Resources Policy 65 (2020) 101543

Contents lists available at ScienceDirect

Resources Policy
journal homepage: [Link]

Assessment of oil refinery performance: Application of data envelopment

analysis-discriminant analysis
Amani Mohammed Atris
School of Environment and Society, Tokyo Institute of Technology, 3-3-6, Shibaura, Minato-ku, Tokyo, 108-0023, Japan

A R T I C L E I N F O A B S T R A C T

Keywords: Refineries are vital not only in the oil and gas industry but also in many other industries. Refinery products are
Refinery essential for many industrial sectors and end-users. However, the market for refinery products has witnessed a
Efficiency significant transformation due to changes made in products to meet market demand and environmental regu­
Data envelopment analysis
lations. This study examines refineries’ operational efficiency and conducts an efficiency-based rank assessment
Discriminant analysis
Operational efficiency
using an unbalanced panel dataset comprised of oil and gas refineries in four global regions (U.S. and Canada;
Europe; Asia-Pacific; Africa and the Middle East) covering 2008 to 2017. This study applies a combination of data
envelopment analysis and data envelopment analysis–discriminant analysis to examine the efficiency-based rank
for oil and gas refineries. A Kruskal–Wallis rank sum test is conducted to examine whether the average efficiency-
based ranks measures change over time and whether they differ among the four regions. Moreover, a Wilcoxon
rank sum test is utilized to investigate whether the adjusted efficiency averages differ between any two of the
regions under study over the 10 years between 2008 and 2017. The results indicate that the U.S. and Canada
display superior performance among the four regions, likely because that region contains major companies and
complex refineries that use highly advanced technology.

1. Introduction consumption. As a result, refineries in Asia, the Middle East, and Eastern
Europe have significantly upgraded their facilities and expanded their
The oil and gas (O&G) industry’s supply chain consists of an upper production scale over the past decade.
sector (exploration and development) and a lower sector (refinery and Refineries must consider their configuration processes in order to
sales). Refineries are vital in the O&G industry. Their products, such as adapt to global market changes. Refinery configuration is the combi­
petrochemicals, gasoline, and diesel, are crucial products for many in­ nation of refinery process units used to make refined products.
dustrial sectors and end-users. Oil refinery products fuel most of the A great number of process units increases refinery complexity, and
world’s transportation modes. Furthermore, industrial and commercial refinery type defines a refinery’s complexity level. There are four types
activities and electricity generation use products refined from crude oil. of refinery: topping, hydro skimming, conversion or cracking, and deep
The market for refinery products has witnessed a significant trans­ conversion or coking refineries. Topping refineries are the simplest type;
formation over the last few decades due to changes in products made to they work only on crude distillation and basic operation process. Hydro-
meet market demand and environmental regulations. Although the skimming refineries include crude distillation process units and more
refining market is mature and globally widespread, the profit margins of advanced units, such as those for catalytic reforming, hydro treating,
refineries in developed countries have dropped significantly in the past and product blending, which upgrade naphtha to gasoline and control
few years due to the high prices of refined oil products. The decline in the sulfur amount. Cracking refineries include not only hydro-skimming
refineries’ profit margins has been driven mainly by the increase in the characteristics but also catalytic cracking and hydrocracking processes,
prices of raw materials (e.g., crude oil prices and proposed local gov­ so that they can convert the heavy crude oil fraction to lighter products
ernment taxes), which has raised the total costs of final refined oil (e.g., gasoline, petrochemical, jet and diesel fuel). Coking refineries are
products. Accordingly, the corporate leaders of refineries are facing the most complex type. They can convert the heaviest oil residuals to
pressures that could lead to future divestitures of their business. On the lighter streams and valuable products. Cracking and coking refineries
other hand, developing countries have witnessed increases in oil are required to meet the market demand for qualified light products.

E-mail address: [Link]@[Link].

