Analysis of technical efficiency of intensive white-leg

IIFET 2016 Scotland Conference Proceedings

1

Analysis of technical efficiency of intensive white-leg shrimp farming in Ninh Thuan, Vietnam: An

application of the double-bootstrap data envelopment analysis

LE, VAN THAP

The Faculty of Economics, Nha Trang University, Vietnam, [email protected], [email protected]

LE, KIM LONG

The Faculty of Economics, Nha Trang University, Vietnam, [email protected]

NGUYEN, TRONG HOAI

The School of Economics, University of Economics, HCM City, Vietnam, [email protected]

ABSTRACT

The paper adopts the double-bootstrap data envelopment analysis developed by Simar and Wilson in 2007

to estimate and explain technical efficiency of intensive white-leg shrimp farming in Ninh Thuan, Vietnam. This

procedure not only produces confidence intervals for technical efficiency but also enables consistent inferences

for factors explaining technical efficiency. From a policy perspective, our results show that there is considerable

room for improvement in technical efficiency in the case of intensive white-leg shrimp farming in Ninh Thuan,

Vietnam. The potential improvement is certainly greater than that using the deterministic data envelopment

analysis, which has been adopted widely in the aquaculture literature for technical efficiency estimation.

Statistically positive influence on technical efficiency is larger farm size, with cultural length and financial

stress being negative influences.

Key words: bootstrap, DEA, technical efficiency, intensive shrimp farming, Vietnam

INTRODUCTION

Vietnam has a long coastline of 3,260 km with coastal area of over 1.5 million hectare and an Exclusive

Economic Zone of more than 1million km2. All along the coastal line, there are many river mouths, lagoons and

bays with abundant aquatic resources which constitute a variety of ecological conditions highly suitable for

aquaculture development, particularly for various types of brackish water and shrimp culture. These natural

features give Vietnam great potential for developing aquaculture using inland water bodies, coastal brackish

water, and marine water areas.

In Vietnam, the fisheries industry is considered a key sector. Approximately 3.4 million people, or

approximately 10% of the labour force, are involved in this sector. One-tenth of export earnings from Vietnam

stem from fisheries products, including aquaculture (see also Long et al. 2008). In 2010, fishery production,

including catching and aquaculture, reached 5.2 million tons, creating more than USD 5.0 billion of export

revenue. Much of the growth in production can be attributed to continued expansion in aquaculture, which

increased from a 30% share of the sector in 1990 to 52% in 2010 (Fisheries Directorate of Vietnam, 2012).

Shrimp farming in Vietnam started to develop from the beginning of 1990s. The area of shrimp production

increased from 93,000 ha in 1990 to 235,000 ha in 2000, with an output of 38,000 metric tons and 103,000

metric tons respectively. In June 2000, Resolution No. 09/NQ-CP was launched by the government, allowing

farmers to convert low productive and saline rice fields, uncultivated areas, and salt pans in coastal areas into

aquaculture ponds. The shrimp farming area has, therefore, grown dramatically and reached 644,000 ha with

463,000 metric tons of shrimp output in 2010 (Fisheries Directorate of Vietnam, 2012; Tran et al., 2013).

Since 2001, white-leg shrimp farming in brackish water has become a major part of coastal economic

development in Vietnam. White-leg shrimp aquaculture, mainly in the central provinces of Vietnam and the

Mekong river delta, grew from 13,455 ha in 2005 to 22,192 ha in 2010 (Fisheries Directorate of Vietnam, 2012).

By 2012, the area of white-leg shrimp aquaculture was 38,169 ha with 177,817 tons of shrimp output.

Specifically, according to the Vietnam Association of Seafood Exporters and Producers (VASEP, 2013), with

only 5.9% of the total aquaculture area, white-leg shrimp occupied 27.3% of the shrimp industry output in 2012.

Due to this rapid development, with the side effects of chemicals, over-feeding, and intensive farming,

white-leg shrimp production has faced several severe risks, especially the spread of shrimp disease and

pollution. Although the potential for high and quick returns on investments makes the shrimp aquaculture

industry very attractive to national leaders and private sector entrepreneurs, production is very unstable (Lebel et

al., 2002). According to Sinh (2006), 30% of shrimp farms in the Mekong river delta experience financial

losses. Den et al., (2007), researching shrimp farms in Bac Lieu province, found that less technically efficient

farms are more likely to experience financial losses. With many shrimp farms in Khanh Hoa, the central area of

Vietnam, experience financial losses, Long and Binh (2013) concluded that white-leg shrimp farms face

mailto:[email protected]





2

especially high risks. Moreover, although the area of white-leg shrimp farming increased by 15.5% from 2011 to

2012, its output has risen by only 3.2% (VASEP, 2013). Therefore, the rapid increase in the number and

capacity of white-leg shrimp farms has raised the issue of whether the resources used by shrimp farming are

excessive and inefficient. However, studies on the input utilization of shrimp farming in Vietnam are still very

limited (Nguyen & Fisher, 2014).

Ninh Thuan is the most southern province in central Vietnam with the potential area of about 1,000 ha for

shrimp farming. The geographic, climatic and hydrological conditions (temperature, rainfall, tidal regime) are

considered, in general, favorable for shrimp aquaculture. The farming of tiger shrimp in Ninh Thuan has been

promoted since the 1990s with a focus on semi-intensive cultivation and an annual crop. The production of

white-leg shrimp farming in Ninh Thuan started in 2001. More recently, new technologies have been adopted as

part of the process of moving towards more intensive farming methods. Nevertheless, the area for white-leg

shrimp farming soared by about 6.2 times from 159 ha in 2006 to 984ha in 2011. Similarly, its output rose from

1,350 tons in 2006 to 7,342 tons in 2011. In 2012, Ninh Thuan had about 850 ha engaged in intensive white-leg

shrimp farming (Fisheries Directorate of Ninh Thuan, 2013). Currently, the intensive white-leg shrimp farming

is considered a key sector in Ninh Thuan province. Approximately 6 thousand people, or approximately 2% of

the labour force, are involved in this sector. The value of aquaculture production, mainly from white-leg shrimp

sector, is about 3,000 billion Vietnamese Dong (150 million USD), occupied 15% of Ninh Thuan’ GDP in 2011

(GSO, 2014). On this basis, it is of importance for policy-makers, farmers and other industry representatives to

know if resources such as land, man-made inputs and human capital are used in an efficient way.

