significant. 6. Conclusions In this paper, we present an approach which allows us to estimate time-varying efficiency levels for individual firms without invoking strong distributional assumptions for inefficiency or random noise. Using a Spanish ports panel, we have applied this methodology to a frontier input distance system. In this way, the operations of cargo handling firms in ports are analyzed by means of a multioutput input distance function estimation using monthly data on firms located at Las Palmas Port, Spain. Both size and traffic mix are firstly shown to be sufficiently diverse as to allow for a reliable estimation of a multioutput translog function which permitted us the calculation of firm and timevariant indices of technical and allocative inefficiency (by using both the parametric and the error component approaches) which further add to the contribution of the paper to the literature. Implementing our approach with typical medium size port terminals data we highlight two main conclusions. The first one is that, it seems there is a relationship between firm size and technical efficiency. The second one is that, our result with respect to allocative efficiency suggests that the port labor-specific regulatory environment impedes adjustments that are needed by the operators. Finally, we have calculated Morishima elasticities. References Atkinson, S., Cornwell, C., 1994. Parametric estimation of technical and allocative inefficiency with panel data. International Economic Review 35, 231–243.

Atkinson, S., Primont, D., 2002. Stochastic estimation of firm technology, inefficiency, and productivity growth using shadow cost and distance functions. Journal of Econometrics 108, 203–225. Ban˜os Pino, J., Coto Milla´n, P., Rodrı´ guez-A´lvarez, A., 1999. Allocative efficiency and over-capitalization: an application. International Journal of Transport Economics 26 (2), 181–199. Battese, G., Coelli, T., 1992. Frontier production functions, technical efficiency and panel data: with application to paddy farmers in india. Journal of Productivity Analysis 3, 153–169. Battese, G., Coelli, T., 1995. A model for technical inefficiency effects in a stochastic frontier production function for panel data. Empirical Economics 20, 325–332. Bonilla, M., Medal, A., Casasu´s, T., Sala, R., 2002. The traffic in spanish ports: an efficiency analysis. International Journal of Economics Transport 29 (2), 215–230. Coelli, T., Prasada Rao, D.S., Battese, G.E., 1999. An introduction to efficiency and productivity analysis. Kluwer Academic Publisher, USA. Cornwell, C., Schmidt, P., Sickles, R.C., 1990. Production frontiers with cross-sectional and time series variation in efficiency levels. Journal of Econometrics 46, 185–200. Coto Milla´n, P., Ban˜os Pino, J., Rodrı´ guez-A´lvarez, A., 2000. Economic efficiency in spanish ports: some empirical evidence. Maritime Policy and Management 27 (2), 169–174. Cuesta, R.A., 2000. A production model with firm-specific temporal variation in technical efficiency: with application to spanish dairy farms. Journal of Productivity Analysis 13, 139–158. Cullinane, K., Song, D.W., Gray, R., 2002. A stochastic frontier model of the efficiency of major container terminals in Asia: assessing the influence of administrative and ownership structures. Transportation Research, Part A 36, 743–762. Cullinane, K., Song, D.W., 2003. A stochastic frontier model of the productive efficiency of korean container terminals. Applied Economics 35, 251–267. Cullinane, K., Wang, T.-F., Song, D.W., Ji, P., 2005. A comparative analysis of DEA and SFA approaches to estimating the technical efficiency of container port. Transportation Research A: Policy and Practice 40 (4), 354–374. De Rus, G., Roma´n, C., Trujillo, L., 1994. Actividad econo´mica y estructura de costes del Puerto de La Luz y de Las Palmas. Ed. Cı´ vitas. Madrid. Espan˜a. Dı´ az, J.J., 2003. Descomposicio´n de la productividad, la eficiencia y el cambio te´cnico a trave´s de la funcio´n de costes cuadra´tica. Una aplicacio´n a la operacio´n de estiba en Espan˜a. Ph.D. Thesis, Universidad de La Laguna. Estache, A., Gonza´lez, M., Trujillo, L., 2002. Efficiency gains from port reform and the potencial for Yardstick competition: lessons from Me´xico. World Development 30 (4), 545–560. Estache, A., Tovar, B., Trujillo, L., 2004. Sources of efficiency gains in port reform. A DEA decomposition of a Malmquist TFP index for Mexico. Utility Policy 12 (4), 221–230. Fa¨re, R., Grosskopf, S., 1990. A distance function approach to price efficiency. Journal of Public Economics 43, 123–126. Fa¨re, R., Grosskopf, S., 1995. Nonparametric tests of regularity, farrell efficiency, and Goodness-of-fit. Journal of Econometrics 69, 415–425. Good, D.H., Ro¨ller, L.H., Sickles, R.C., 1995. Airline efficiency differences between Europe and the US: implications for the