[Link]
Received 29 August 2019; Received in revised form 7 November 2019; Accepted 13 November 2019
Available online 25 November 2019
0301-4207/© 2019 Elsevier Ltd. All rights reserved.
A. Mohammed Atris Resources Policy 65 (2020) 101543

Refinery complexity depends on several factors, including product de­ and discuss policy implications for the refineries industry operations in
mand, local economics and regulations, and crude oil composition (e.g., every region by obtaining empirically efficiency based rank measures.
heavy/light, sour/sweet). All of these factors must be considered in This study uses a combination of data envelopment analysis (DEA)
order to enhance efficiency and meet rapidly changing regional and and DEA-discriminant analysis (DEA-DA). DEA classifies refineries into
global market trends. The key goal is to gain higher market share and two categories (efficient and inefficient) based on their efficiency scores.
achieve a higher level of profitability in order to maintain competi­ Then, DEA-DA is utilized to evaluate all refineries’ operational effi­
tiveness in the market. ciency scores and ranks to get an adjusted efficiency score for each re­
The profitability of refineries depends on many factors (see Ban­ finery. DEA-DA reduces the number of efficient refineries and generates
dyopadhyay et al. (2019); Cerda � et al. (2018); Zhao et al. (2017); Kor­ a single efficient decision-making unit (DMU) to present a wild industry
otin et al. (2017); Zhang et al. (2001)) such as the prices of crude oil and assessment based upon the efficiencies ranks. Further, due to a lack of
other raw material inputs, the characteristics of the regional markets in statistical inference methods used to complement DEA and DEA-DA
which the refinery operates and sells its products, refinery operations, computations, a Kruskal–Wallis rank-sum test is conducted to examine
process capacity, process complexity, the current structure of the re­ whether the average adjusted efficiency-based ranks measures change
finery industry, and the operational efficiency of the refinery company.1 over time and whether they differ among the four regions. Moreover, a
Global market regulations and standards, which require certain Wilcoxon rank-sum test is utilized to investigate whether the adjusted
environmental performance levels for refinery operations and major efficiency averages differ between any two of the regions under study
refined products, affect all refineries. Thus, refineries must take a global over the ten years between 2008 and 2017. Combining these methods
perspective when facing global market challenges. aids the decision making of corporate leaders and policymakers because
One global environmental protection factor is that the International it allows them to compare operational performance among regional
Maritime Organization (IMO), launched in October 2016, will become refineries in terms of their efficiency level and of their overall ranking.
effective in January 2020. This initiative aims to reduce the amount of Corporate leaders tend to pay more attention to their firms’ ranking in
sulfur emissions in bunker fuel from its current 3.5% to a maximum of their industry sector than to their comparative level of efficiency. Effi­
0.5%.2 Hence, refineries have to produce a low-sulfur product in order to ciency measures and rankings reveal the position of a refinery within the
meet the environmental regulation. The downstream oil market has to industry and indicate which refineries have room for performance
adapt to this regulation, and also become more flexible in the face of improvement. As mentioned, our performance assessment examines
global challenges by enhancing refineries’ ability to meet the global differences across four regions: the U.S. and Canada, Europe, the Asia-
market requirements imposed by falling oil prices (an uncontrollable Pacific, and Africa and the Middle East.
factor), new regulations on sulfur in marine bunker fuel, and other The remainder of this paper is organized as follows. Section 2 re­
factors such as regional expansion (e.g., the Asia-Pacific region and the views the literature on DEA studies of O&G refineries. Section 3 dis­