In spite of the current successes witnessed by the white-leg shrimp farming, many challenges are

continuing to set back the growth of this aquaculture in Ninh Thuan. Firstly, the area for shrimp farming has

almost reached the maximum limit of 1000 ha in 2012. Secondly, the shrimp harvested per ha has decreased

over the period of 2010 – 2012. Moreover, the spread of shrimp disease and pollution from sewage ponds have

created challenges for policy makers and threaten the sustainable development of shrimp aquaculture. In

addition, a lack of technical knowledge, a low educational level, inexperienced managers/farmers, financial

stress, a lack of market information are some of the factors that are hindering the development of the shrimp

farming (Fisheries Directorate of Ninh Thuan, 2013). Therefore, research on the efficiency of intensive white-

leg shrimp farms and its determinants is vital for sound policy recommendations and the redirection of

managerial actions for the sustainable development of shrimp aquaculture in Ninh Thuan, and also Vietnam

more broadly.

For several decades, efficiency measurement in production has received particular attention with the aim of

accelerating economic development, especially in developing countries. Economic efficiency, especially inter-

farm differences in efficiency, is one of the major factors explaining differences in white-leg shrimp farm

survival and growth. Thus, factors explaining and determining differences in economic efficiency and changes

in efficiency between white-leg shrimp farms are of major interest to owners and policy makers as they strive to

improve earnings and improve the chances of farm survival. This study is undertaken to improve understanding

of the inter-farm differences in and opportunities to improve white-leg shrimp farming efficiency in utilizing

limited input resources to achieve farms’ objectives in Ninh Thuan province.

The two most popular techniques employed in efficiency measurement are: (i) data envelopment analysis

(DEA), a nonparametric technique adopting a linear programming approach; and (ii) stochastic frontier analysis

(SFA), a parametric technique using of an econometric method. The advantage of DEA over SFA is that DEA is

nonparametric and does not require any parametric assumptions on the structure of production technology.

However, the deterministic DEA approach does not have a solid statistical foundation behind it and is sensitive

to outliers (Coelli et al., 2005). Both DEA and SFA have been used to study the efficiency of aquaculture farms

in different countries (for reviews, see Sharma & Leung, 2003 and Iliyasu et al., 2014). The application of the

SFA approach, however, predominates in efficiency studies on aquaculture, probably due to the stochastic

nature of aquatic culture (Iliyasu et al., 2014).

To overcome these drawbacks of the deterministic DEA approach, Simar and Wilson (2000) have

introduced bootstrapping into the DEA framework. Their method, based on statistical well-defined models,

allows for consistent estimation of the production frontier, corresponding technical efficiencies, as well as

standard errors and confidence intervals. Moreover, Simar and Wilson (2007) proposed a procedure based on a

double bootstrapping technique, which provides both confidence intervals for efficiency estimates and

consistent inferences for factors explaining efficiency. Algorithm 2 of the double bootstrap DEA developed by

Simar and Wilson in 2007 has been applied empirically to several agriculture studies. For example, Latruffe et

al., (2008) used this method to estimate the technical efficiency of crop and livestock farms in the Czech

Republic. Balcombe et al., (2008) applied it to examine the sources of efficiency in Bangladesh rice farming.

The method was also used in Olson and Vu (2009) for agriculture farms in Minnesota, U.S.A. Keramidou and

Mimis (2011) also adopted double bootstrap DEA to investigate the sources of efficiency in the Greek poultry

sector.


3

Despite this development, the application of the DEA bootstrapping technique has thus far been limited in

analyzing the efficiency of aquaculture. There are only the works of Chang et al. (2010) and Iliyasu et al. (2016)

used this technique for estimating biased corrected technical efficiency in aquaculture. Most other studies have

employed the deterministic DEA model to analyze the technical efficiency of aquaculture (Iliyasu et al., 2016).

This, therefore, motivates the use of Algorithm 2 of the double bootstrapping DEA developed by Simar and

Wilson (2007) for both biased corrected technical efficiency estimates and consistent inferences for factors

explaining efficiency in this study.

This paper extends previous aquaculture studies by adopting the double bootstrap DEA model in

comparison with the deterministic DEA approach to analyze efficiency in the case of white-leg shrimp farming

in Ninh Thuan, Vietnam. The specific objectives of this study are as follows: (i) to use the double bootstrap

DEA procedure to correct the bias generated by the deterministic DEA method; (ii) to identify factors that are

significant in explaining differences in levels of technical efficiency between shrimp farms.

METHODOLOGY

Study area and data collection

Ninh Thuan is the province in central Vietnam. It consists of 6 districts, Phan Rang, Bac Ai, Ninh Son,

Ninh Hai, Ninh Phuoc, Thuan Bac and Thuan Nam. The province has a total area of 3,360 km2 and a 105 km

coastline, along which are quite a number of lagoons and river mouths that facilitate the farming of aquaculture

products. As the province with the lowest annual rainfall (average 750 mm), few waterways and seawater of a

stable but high salinity level, Ninh Thuan enjoys several advantages in developing aquaculture products of high

economic value. The intensive white-leg shrimp aquaculture is mainly found in Ninh Hai and Thuan Nam

districts (Fisheries Directorate of Ninh Thuan, 2013).

In the present study, a total of two districts, Ninh Hai and Thuan Nam, were purposely selected due to their

high concentration of intensive shrimp farmers. Accordingly, these two districts are believed to represent the

province fairly well. From the population of 442 registered intensive white-leg shrimp households, a sample of

110 farmers (about 25%) were randomly selected from the list of farmers obtained from the Fisheries

Directorate of Ninh Thuan. Each farming household, represented by the owner and/or manager, gave a face-to-

face interview to two research team members from the Faculties of Economics and Aquaculture, Nha Trang

University in early 2005. Data were collected using a project-designed questionnaire enquiring about the

relevant socioeconomic characteristics, farm-specific information, outputs produced, and inputs used for the

operating year of 2014. A pilot study was first conducted to validate the questionnaire and the necessary

adjustments and changes were made. A total of 110 questionnaires were administered but only 102 were used in

the analysis due to incomplete responses.