ARTICLE IN PRESS A. Rodrı´guez-A´lvarez et al. / Int. J. Production Economics 109 (2007) 149–161 pace of EC integration and domestic regulation. European Journal of Operational Research 80 (3 February 2), 508–518. Heaver, T., 1995. The implications of increased competition among ports for port policy and management. Maritime Policy and Management 22, 125–133. Heshmati, A., Kumbhakar, S.C., 1994. Farm heterogeneity and technical efficiency: some results from swedish dairy farms. Journal of Productivity Analysis 5 (1), 45–61. Heshmati, A., Kumbhakar, S.C., Hjalmarsson, L., 1995. Efficiency of the Swedish pork industry: a farm level study using rotating panel data, 1976–1988. European Journal of Operational Research 80 (3), 519–533. Jara-Dı´ az, S., Tovar, B., Trujillo, L., 2005. Marginal costs, scale and scope for cargo handling firms in Spain. Transportation 32, 275–291. Kumbhakar, S.C., 1990. Production frontiers, panel data and time-varying technical inefficiency. Journal of Econometrics 46, 210–211. Lee, Y.H., Schmidt, P., 1993. A production frontier model with flexible temporal variation in technical efficiency. In: Fried, H., Lovell, C.A.K., Schmidt, S. (Eds.), The Measurement of Productive Efficiency: Techniques and Applications. Oxford University Press, New York. Liu, Z., 1995. The comparative performance of public and private enterprise. The case of British ports. Journal of Transport Economics and Policy, 29 (3), 263–274. Lovell, C.A.K., 1996. Applying efficiency measurement techniques to the measurement of productivity change. Journal of Productivity Analysis 7, 329–340. Martı´ n, M., 2002. El sistema portuario espan˜ol: regulacio´n, entorno competitivo y resultados. Una aplicacio´n del ana´lisis envolvente de datos. Ph.D. Thesis, Universitat Rovira I Virgili. Martı´ nez Budrı´ a, E., Dı´ az Armas, R., Navarro Iban˜ez, M., Ravelo Mesa, T., 1999. A study of the efficiency of Spanish port authorities using data envelopment analysis. International Journal of Transport Economics XXVI (2), 237–253. Notteboom, T.E., Coeck, C., Van den Broeck, J., 2000. Measuring and explaining relative efficiency of container terminals by means of bayesian stochastic frontier models. International Journal of Maritime Economics 2 (2), 83–106.

161

Pestana, C., 2003. Incentive regulation and efficiency of Portuguese port authorities. Maritime Economics and Logistics 5, 55–69. Rodrı´ guez-A´lvarez, A., Lovell, C.A.K., 2004. Excess capacity and expense preference behaviour in National Health systems: an application to the Spanish public hospitals. Health Economics 13 (2), 157–169. Rodrı´ guez-A´lvarez, A., Ferna´ndez-Blanco, V., Lovell, C.A.K., 2004. Allocative inefficiency and its cost: the case of Spanish public hospitals. International Journal of Production Economics 92 (2), 99–111. Roll, Y., Hayuth, Y., 1993. Port performance comparison applying data envelopment analysis (DEA). Maritime Policy and Management 20 (2), 153–161. Song, D.-W., Cullinane, K., Roe, M., 2001. The productive efficiency of container terminals: an application to Korea and the UK. Ashgate. Tongzon, J.L., 2001. Efficiency measurement of selected australian and other International Ports using data envelopment analysis. Transportation Research, Part A 35, 113–128. Tongzon, J.L., Heng, W., 2005. Port privatization, efficiency and competitiveness: some empirical evidence from container ports (terminals). Transportation Research, Part A 39, 404–424. Tovar, B., 2002. Ana´lisis multiproductivo de los costes de manipulacio´n de mercancı´ as en terminales portuarias. El Puerto de La Luz y de Las Palmas. (A multioutput cost analysis for cargo handling services in port terminals. La Luz y de Las Palmas’ Port) Ph.D. Departamento de Ana´lisis Econo´mico Aplicado. Universidad de Las Palmas de Gran Canaria. Espan˜a (available on: /http://www.eumed.net/tesis/ btf/index.htmS). Tovar, B., Trujillo, L., Jara-Dı´ az, S., 2004. Organization and regulation of the port industry: Europe and Spain. In: CotoMillan, P. (Ed.), Essays on Microeconomics and Industrial Organisation, second Ed. Physica-Verlag. A springer-Verlag Company, Germany. Valentine, V.F., Gray, R., 2001. The measurement of port efficiency using data envelopment analysis. Ninth World Conference on Transport Research, Seoul, Korea.