Middle East). cusses the study’s dataset, DEA and DEA-DA formulations, and rank sum
Refinery market capacity varies across regions. Fig. 1 shows the ca­ tests (Kruskal–Wallis and Wilcoxon). Section 4 discusses the study’s
pacity trends in the four regions examined in this study,3 showing that empirical results on the refineries’ efficiency and rankings based on our
the Asia-Pacific region has the highest capacity, followed by the U.S. and regional classification. Section 5 presents the study’s conclusion and
Canada, Europe, and the Middle East and Africa. Asia Pacific and the outlines potential future research possibilities.
Middle East and Africa have seen increasing trends over the last 10
years. On the other hand, the capacity of refineries in Europe has 2. Literature review
decreased since 2012, while capacity in the U.S. and Canada region is
relatively stable, with only a slight increase since 2015. 2.1. DEA studies applied to O&G refineries
These regional capacity trends are consistent with future market
expectations according to which Asia is expected to lead, with a capital The DEA method is widely used in many studies as a holistic
expenditure (capex) of US$254 billion allocated for announced and approach to evaluating the relative performance of a group of peer units,
planned projects up to 2022; this is followed by Africa and the Middle called DMUs. Glover and Sueyoshi (2009) reviewed prior studies and
East, with US$117 billion and US$88 billion, respectively.4 On the other documented the contribution of Professor William W. Cooper in terms of
hand, Western Europe has seen many refinery shutdowns, and a the historical development of the DEA. Sueyoshi et al. (2017) conducted
maximum of 800,000 barrel/day is at risk of shutdown because of sig­ a survey of 693 studies that applied the DEA to energy and the envi­
nificant challenges caused by changes in crude oil prices, decreasing ronment, classifying them into groups according to topic. Mardani et al.
demand for refineries’ products, and environmental regulations.5 These (2018) summarized 145 DEA studies on energy and environmental is­
regional factors are affecting refinery capacity. sues. They provided an overview of studies using different DEA appli­
The purpose of this study is to investigate O&G refineries’ opera­ cation types along with an explanation of their inputs and outputs.
tional efficiency in four global regions (U.S. and Canada, Europe, the Furthermore, many studies have applied the DEA in combination
Asia-Pacific, and Africa and the Middle East) using an unbalanced panel with a productivity index. For example, Chan Oh and Hildreth (2014)
dataset comprised of 696 global O&G refineries between 2008 and 2017, measured technical improvements of energy efficiency in the automo­
tive industry. Yang and Li (2017) conducted efficiency evaluations and
policy analysis of industrial wastewater control in China. Hsu (2013)
1
See: THAILAND INDUSTRY OUTLOOK 2018–20, Refinery Industry. performed an international comparison of the efficiency of government
([Link] health expenditures. Fa €re et al. (2004) evaluated environmental per­
26ba58c65/IO_Refinery_2018_EN.aspx). formance using a formal index number (a kind of environmental pro­
2
See ([Link] ductivity index), and Chowdhury et al. (2014) investigated productivity,
aper_Sulphur-[Link]). efficiency, and technological changes with and without case-mix used as
3
Data source: BP Statistical Review of World Energy (2018)([Link]
output categories in Ontario hospitals between 2002 and 2006. Feng
[Link]/content/dam/bp/business-sites/en/global/corporate/pdfs/energy-eco
et al. (2018) analyzed the sources of green total-factor productivity
nomics/statistical-review/[Link]).
4
See ([Link] (GTFP) changes and its inefficiency of China’s metal industry (MI) from
a-drive-global-refinery-capacity-additions-capex/). 2000 to 2015, from regional and provincial view. Song et al. (2018)
5
See ([Link] proposed a comprehensive decomposition framework that integrated
oil/091418-more-refineries-at-risk-of-closure-in-europe-new-capacities-else production-theoretical decomposition analysis with index decomposi­
where-platts-summit). tion analysis to distinguish the driving factors of CO2 emissions from