The empirical model

DEA is essentially a linear programming technique used for measuring the relative performance of farms

for which the presence of multiple inputs and outputs makes comparisons difficult. It involves the identification

of farms that in a relative sense operate in the most optimal manner, and then uses this information to construct

an efficiency frontier over the available data set. DEA uses this efficient frontier to calculate the efficiency of

the other farms that do not fall on the efficient frontier (Coelli et al., 2005).

Technical efficiency relates to the degree to which a farm produces the maximum feasible output from a

given bundle of inputs, or uses the minimum feasible amount of inputs to produce a given level of output. These

two definitions of technical efficiency lead to what are known as output-oriented and input-oriented efficiency

measures respectively. Orientation (input or output) is selected according to which factors farmers have most

control over (see also Banker et al., 1984 and Nielsen, 2011). In this study, an input-oriented DEA model was

used. The input orientation approach to DEA for efficiency analysis has also been chosen by some aquaculture

studies for developing countries, such as those of Alam and Jahan (2008), Alam (2011), and Nguyen and Fisher

(2014). The orientation aims to reduce the input amount by as much as possible while keeping at least the

present output level. This is a sensible assumption for white-leg shrimp farmers in developing countries, like

Vietnam, where saving costs in production is very important due to their limited finances.

Next, the CRS DEA model is only appropriate when the farm is operating at an optimal scale (see also

Coelli et al., 2015 and Nielsen, 2011). Alam (2011) suggested that aquaculture operations, unlike industry,

follow law of variable proportions. Factors such as imperfect competition, constraints on finance, socio-

economic limitations of the Ninh Thuan’s shrimp farmers may cause the farm to not operate at an optimal scale

in practice. The VRS DEA model for production technology is adopted in this study for the case of the intensive

white-leg shrimp farming in Ninh Thuan, Vietnam.


4

Table 1. Description of the variables in the models

Variables Description Unit

Production model for DEA framework

Output (Y) Total quantity of shrimp produced per ha per year Kilogram

Input (X)

Seed (X1) Fingerlings stocked in the farm per ha per year 1000ind

Feed (X2) Total quantity of feed used per ha per year Kilogram

Labor (X3) Total number of man-hours per ha per year Hour

Chemicals (X4) Total amount of fertilizer/chemical applied per ha per year Kilogram

Power (X5) Total Kw of electricity per ha per year Kw

Farm specific variables

Farm size (Z1) Total area for shrimp aquaculture of the farm Hectare

Financial stress (Z2) Borrowing for production cost (1 = yes, 0 = otherwise) Dummy

Culture length (Z3) The length of shrimp farming per year Day

Experience (Z4) Years the shrimp farmer/manager spent in shrimp farming Year

Training (Z5) Technical training from extension agents (1 = yes, 0 = otherwise) Dummy

Education (Z6) Level of education of shrimp farmer/manager

(1 = college or higher, 0 = otherwise)

Dummy

The outputs and inputs for DEA framework in this study were selected based on informed choice from the

existing literature (see Iliyasu et al., 2014; Nguyen and Fisher, 2014). The production technology has, therefore,

been considered to be comprising of one output and five main inputs. Five most important inputs include labor,

seed, feed, fertilizer/chemical and power which were assumed to adequately represent the white-leg shrimp

production technology in the sample areas (see also Nguyen and Fisher, 2014 for shrimp farming in Mekong

river delta, Vietnam). The output is harvested shrimp in kilogram (see Table 1 for details). Therefore, the inputs

and output relationships have been examined using the VRS DEA framework as follows.

Consider a farm with five inputs X (labor, seed, feed, fertilizer/chemical and power) and output Y (shrimp

harvested). For the jth farm out of 102 farms, the input-based technical efficiency index for DEA framework,

TEj, is defined as:

such that:

(Eq. 1)

where TEj = θj is the technical efficiency value ranging from 0 to 1, and λ is a vector of constant (weights) that

defines the linear combination of the peers of the jth farm. The first constraint is with respect to the output of

white-leg shrimp farming. The term Yj , tons of shrimp harvested, on the left-hand side of the constraint is the

vector of output of the jth farmer compared to the output vector of the theoretically efficient farmer (Yλ). This

constraint implies that the theoretically efficient shrimp farm produces an amount of output that is greater than

or equal to the actual output produced by the jth farm given the same amount of inputs.

The second constraint in (1) concerns the input of white-leg shrimp farming. Five main inputs namely,

seed (1000ind), labor (man-hour), feed (kg), fertilizer/chemical (kg) and power (Kw of electricity) are

incorporated into the VRS DEA model in (1). The term θjXj on the left-hand side of the constraint represents the

actual level of input of the jth farm multiplied by its level of efficiency (θj). Next, the term Xλ represents the

minimum quantity of input of the theoretically efficient shrimp farm uses, given the actual level of output

produced by the jth farm. If the solution in (1) is smaller than one, then the level of input used by that particular

shrimp farm can be reduced further to as low as Xλ to produce the same level of output. It means that there is a

degree of technical inefficiency in that particular shrimp farm. On the other hand, if the solution in (1) turns out

to be θj = 1, then the level of input of that particular shrimp farm is as small as the quantity of input used by the

theoretically efficient farm in producing the same level of output. The shrimp farm is, therefore, technically

efficient.

The third constraint in (1) is the convexity constraint, =1, for assuming variable returns to scale.

The above VRS DEA model gives us the result of TE_VRS. Next, we can easily have CRS DEA model by

relaxing the convex constraint in (1). Similarly, the CRS DEA model gives us the result of TE_CRS. The scale

efficiency (SE) of a white-leg shrimp farm can be calculated by dividing the technical efficiency estimates under

CRS to the technical efficiency estimates under VRS. It means that SE = TE_CRS/TE_VRS. In general, 0 ≤ SE

≤ 1, with SE = 1, meaning that the farm is operating at an efficient economy of scale. SE < 1 implies that the

input utilized by the farm is not efficient in scale (see Coelli et al., 2005).


5

Apart from the technology used in aquaculture, individual farm characteristics may also have an impact on

technical efficiency estimated in the first stage DEA in (1). For example, farmers not attending training in

farming techniques are expected to perform worse than those who do (provided that they are operating under

otherwise similar conditions). Iliyasu et al. (2014) recognized that the majority of aquaculture studies use a

deterministic two-stage DEA approach, in which efficiency is estimated in the first stage, and then the estimated

efficiencies are regressed on the individual farm characteristics with Tobit regression in the second stage. This

approach is often called the deterministic two-stage DEA analysis.

