Estimating Efficiency under Alternative DEA Models

With no prior empirical evidence on scale properties of container-port production, we use alternative DEA models to examine the effects of model choice on efficiency estimates. DEA-CCR and DEA-BCC models have been chosen to analyse terminal efficiency under constant and variable returns to scale, respectively. We also use both output and input orientations despite the latter being the selected orientation in the context of this research. Appendices 14 to 21 report the estimates of technical and scale efficiencies for different DEA models and type of data used.

For the DEA panel analysis (inter-temporal DEA), the results show that 44 DMU-years out of 420 in the sample are identified as efficient (efficiency score of 1 or 100%) under the CCR model compared with 93 units identified as efficient under the DEA-BCC model. For the DEA cross-sectional analysis (contemporaneous DEA), the results show that a total of 63 and 161 terminals, all years included, are identified as efficient when the DEA-CCR and the DEA-BCC models are applied, respectively. These results confirm that while the same set of DMUs are identified as efficient under both input and output orientations, the DEA-CCR models are more restrictive and yield lower efficiency scores than the DEA-BCC models, with respective average efficiency scores of 67% and 78.3% in the inter-temporal (input-oriented) analysis and 65.1% and 90.8%

in the contemporaneous (input-oriented) analysis. The Spearman's rank order correlation coefficient between the efficiency rankings derived from DEA-CCR and DEA-BCC analyses is 0.67 and 0.92 when input and output orientations are applied, respectively. This indicates that the efficiency estimates yielded by the two approaches follow the same pattern across sampled terminals.

Despite the general trend of relatively high operational efficiency, some terminals depict extremely low efficiency scores. JNCT-2000 has scored the lowest efficiency rating in the sample, with a value of 0.068 in both the CCR-I contemporaneous and DEA-CCR-I inter-temporal analyses. In addition to JNCT, 29 DMUs have scored lower than 30% efficiency rating in the DEA-CCR-I contemporaneous model and 19 DMUs in the CCR-I inter-temporal model. Of these low scores, twelve (12) have been recorded in the first year of the study (2000) under the CCR-I contemporaneous model against nine (9) in the CCR-I inter-temporal model. Further investigations show that the latter 9 terminals (MDCT, TOCT, NP, JNCT, MPE, TT, ASCT, SACT, and CCT) have either started operations in the year 2000 or undergone extensive expansion in that year.

Other noticeable cases include CT3, which has experienced a significant drop in its efficiency in 2005 due to a period of slow activity following the transfer of ownership from CSX World Terminals to DP World (CT3 efficiency scores in 2005 are 32.8% in the CCR-I contemporaneous model and 17.9% in the CCR-I inter-temporal model).

Such findings support the argument that DEA and other benchmarking techniques tend to favour small or fully ‘utilised’ terminals against newly operated terminals and those expanding or investing in new facilities. Further discussion on the impact of incremental investment on container terminal efficiency is presented in subsequent sections.

To confirm that the above terminal DMUs are mere outliers and are not likely to affect the general results, we run a sensitivity analysis through excluding these DMUs from the sample. An outlier is an observation that does not follow the general behaviour of the analysed units but can cause significant problems especially in extreme point methods such as DEA. The results of the sensitivity analysis show no major change in average efficiency estimates or in the rankings of DMUs in the sample, which indicates that the above outliers have no influence on the position or stability of the frontier.

Turning to the comparison of efficiency estimates yielded from alternative DEA models, Figure 22 depicts the year-by-year evolution of average terminal efficiency under both contemporaneous and inter-temporal analyses. It shows a general upward trend for average efficiency estimates until 2003, followed by an almost flat trend in 2004, a sharp downward trend in 2005, and a return to the ascendant trend in 2006.

Since most security measures have been introduced in late 2004, the results from Figure 22 may suggest a possible negative impact of procedural security on port efficiency, but a definitive conclusion requires the estimation of a TFP index for assessing productivity change before and after the implementation of the new security regulations.

Figure 22: Year-by-year (2000-06) evolution of average terminal efficiency (Based on input-oriented efficiency ratings)

Figure 23 shows the relationship between mean terminal efficiency scores and their standard deviations and indicates low negative correlation coefficients of

for DEA-CCR contemporaneous analysis, for DEA-BCC contemporaneous analysis, for DEA-CCR-I inter-temporal analysis, and for DEA-BCC-I inter-temporal analysis. A two-sided test of significance reveals that the correlation coefficients are statistically significant at the 5% confidence level, implying that the efficiency of container terminals in the sample does not exhibit similar levels of variation over time. This means that the more efficient terminals tend to have less relative variability over time compared with the less efficient terminals. These findings are in contrast with the results of previous port literature (e.g. Valentine and Gray, 2001;

Song et al., 2003; Cullinane et al., 2001) which have found similar levels of fluctuation over time between the efficiency of sampled terminals irrespective of their level of average efficiency. This may be due to the sampling procedure used in most port benchmarking studies where DMUs are usually selected from top-ranked ports in terms of throughput or from ports located within the same country or region.

Figure 23: Relationship between mean efficiency and standard deviation (Input-oriented efficiency ratings)