2
A. Mohammed Atris Resources Policy 65 (2020) 101543

Fig. 1. Refining capacity of the four regions.

China’s iron and steel industry between 2000 and 2014. Further, they They found that results obtained without using undesirable output as a
analyzed the different characteristics and drivers of CO2 emissions at the basis could be misleading and that environmental regulations seemed to
national, regional and provincial levels. Feng et al. (2019) examined the be less effective for efficient refineries. The study also showed the re­
sustainability of China’s MI since the 21st century through an overview finery age, as an uncontrollable variable, had no significant effect on the
and an analysis of GTFP to elucidate the current situation of the energy environmental efficiency of the refinery.
and environment in MI Sector from 2000 to 2015. Ma et al. (2019) Li et al. (2017) established a sustainability evaluation index system
examined the impact of government regulation on energy and CO2 for refinery firms comprising 17 representative indexes for refineries’
emissions performance in China’s mining industry. sustainability development. They applied a DEA-based model to a sus­
Our review finds that DEA studies on O&G companies, particularly tainability assessment of 15 refineries in China. They provided recom­
studies on O&G refineries, are limited compared with DEA studies for mendations for enhancing the refineries’ sustainability based on the
other applications, despite the importance of refineries’ products and results, such as reducing the total cost per unit, enhancing investments
their massive economic impact in the petroleum and petrochemical in­ in science and technology to maintain continuing innovation, increasing
dustry. Table 1 summarizes DEA studies on O&G refineries, including the social contribution rate and employee benefits, and considering
the authors’ names; the studies’ methodology, inputs, and outputs used energy-conservation and emissions-reduction policies in order to ach­
for efficiency assessment; and a brief description of each study. ieve clean production. Song and Zhang (2009) presented a novel method
Azadeh et al. (2017) evaluated the reciprocal impacts of managerial of evaluating and predicting oil refineries’ performance based on DEA
and organizational factors and RE for 41 gas refineries in Iran using DEA and SVM models. They found that SVM could be used to predict whether
and statistical methods. The DEA results indicated that two RE factors an oil refining enterprise was DEA-efficient because the SVM’s predic­
(learning and flexibility) had the most influence on the managerial and tion result was identical to that of the DEA model. Al-Najjar and
organizational factors and that the managerial factors had a weaker Al-Jaybajy (2012) utilized DEA to examine the relative efficiency of a
influence on RE than the organizational factors had. The results of the sample of oil refineries in Iraq from 2009 to 2010. They proposed using
statistical methods showed that the data were reliable and revealed a their research on oil refineries in Iraq in order to determine how best to
strong direct relationship between the two main factors (organizational apply DEA to measure efficiency and overcome efficiency problems.
and managerial). Azadeh et al. (2015) applied a combined ANN and Mekaroonreung and Johnson (2010) measured the technical efficiency
multivariate approach to evaluate the performance of five gas refineries of 113 U.S. oil refineries in 2006 and 2007 by comparing multiple DEA
in Iran from 2005 to 2009 by considering two cases (financial indicators methods. They considered undesirable output in the production process
and financial/non-financial indicators). The refineries were ranked and found that domestic refineries could improve efficiencies regardless
using DEA, PCA, ANN, and NTX. A sensitivity analysis found that the of the DEA assumptions used and that environmental regulations
DEA was the most noise-resistant of the methods and that the results for reduced the amount of potentially desirable outputs produced by some
the financial indicators and those for the combined operational in­ facilities. Managi et al. (2004) examined the total factor productivity of
dicators differed slightly. offshore O&G production in the Gulf of Mexico from 1947 to 1998. The
Bevilacqua and Braglia (2002) evaluated the environmental effi­ study utilized DEA to measure productivity and decomposed it into
ciency of seven AgipPetroli oil refineries owned by Italy’s Eni Group technological change, efficiency change, and scale change. They applied
Company. In particular, they examined the environmental impact of DEA to the dataset. The results indicated that productivity declined for
their air emissions from 1993 to 1996. They confirmed that an appli­ the first 30 years of the study period but that technological change
cation of an environmental management system (EMS) could generate outpaced depletion, causing productivity to increase rapidly, particu­
various benefits, such as reduced liability, improved public image, better larly in the last five years of the study period. Sueyoshi (2000) proposed
compliance, reduced costs, and better access to capital. Francisco et al. a stochastic DEA model that included future information, called a “DEA
(2012) used DEA models to evaluate environmental efficiency on the future analysis.” The suggested approach was used to plan the restruc­
basis of an undesirable output associated with production processes. turing strategy of a Japanese petroleum company. The results indicated

3
A. Mohammed Atris Resources Policy 65 (2020) 101543

Table 1
DEA studies applied to oil and gas refineries.
Authors Methodology Description Inputs Outputs