As existing aquaculture studies undertaking efficiency analysis using deterministic two-stage DEA have

not considered its statistical properties, we need to be careful when attempting to draw broad policy implications

based on their findings, especially due to the stochastic nature of aquatic culture. Zhang and Bartels (1998)

claimed that the deterministic DEA estimation of technical efficiency is sample specific. This is then

compounded by the fact that sample size and the number of variables included in model specifications impact

the estimates. Moreover, Simar and Wilson (2000) proposed that the deterministic DEA method will yield

sample estimates of efficiency that will positively exaggerate the level of efficiency within a sample of data.

Simar and Wilson (2007) criticized the deterministic two-stage DEA approach because the DEA efficiency

estimates are biased and serially correlated, therefore invalidating conventional inferences in the second stage.

Simar and Wilson (2007) proposed Algorithm 2 with a double bootstrap procedure that gives confidence

intervals for the efficiency estimate and also enables consistent inference within models explaining technical

efficiency. Bootstrapping is a way of testing the reliability of a data set by creating a pseudo-replicate data set.

Random samples are obtained by sampling with replacements from the original data set, which provides an

estimator of the parameter of interest. The rationale behind bootstrapping is to simulate a true sampling

distribution by mimicking the data generating process (see Balcombe et al., 2008; and Olson & Vu in 2009 for

further discussion). Simar and Wilson (2011) emphasized that the double bootstrap methods developed in

Simar&Wilson (2007) provide the only feasible means for inference in the second stage regression. However,

there are not any studies so far that apply this double bootstrapping technique for aquaculture data.

This study adopts Algorithm 2 in Simar and Wilson (2007) which is described in the endnote.i To explain

differences in levels of technical efficiency between white-leg shrimp farming practices, the procedure proposed

by Simar and Wilson (2007) as follows:

Taking a transformation of the technical efficiency estimated in (1) , is the reciprocal of

technical efficiency value as dependent variable for the second-stage with truncated regression such as:

(Eq. 2)

where is, called the technical efficiency score (see also input efficiency score developed by Shephard (1970)

with a distance function), with the range from 1 to infinity and Zj is a vector of individual farm characteristics

assumed to affect the choice and use of output and inputs, is a vector of parameters to be estimated, and is a

continuous i.i.d. random variable, distributed as with left truncation at for each farm j, and

assumed to be independent of Zj. Adopting the truncated regression for (2), this procedure avoid boundary

problems since the estimated value of technical efficiencies are typically defined on the interval [0,1], with, in

general, few values, if any, close to 0 but some values of 1 (see Burgress and Wilson, 1998 for further

discussion). Johnson and Kuosmanen (2012) emphasized that Simar and Wilson (2007) advocated the use of the

truncated regression model that takes into account explicitly the bounded domain of the DEA efficiency

estimates.

Next, individual farm characteristics for white-leg shrimp farming practices in Ninh Thuan include the

farmers/managers’ experience, the length of a production circle, farm size, education of the farmers/managers,

training of the farmers/managers and financial stress (see Iliyasu et al., 2014 for a review of factors explaining

technical efficiency in aquaculture). These shrimp farm specific variables are descripted in details in Table 1.

Moreover, since the dependent variable in (2) is the reciprocal of technical efficiency, so-called the technical

efficiency score, a positive relationship between a shrimp farm specific variable and technical efficiency exists

if the sign of the coefficient is negative, and negative if the coefficient is positive (see also Balcombe et al.,

2008).

The truncated bias corrected maximum likelihood method is used to identify individual farm characteristics

that have significant explanatory effect on differences in technical efficiencies in (2) using the double bootstrap

estimation. The rDEA package created by Simm and Besstremyannaya in 2015 is employed in this study for the

double bootstrap DEA model with L1 = 100 interactions for the first loop and L2 = 2000 for the second loop of

Algorithm 2 (see endnote i for details). For comparison, the deterministic two-stage DEA analysis with Tobit

regression is also conducted for the white-leg shrimp farming in Ninh Thuan. We used the Benchmarking

package created by Bogetoft and Otto (2010) for the deterministic two-stage DEA approach. Both rDEA and

Benchmarking packages are linked to the R Project for Statistical Computing.


6

EMPIRICAL RESULTS AND DISCUSSION

Descriptive statistics of the data

Table 2 presents the summary statistics of the data collected in the study. The farm specific variables

provide a summary of the characteristics of these intensive white-leg shrimp farms in Ninh Thuan province. The

average experience of the sample shrimp farmers/managers is 11.31 years; 23 percent of shrimp

farmers/managers have the educational level of college or higher; 67 percent of farmers have participated in

extension services; 75 percent of farms had a loan on production cost. Average farm size is 0.85 ha ranging from

0.3 to 6 ha.

Table 2. Summary statistics of variables in the models

Variables Mean SD Min Max

Production model

Output (Y) 29930.69 11345.63 5260.00 50010.00

Input (X)

Seed (X1) 4323.73 1829.58 500.00 9900.00

Feed (X2) 46465.69 20814.28 6300.00 87300.00

Labor (X3) 7562.93 2908.29 2316.00 13714.00

Chemicals (X4) 105.56 55.55 9.50 321.3

Power (X5) 379926.54 231735.31 25320.00 932224.00

Farm specific variables

Farm size (Z1) 0.85 0.69 0.3 6

Financial stress (Z2) 0.75 0.43 0 1

Culture length (Z3) 224.80 61.18 55.0 330

Experience (Z4) 11.31 4.10 4.0 24

Training (Z5) 0.67 0.47 0 1

Education (Z6) 0.23 0.42 0 1

The output measured was shrimp in kilograms per ha per year. The mean shrimp output harvested is

29930.69 kg (min. 5260.00 kg; max. 50010.00 kg). All main production system inputs (five in total) were

measured. Human labor input is measured as the number of man-hours per ha for various activities, and it

includes all hired and family labor, assuming that 1 day consists of 8 h work. The mean number of hours spent

per ha per year is 7562.93 (min. 2316.00 hours; max. 13714.00 hours). Seed is measured as the physical

quantity of seeds in 1000 individuals per ha per year. There are on average 4323.73 post-larvae (i.e., individual

shrimp) per ha (min. 500.00; max. 9900.00). Feed is one of the most important components of aquaculture and

constitutes more than 50% of production cost. The amount of feed used is measured in kg. The average quantity

of feed used by the sample fish farmers is 46465.69 kg per ha per year. The amount of fertilizer/chemical input

is measured as kg applied and includes calcium carbonate, urea, and diammonium phosphate. The mean of

fertilizer/chemical input used for a year is 105.56 kg per ha. Finally, power devoted to shrimp production is

measured in kW of electrics per ha. The average number of electrics spent per year is 379926.54 kW.

Efficiency estimation

Table 3 presents estimated levels of efficiency for white-leg shrimp aquaculture in Ninh Thuan for both

deterministic and double bootstrapping approaches. Over all farms, the deterministic DEA estimate for average

technical efficiency is 0.79 (TE_VRS). This means that, on average, white-leg shrimp farming households can

reduce their inputs by 21% without changing the level of their output. Furthermore, the mean value of scale

efficiency of 0.87, calculated by TE_CRS/TE_VRS with the deterministic estimation, implies that there are a

number of shrimp farms demonstrating improper operational scale.

Table 3. Deterministic and double bootstrap DEA estimates

Description Deterministic Double bootstrap

TE_CRS TE_VRS TE_CRS TE_VRS

Mean 0.69 0.79 0.63 0.73

Median 0.70 0.80 0.63 0.75

Min 0.39 0.40 0.35 0.37

Max 1.00 1.00 0.88 0.91

Upper 95% CI for Mean - - 0.67 0.80

Lower 95% CI for Mean - - 0.59 0.68

From these initial estimates, we apply the double bootstrap in Simar and Wilson (2007) to correct for the

bias in technical efficiency. The results in Table 3 show that the biases in the uncorrected results are quite


7

considerable. Specifically, the deterministic TE_VRS estimate of 0.79 means that with a given output, an

average farm could reduce its input by 21% if the technical efficiency value were improved to 1. The bias-

corrected TE_VRS estimate of 0.73, however, suggests an expected input reduction of 27% for the double

bootstrap estimate. It is clear that the amount of input saving is considerable for the case of white-leg shrimp

aquaculture in Ninh Thuan. Interestingly, the mean value of scale efficiency of 0.86 calculated by

TE_CRS/TE_VRS with the double bootstrap is almost the same as the deterministic estimation. This means that

the value of scale efficiency may be not biased by the deterministic DEA.

Moreover, we also provide interval estimates of technical efficiency. The width of the 95% confidence

intervals is 0.12 for the mean of TE_VRS using the double bootstrap method. The lower and upper bounds of

the 95% confidence interval for the bias-corrected TE_VRS in the double bootstrap are 0.68 and 0.80. Clearly,

as a result of estimating the bias-corrected measures and interval estimates of technical efficiency, our results

can be viewed by policy makers with increased confidence.

Table 4. Mean comparison and correlations of efficiency rankings

Efficiency

Mean

t-ratio

Spearman rank

correlation ( )

Kruskal-Wallis

rank sum test Deterministic Double

bootstrap

TE_CRS 0.69 0.63 19.028*** 0.980*** 14.748***

TE_VRS 0.79 0.73 20.022*** 0.981*** 9.615*** ***, **, * Significant at 1%, 5% and 10% levels, respectively.

As the efficiencies estimated from the deterministic and double bootstrap DEA are not independent, we use

the paired difference t-test, the Kruskal–Wallis rank sum test, and Spearman’s rank correlation to compare the

estimates (see Bogetoft & Otto, 2010 for details). The results are presented in Table 4. Based on the paired t-

tests and the Kruskal–Wallis rank sum tests, on average, the efficiencies estimated from the deterministic DEA

– both TE_CRS and TE_VRS – are statistically significantly higher than in the double bootstrap method. This

finding supports the proposition of Simar and Wilson (2000). Latruffe et al. (2008) and Olson & Vu (2009) also

found the same relationship in their agricultural studies. Next, the Spearman correlation coefficients for the

efficiency rankings of the sample shrimp farms are also positive and statistically significant. The result implies

that the efficiencies calculated in both methods are not independent. In other words, the efficiency rankings of

intensive white-leg shrimp farms in Ninh Thuan are consistent in both estimation methods.

Factors associated with technical efficiency

From a policy perspective, the estimates of efficiency propose that there is considerable room for

improvement in technical efficiency as a means of making up the gap in yield between shrimp farms. Therefore,

it would appear sensible to examine the determinants of technical efficiency for the case of white-leg shrimp

farming in Ninh Thuan.

Table 5. Determinants of technical efficiency score(a): double bootstrap estimation

Variables Coefficients Lower 95% CI Upper 95% CI Lower 90% CI Upper 90% CI

Intercept 0.2923 -0.8719 1.1264 -0.6147 1.0051

Farm size -0.3149** -0.7216 -0.0521 -0.7098 -0.1053

Financial stress 0.3538** 0.0292 0.8300 0.0665 0.7075

Culture length 0.0030** 0.0007 0.0059 0.0010 0.0053

Experience 0.0174 -0.0174 0.0539 -0.0111 0.0482

Training -0.0757 -0.3440 0.1969 -0.3062 0.1572

Education -0.0089 -0.3822 0.3375 -0.3183 0.3041

**, * Significant at 1%, 5% and 10% levels, respectively. (a) Technical efficiency score is the reciprocal of technical efficiency value

We now turn to the results that attempt to explain the sources of technical efficiency. The truncated bias-

corrected maximum likelihood method is used to identify individual farm characteristics that have a significant

explanatory effect on the differences in efficiencies using the double bootstrap estimates. As mentioned earlier,

the dependent variable in the double bootstrap model is the technical efficiency score or the reciprocal of

technical efficiency value. Therefore, there is a positive relationship between an individual farm characteristic

and efficiency if the sign of the coefficient is negative, and a negative relationship if the coefficient is positive.


8

For comparison, the Tobit regression in the second stage of the deterministic DEA approach is also

conducted. To be consistent with the double bootstrap DEA method, the dependent variable in the Tobit

regression is also the reciprocal of the estimates of technical efficiency in (1), so the predicted impact is the

opposite of the signs for the estimated coefficients.