Azadeh et al. (2017) - DEA This study measured the mutual impact of - Commitment management - Managerial factors
- Statistical methods Resilience Engineering (RE) and managerial and - Learning - Organizational factors
(Spearman’s test & organizational factors for 41-gas refineries in - Awareness
alpha test) Iran. - Flexibility
- Self-organization
- Redundancy
Azadeh et al. (2015) - DEA This study presented a combined approach for - Number of personnel - Return on sales
- Principal evaluating the performance of 5 gas refineries in - Total costs (except the cost - Operating earnings
Component Iran over the period 2005–2009. of goods sold (COGS)) - Net income
Analysis (PCA) - Personnel education cost - Return on assets
- Numerical - R&D cost - Capital return
Taxonomy (NTX) - Fixed non-current assets - The amount of gas sent to the torch/the
- Artificial Neural - Stock turnover amount of received sour or sweet gas
Network (ANN) - Asset turnover ratio - Operating capacity divided by nominal
- Statistical methods - Current assets turnover capacity
(T-test) ratio - Operating capacity of each LPG
- Amount of refinery’ s fuel production unit divided by design
consumption/amount of capacity of each LPG production
received sour or sweet gas - Operating capacity of each refrigeration
and dew point control unit divided by
design capacity of each refrigeration and
dew point control unit
- Operating capacity of each sulfur
production unit divided by design
capacity of each sulfur production unit
- Operating capacity of each liquids
stabilization unit divided by design
capacity of each liquids stabilization unit
- Operating capacity of each dehydration
unit divided by design capacity of each
dehydration unit.
Bevilacqua and DEA This study examined the environmental - Power planet fuel - SO2
Braglia (2002) performance of the seven AgipPetroli oil consumption - NOx
refineries set up in Italy from 1993 to 1996 - Oil processed - Total Suspended Particles (TSP)
- Refining planet fuel - Volatile Organic Compounds (VOC)
consumption - CO
- CO2
Francisco et al. DEA This study assessed the environmental efficiency - Idleness percentage of the - Desirable: (refinery production volume)
(2012) for 10 oil refineries in the public sector in Brazil operating plant - Undesirable: (generated effluents)
in 2004 - The amount of water
consumed
Li et al. (2017) DEA This study examined the sustainable performance - Asset-liability ratio - Return on assets
for 15 refineries in China - Comprehensive energy - Asset turnover
consumption per unit of - Investment intensity in science and
output technology
- Entire cost per unit - R&A personnel
- Solid waste emissions per - Comprehensive commodity rate
unit of output - Environmental protection cost per 10
- Wastewater emissions per thousand Yuan of output
unit of output “three wastes”
- Waste gas emissions per 1 disposal rates
unit of output 2 social contribution rate
- Employee turnover rate 3 income per capital.
SONG and ZHANG. DEA This study presented a method for evaluating and - Manpower cost index - Return on average capital employed
(2009) Support Vector predicting oil refineries performance, using two - Operation cost index - Operation cost added value index
Machine (SVM) deferent techniques. - Manpower cost added value index
- Assets added value index
Al-Najjar and DEA This study measured the relative efficiency of 12 - Crude oil - Naphtha
Al-Jaybajy (2012) oil refineries in Iraq over the period 2009–2010. - Workforce - Gasoline
The study revealed that there is a waste or - Electricity - Kerosene
underutilization of resources at the inefficient - Land - Fuel oil
refineries.
Mekaroonreung and DEA- variable returns This study Investigated the technical efficiency of - Equivalent distillation as a - Gasoline
Johnson (2010) to scale (VRS) 113 US oil refineries in operation over two years, proxy of capital - Distillate
2006–2007. They found that domestic refineries - Energy - Toxic release
could improve efficiencies regardless of the - Crude oil
different DEA assumptions. Further,
environmental regulations reduced the amount of
potentially desirable outputs produced by some
facilities.
Managi et al. (2004) DEA This study examined the hypothesis that - Number of platforms - Oil production
technological change has offset depletion for - Ave. platform size - Gas production
offshore oil and gas production in the Gulf of Number of exploration wells
Mexico using a unique micro-level data set over
(continued on next page)

4
A. Mohammed Atris Resources Policy 65 (2020) 101543

Table 1 (continued )
Authors Methodology Description Inputs Outputs

the period 1947–1998. The results indicated that - Number of development

the technological change has outpaced depletion wells
and productivity has increased rapidly, - Average drilling distance for
particularly in the most recent five years of the exploratory wells
study period. - Average drilling distance for
development wells
- Produced water
- Weighted innovation index
- Horizontal & directional
drilling (exploratory)

- Horizontal & directional

drilling (development)
Sueyoshi (2000) A stochastic DEA This study suggested the DEA approach for - Number of employees - Gasoline
planning the restructuring strategy of a Japanese - Size of gas station - Petrol
petroleum company in 1998–1999. The results - Operation cost
showed that large gas stations operated more
efficiently than small ones.