Table 6. Determinants of technical efficiency score(a): Tobit regressed estimation

Variables Coefficient Standard error T-value P-value

Intercept 0.9788** 0.1818 5.384 0.0000

Farm size -0.1960** 0.0774 -2.530 0.0114

Financial stress 0.1316* 0.0748 1.758 0.0787

Culture length 0.0015** 0.0005 2.778 0.0054

Experience 0.0053 0.0084 0.636 0.5245

Training -0.0523 0.0662 -0.789 0.4299

Education 0.0348 0.0835 0.417 0.6769 **, * Significant at 1%, 5% and 10% levels, respectively. (a) Technical efficiency score is the reciprocal of technical efficiency value

Our estimation reveals some interesting findings. First of all, the farm size has a significant positive impact

on technical efficiency in both double bootstrap and Tobit estimations within the 95% confidence interval. A

positive relationship between farm size and technical efficiency for Ninh Thuan’s intensive shrimp farming

indicates that the larger the farm size, the more efficiently the farmers utilize their inputs. More often than not,

large farms tended to reap the benefits of economies of scale (see also Iliyasu et al., 2014). This is in line with

the findings reported by some aquaculture studies such as Dey et al. (2005) for intensive freshwater pond poly-

culture production in Vietnam and Amos (2007) for crustacean production in Nigeria.

In the case of financial stress, we find a negative relationship with technical efficiency. This estimate is

statistically significant for the double bootstrap within the 95% confidence interval. This implies that the shrimp

farms with a loan for production costs are less technically efficient than others in Ninh Thuan. This relationship

is also found in Latruffe et al. (2008) for the case of agriculture farming in the Czech Republic, a country with

imperfect national capital markets similar to Vietnam. The negative impacts of financial stress on technical

efficiency can be explained by agency theory proposed by Jensen and Meckling (1976) for the case of the

intensive white-leg shrimp farming with high capital investment and operational expenses in Ninh Thuan.

Shrimp farm households with a loan for production costs from food suppliers, commercial banks and even

unofficial financial markets bear high costs for receiving credit with tight schedules for paying the loans off.

There are only 32% of households with a loan for production cost in our sample who can borrow money from

commercial banks because of (i) limited assets for mortgage and (ii) complicated borrowing procedures for a

high risky farming. Therefore, the scope of farmers’ management decisions is restricted and efficiency is

reduced. The Tobit regressed estimation also displays the expected coefficient for this variable, complying with

the double bootstrap result. However, the estimate in the Tobit regression is statistically at the 90% confidence

interval.

As expected, cultivation length in a year has a significant negative impact on technical efficiency in both

double bootstrap and Tobit estimations within the 95% confidence interval. This result for the case of intensive

shrimp farming in Ninh Thuan may be due to two reasons. Firstly, farmers who keep their shrimp longer than

necessary are more likely to use greater inputs with little or no gain in additional output, thus becoming

inefficient (see also Iliyasu et al., 2014). This relationship is also found in some other aquaculture studies, such

as Tan et al. (2011) and Alam (2011). Secondly, farmers with longer culture period (intensive use of land) may

face with the risks of the spread of shrimp disease and pollution. They are, therefore, more likely becoming

inefficient. Nguyen and Fisher (2014) also found that pollution has a negative effect on technical efficiency for

the case of shrimp farming in Mekong river delta, Vietnam.

Contrary to expectation, we see that experience has a negative impact on technical efficiency for both

estimations. This implies that the shrimp farming households with greater experience are more technically

inefficient than others in Ninh Thuan. This relationship is also found in some other aquaculture studies, such as

those of Onumah (2010) and Onumah et al. (2009) for the stochastic production frontier model and Nguyen and

Fisher (2014) for the deterministic DEA estimation. The negative impacts of experience on technical efficiency

can be explained by the leapfrogging hypothesis (Brezis et al., 1993). Inexperienced farmers may bypass

investment in outdated technology, and benefit from new and improved technology, thereby making their

operation more technically efficient. Experienced farmers may not find it beneficial to invest immediately in

new and improved technology (see also Iliyasu et al., 2014). However, the estimate is statistically very weak for

both models in the case of Ninh Thuan’s intensive shrimp aquaculture.

The fifth explanatory variable is technical training from extension agents. We can see that for both

estimations, the variable is positively related to higher levels of technical efficiency. Fish farmers who receive


9

technical training from extension agents may acquire more managerial skills and may therefore be more

technically efficient than those who rely only on their lengthy experience. This result is also found in Cinemre et

al. (2006). Interestingly, the estimate is statistically very weak for both estimations. This may be explained by

the majority of households in our sample (67%) who have attended extension training courses. However, the

quality of the extension training for shrimp aquaculture in Ninh Thuan should be re-assessed. Technical training

from extension is, even, found a negative but insignificant effect on technical efficiency, which is explained by

the inappropriate training for the case of intensive shrimp farming in the Mekong river delta of Vietnam

(Nguyen & Fisher, 2014).

Next, education is positively related to a farm’s technical efficiency in the double bootstrap estimation.

Educated shrimp farmers can be expected to comprehend information and have better managerial skills, thereby

making them more technically efficient. This result conforms to the finding obtained for the case of aquaculture

farms in India by Roy and Rens (2008). The Tobit regression for a deterministic estimation displays the

unexpected coefficient for this variable. However, this relationship in Ninh Thuan’s shrimp aquaculture is not

statistically significant for either model at the 90% confidence interval.

Overall, the comparison of the double bootstrap estimation in Table 5 with Tobit estimation using non bias

corrected scores in Table 6 shows only slight differences. However, stronger relationships result from the

application of Algorithm 2 of Simar and Wilson’s (2007) procedure. This result is also found in Latruffe et al.

(2008) for the case of agriculture farming in the Czech Republic.

CONCLUSIONS

This study employs the double bootstrap DEA model proposed by Simar and Wilson (2007) to determine

the variability in technical efficiency estimates and to correct for the bias inherent in the deterministic

measurement for the case of shrimp farming in Ninh Thuan, Vietnam. The bias-corrected point estimate of

technical efficiency is 0.73, and at the 95% confidence interval is estimated to be 0.68 at the lower limit and

0.80 at the upper limit. This result suggests that there is considerable room for improvement in technical

efficiency in the sample of farms analyzed. Moreover, the mean estimate of technical efficiency of 0.73 using

the double bootstrap approach is statistically significantly lower than that of 0.79 adopting the deterministic

DEA for the case of Ninh Thuan’s white-leg shrimp farming. The potential improvement in technical efficiency

in this study is therefore certainly greater than that using deterministic DEA, which has been adopted widely in

the aquaculture literature.

An improvement in technical efficiency among these white-leg shrimp farmers can help to reduce the gap

in yield between the most and the least efficient farmers. The analysis also reveals that the factors that could

enhance technical efficiency are education, extension training, farming using earthen ponds, and the increased

size of farms. The variables that are negatively related to technical efficiency, and hence hamper farm

performance, are financial stress, farmer experience and a longer cultivation period. From a methodological

point of view, the comparison of the double bootstrap estimation and the Tobit regressed estimation shows only

slight differences. Nevertheless, stronger relationships result from the application of Simar and Wilson’s (2007)

algorithm. The results reported here suggest that policies for credit and land are the leading constraints on

improved productivity in Ninh Thuan shrimp farming. To help farmers avoid keeping shrimp in ponds longer

than necessary, market information on shrimp output should be clear. Further development of intensive shrimp

farming in Ninh Thuan should be undertaken with care of risks of the spread of shrimp disease and pollution.

Improving the quality of technical training for white-leg shrimp farming in Ninh Thuan may also be considered.

Finally, attempts to alleviate these constraints may be more effective when targeted at less experienced and

better educated farmers, or perhaps family members.

Finally, our findings with the double bootstrap DEA do not differ substantively those of the deterministic

DEA analysis. Therefore, it is the contention here that the findings of previous studies, employing a

deterministic DEA two step approach, largely remain valid. However, it is advisable to use Algorithm 2 of the

Simar and Wilson (2007) double bootstrap procedure in further applied research on technical efficiency in

aquaculture studies where the random noise in production is significant, as this can increase the confidence that

policy makers can place in the results generated.

REFERENCES

Alam, F. 2011. Measuring technical, allocative and cost efficiency of pangas (Pangasius hypophthalmus:

Sauvage 1878) fish farmers of Bangladesh. Aquaculture Research, 42(10), 1487–1500.

Alam, F., & Jahan, M. 2008. Resource allocation efficiency of the prawn-carp farmers of Bangladesh.

Aquaculture Economics & Management, 12(3), 188-206.

Amos, T. T. 2007. Production and productivity of Crustacean in Nigeria. Journal of Social Science, 15(3), 229–

233.


10

Balcombe, K., Fraser, I., Latruffe, L., Rahman, M., & Smith, L.,2008. An application of the DEA double

bootstrap to examine sources of efficiency in Bangladesh rice farming. Applied Economics, 40(15),

1919–1925.

Banker, R.D., Charnes, A., & Cooper, W.W. 1984. Some models for estimating technical efficiency and scale

inefficiency in data envelopment analysis. Management Science, 30, 1078–1092.

Bogetoft, P., & Otto, L. 2010. Benchmarking with DEA, SFA, and R (Vol. 157). Springer Science & Business

Media, New York, USA.

Burgess, J.F., Wilson, P.W. 1998. Variation in inefficiency among US hospitals. Canadian Journal of

Operational Research and Information Processing (INFOR) 36, 84–102.

Brezis, E. S., Krugman, P. R., & Tsiddon, D. 1993. Leapfrogging in international competition: A theory of

cycles in national technological leadership. American Economic Review, 83(5), 1211–1219.

Chang, H., R.N. Boisvert, & L. Hung 2010. Land subsidence, production efficiency and the decision of

aquaculture firms in Taiwan to discontinue production. Ecological Economic, 69(12), 2448–2456.

Cinemre, H. A., Ceyhan, V., Bozoglu, M., Demiryurek, K., & Kilic, O. 2006. The cost efficiency of trout farms

in the black sea region, Turkey. Aquaculture, 251(2), 324–332.

Coelli, T. J., Rao, D. S. P., & Battese, G. E. 2005. An introduction to efficiency and productivity analysis.

Plymouth, Massachusetts: Kluwer Academic Publishers.

Den, T. D., Ancev, T., & Harris, M. 2007. Technical efficiency of prawn farms in the Mekong Delta, Vietnam.

51st AARES Annual Conference, February 13–16, 2007. Queenstown, NZ: Australian Agricultural and

Resource Economics Society.

Fisheries Directorate of Vietnam 2012. Summary report of planning for Vietnam’s fisheries development to

2020 and the vision of 2030 (in Vietnamese). Hanoi, Vietnam.

Fisheries Directorate of Ninh Thuan 2013. Planning for Ninh Thuan’s fisheries development to 2020 (in

Vietnamese). Ninh Thuan, Vietnam.

GSO. 2014. Ninh Thuan’s Statistic Yearbook (in Vietnamese). General Statistics Office, Vietnam.

Iliyasu, A., Mohamed, Z. A., Ismail, M. M., Abdullah, A. M., Kamarudin, S. M., & Mazuki, H. 2014. A review

of production frontier research in aquaculture (2001–2011). Aquaculture Economics &

Management, 18(3), 221–247.

Iliyasu, A., Mohamed, Z. A., & Terano, R. 2016. Comparative analysis of technical efficiency for different

production culture systems and species of freshwater aquaculture in Peninsular Malaysia. Aquaculture

Reports, 3, 51-57

Jensen, M., & Meckling, W. 1976. Theory of the firm: Managerial behavior, agency costs and ownership

structure. Financial Economics, 3, 305–360.

Johnson, A.L., Kuosmanen, T. 2012. One-stage and two-stage DEA estimation of the effects of contextual

variables. European Journal of Operation Research 220, 559-570.

Keramidou, I., & Mimis, A. 2011. An application of the double-bootstrap data envelopment analysis to

investigate sources of efficiency in the Greek poultry sector. World's Poultry Science Journal, 67(04),

675–686.

Latruffe, L., Davidova, S., & Balcombe, K. 2008. Application of a double bootstrap to investigation of

determinants of technical efficiency of farms in Central Europe. Journal of Productivity

Analysis, 29(2), 183–191.