that large gas stations worked more efficiently than small ones. examines more firms than have been examined in previous studies; the
Moreover, the O&G industry is a part of the mining, quarrying and larger dataset allows a broader analysis. To address the shortcomings of
O&G extraction sector, where crude oil extraction is considered as liquid previous O&G studies, this study uses a unique unbalanced panel dataset
minerals. Therefore, it is worthy of mentioning the other applications of for O&G refineries in the four regions listed above covering the 10 years
DEA on mining and quarrying industries. For example, Zhu et al. (2018) from 2008 to 2017. Further, this study reflects the impact of global
utilized slacks-based global DEA to analyze the GTFP of mining and market challenges and the requirement on refinery efficiency. Then, this
quarrying industry and its five sub-industries in China from 1991 to study addressed some practical suggestions to improve the overall re­
2014 concerning to technology, scale, and management. Further, they finery operational efficiency, which have gained importance in modern
employed a Malmquist index to distinguish the key factors that account society as one of the primary energy efficiency issues.
for the changes of GTFP. The results indicated that the GTFP of the Moreover, this study investigates differences in the adjusted
mining and quarrying industry improved during the period of study due efficiency-based ranks among four regions by the Kruskal-Wallis rank
to technological progress, whereas declining in scale efficiency and sum test, in which the null hypothesis is that there are no differences in
management efficiency curb the growth of GTFP. Li et al. (2019) used a the average adjusted efficiency levels among the four regions or between
modified dynamic DEA SBM model to estimate the coal production ef­ different periods. As the mathematical structure of DEA applications
ficiencies and land damage in 24 Chinese provinces from 2011 to 2016. lack statistical inferences, the Kruskal–Wallis rank sum test is required to
They found that seven provinces were fully efficient in all the study’s investigate the null hypothesis where different groups or periods have
years, efficiencies of coal industry labor force, fixed assets, and coal the same distribution for average adjusted efficiency levels.6 This study
production dropped significantly. However, the land damage was un­ also utilizes a Wilcoxon rank sum test to check for differences in average
clear, and Beijing and Guangxi had the least mining quantities, but the adjusted efficiency levels between region pairs, where the null hypoth­
greatest need for land damage improvements. Roman et al. (2017) esis is that there is no apparent difference in the average adjusted effi­
applied two-step DEA to evaluate the efficiency of the mining and ciency levels between any two regions.
quarrying industry of Visegrad 4 countries between 2011 and 2015. The
results revealed that the Slovak Republic is the most inefficient country.
In addition, they employed the double bootstrapped efficiencies that 2.3. Hypotheses
reduce lacks of DEA, and then truncated regression. The results of the
proposed model show that Gross investments in machinery, equipment, This study proposes three null hypotheses based on the results of the
construction and alteration of buildings and Human Development Index previous studies on O&G refineries discussed above.
(HDI) have a positive impact on efficiency. Wang and Zhang (2018) The operational efficiency of refineries over the world is determined
investigated four key conceptual dimensions of innovation for relative by various vital factors, such as proximity to oil sources, the intensity of
technological innovation of China’s coal mine and developed groups of oil product consumption, oil transportation routes, and refinery con­
indicators to assess the performance of coal mine’s technological inno­ struction. Furthermore, refineries vary by type and oil-refining tech­
vation activities. Then, they developed an index system based on DEA to nology, which condition the complexity level of a refinery.
audit the progress of innovation and development. They found that the A refinery’s complexity is reflected in its oil-refining capacity utili­
proposed index system and model could effectively assess coal mine’s zation and the technology used by the refinery to obtain highly purified
relative technological innovation capability and get the input gap to­ oil products. Differences in complexity levels are driven mainly by the
wards efficiency, safety, and sustainability. Hosseinzadeh et al. (2016) regional and locational characteristics of the refineries. Therefore, based
applied bootstrap DEA to distinguish the balance of efficiency gains and on these refinery characteristics, we propose that there are significant
losses for 33 Australian mining firms from 2008 to 2014. The results differences in average adjusted efficiency levels among the four regions
indicated that mining companies involved in exploration and extraction and between any two regions. Our first hypothesis is thus as follows:
activities have been less efficient than those involved in metal process­ H0. The average adjusted efficiency measures are invariantly distrib­
ing or mining services. uted between the four regions. There is no difference between the av­
erages of their adjusted efficiency levels.
2.2. Contribution of this study

A careful review of DEA studies on O&G refineries indicates that, to 6

Bootstrap DEA is proposed by Simar and Wilson (1998, 2000) as a way to
the best of our knowledge, no previous study has applied a combination equip DEA with statistical inference using a sample technique. This study does
of DEA and DEA-DA to O&G refineries. Furthermore, this study not use such an application.