Lebel, L., Tri, N. H., Saengnoree, A., Pasong, S., Buatama, U., & Thoa, L. K. 2002. Industrial transformation

and shrimp aquaculture in Thailand and Vietnam: Pathways to ecological, social, and economic

sustainability? AMBIO: A Journal of the Human Environment, 31(4), 311–323.

Long, L. K., Flaaten, O., & Anh, N. T. K. 2008. Economic performance of open-access offshore fisheries—The

case of Vietnamese longliners in the South China Sea. Fisheries research, 93(3), 296-304.

Long, L. K., & Binh, P. T. 2013. Profitability analysis for white leg shrimp aquaculture farms in Ninh Hoa

town, Khanh Hoa province (in Vietnamese). Journal of Fisheries Science and Technology, 2, 62–66,

Nha Trang University, Vietnam.

Nguyen, K. T., & Fisher, T. C. 2014. Efficiency analysis and the effect of pollution in the Mekong river

delta. Aquaculture Economics & Management, 18(4), 325–343.

Nilsen, O. B. 2010. Learning-by-doing or technological leapfrogging: A production frontier and efficiency

measurement in Norwegian salmon aquaculture. Aquaculture Economics & Management, 14(2), 97–

119.

Nielsen, R. 2011. Green and technical efficient growth in Danish fresh water aquaculture. Aquaculture

Economics & Management, 15(4), 262-277.

Olson, K., & Vu, L. 2009. Economic efficiency in farm households: Trends, explanatory factors, and estimation

methods. Agricultural Economics, 40(5), 587–599.

Onumah, E. E. 2010. Elements which delimitate technical efficiency of fish farms in Ghana. Journal of the

World Aquaculture Society, 41(4), 506–513.


11

Onumah, E. E., Hoerstgen-Schwark, G., & Brümmer, B. 2009. Productivity of hired and family labour and

determinants of technical inefficiency in Ghana's fish farms (No. 0907).

Diskussionspapiere/Department für Agrarökonomie und Rurale Entwicklung.

Roy, A. K., & Jens, N. 2008. Econometric approach for estimation of technical efficiency of aquaculture farms.

In Roy, A. K., & Serangi, N. (Eds.), Applied bioinformatics, statistics and economics in fisheries

research (pp. 501–518). New Delhi: New India Publishing Agency.

Sharma, K. R., & Leung, P. 2003. A review of production frontier analysis for aquaculture

management. Aquaculture Economics & Management, 7(1–2), 15–34.

Shephard, R.W. 1970. Theory of Cost and Production Function. Princeton University Press, Princeton.

Simar, L., & Wilson, P. 2000. Statistical inference in nonparametric frontier models. The state of the art. Journal

of Productivity Analysis, 13(12), 49–78.

Simar, L., & Wilson, P. W. 2007. Estimation and inference in two-stage, semi-parametric models of production

processes. Journal of Econometrics, 136(1), 31–64.

Simar, L., Wilson, P.W. 2011. Two-stage DEA: caveat emptor. Journal of Productivity Analysis 36 (2), 205–

218.

Simm, J., & Besstremyannaya, G. 2015. Package ‘rDEA’, (https://cran.r-

project.org/web/packages/rDEA/rDEA.pdf) (accessed 07.12.15).

Sinh, X.L. 2006. Major considerations on the fishery sector in the Mekong Delta of Vietnam. Presentation to the

CARD workshop, July 15–20, 2006, Cantho, Vietnam.

Tan, R.L., Y.T. Garcia, M.L. Dator, I.M.A. Tan, & D.E. Pemsl. 2011. Technical efficiency of Genetically

Improved Farmed Tilapia (GIFT) cage culture operations in the lakes of Laguna and Batangas,

Philippines. Journal of the International Society of Southeast Asian Agricultural Sciences, 17(1), 194–

207.

Tran, N., Bailey, C., Wilson, N., & Phillips, M. 2013. Governance of global value chains in response to food

safety and certification standards: The case of shrimp from Vietnam. World Development, 45, 325–

336.

VASEP. 2013. Report for Vietnamese shrimp exports in 2012 and forecast for the year 2013 (In Vietnamese).

Hanoi, Vietnam.

Zhang, Y., & Bartels, R. 1998. The effect of sample size on the mean efficiency in DEA with an application to

electricity distribution in Australia, Sweden and New Zealand. Journal of Productivity Analysis, 9,

187–204.

https://cran.r-project.org/web/packages/rDEA/rDEA.pdf

https://cran.r-project.org/web/packages/rDEA/rDEA.pdf


12

i The double bootstrap DEA model

Following Simar and Wilson’s (2007) Algorithm 2, the procedure modified for input-oriented VRS DEA

efficiency in this study is as follows:

Step 1: Estimate DEA input-oriented technical efficiency for all white-leg shrimp farms in the sample data set

using (1). Next, calculate the reciprocal of technical efficiency value, so-called technical efficiency score with

the range from one to infinity , for the jth farm.

Step 2: Use the method of maximum likelihood to estimate of as well as of in the truncated regression

of on Zi when , in which Zi is the vector of white-leg shrimp farm specific variables (individual farm

characteristics) defined in Table 1.

Step 3: For each j = 1,…, n, repeat the following four steps (i–iv) L1 to yield a set of bootstrap estimates

.

(i) For each j = 1,…, n, is drawn from .

(ii) For each j = 1,…, n, compute .

(iii) Construct a pseudo data set where and .

(iv) Using the pseudo data set and (1), calculate pseudo efficiency estimates for all j = 1,…, n.

Step 4: For each j = 1,…, n, calculate the bias-corrected estimator where the bias term is

.

Step 5: Adopting truncated maximum likelihood, regress on Zj to calculate estimates and .

Step 6: Repeat the following three steps (i–iii) L2 time to generate a set of bootstrap estimates

.

(i) For each j = 1,…, n, is drawn from .

(ii) For each j = 1,…, n, compute .

(iii) Adopting truncated maximum likelihood, regress on Zj to yield estimates and .

Step 7: Use the bootstrap estimates K and the estimates and generated in Step 5 to construct confidence

intervals for and . The percent confidence interval of the jth element of vector is constructed as

the such that the estimated confident interval for is

.

Documents

Analysis of technical efficiency of intensive white-leg