5
A. Mohammed Atris Resources Policy 65 (2020) 101543

The second hypothesis is as follows: period. Higher revenue indicates that the company is successful in its
business sector.
H0. The average adjusted efficiency levels in any two regions are
� Net income: This variable indicates the net amount of money (profit)
equal. No statistically significant differences are observed between any
that the company earned in a specific period. This variable is
two regions.
calculated by subtracting all the business expenses from the amount
Over the last few decades, international oil refinery regulations have
of total revenue. Increased net income indicates a net increase in
been strengthened in order to reduce pollution. Environmental regula­
shareholders’ equity, which will be distributed as dividends among
tion by the IMO is one of the most significant challenges facing refineries
the shareholders.
all over the world. The IMO is pressuring refineries to produce low-
� Total enterprise value (TEV): This determines the overall economic
sulfur products. For instance, refineries in North America and Europe
value of the company. It is calculated as the market price of a stock
have to reduce their sulfur oxide content to levels under 0.1%. While
multiplied by the total number of shares outstanding. This variable is
Asia and the Middle East are witnessing the construction of many
an important measure for analyzing potential takeover targets.
complex refineries and anticipating a projected capacity increase of 7
million additional barrels per day by 2023, the IMO is radically changing
Table 2 presents descriptive statistics for the data on refining com­
refineries’ processes (complexity) in order to reduce the sulfur content
panies operating in the U.S. and Canada, Europe, the Asia-Pacific, and
of fuel oil used in ships from 3.5% to a maximum of 0.5%. The IMO’s
Africa and the Middle East. The average, standard deviation, minimum,
regulations will come into effect in January 2020, which is outside our
and maximum are denoted by “Avg.,” “S.D.,” “Min.,” and “Max.,”
study period. Therefore, the third null hypothesis is as follows:
respectively. In 2008, the Asia-Pacific region exceeds the other three
H0. The average adjusted efficiency measures are invariantly distrib­ regions in all average production factors. This is due to the increased
uted over the 10 years of the study period. No statistically significant demand for refinery products in their markets, particularly China and
shifts in average adjusted efficiency levels occur over time. India. Large-scale upgrading is ongoing as a response to this increase in
market demand. From 2014 to 2017, U.S. and Canada region shows the
3. Methodology highest values for all production factors except total debt and net income
in 2014 and 2016. Since 2014. U.S. shale oil has created a boom in the
3.1. Data crude oil industry in this region and led the US to become a crude oil
exporter in 2017 for the first time. Shale oil made up more than 1/3 of
This study uses an unbalanced panel dataset comprising O&G re­ the onshore production of crude oil in the U.S.
fineries located in the four regions listed above. The dataset was The Asia-Pacific region has the highest total debt among all regions
collected using S&P CAPITAL IQ PLATFORM7 covering 2008 to 2017. in 2008, and from 2011 to 2014, the values were $2421.4, $2492.8,
This study uses two DEA models: the first is the DEA radial model (input $2788.6, $2320.8, and $2277.5 million US, respectively. The U.S. and
oriented), which is used for efficiency assessment; the second is the DEA- Canada region shows the highest total revenue from 2010 to 2017, at
DA model, which is used for group classification and ranking via a $11611.1, $16002.6, $24897.1, $28504.8, $20873.1, $19835.8,
financial dataset that contains four inputs and three outputs. These are $17426.9, and $22248.6 million US$, respectively. The input and
summarized below. output datasets show the highest average levels for all regions in 2017,
[Four Inputs]. except for total revenue in 2013, which displays the highest value
($16254.1 million US).
� Number of employees: This variable reflects one of the main input
factors used in the production process, along with capital and ma­ 3.2. Models
terials. This variable is often used as an index for company size and
scale. It is the main factor considered in M&A (mergers and acqui­ 3.2.1. DEA
sitions) decisions because their main target is often to reduce the The DEA method is a holistic approach to evaluating firms’ efficiency
number of employees and labor costs. levels which DMUs use multiple inputs and outputs to produce goods/
� Total assets: This variable reflects the total amount of investment, services. The DEA is unique in that it compares the performance of each
cash, equipment, receivables, and all other assets as reported in the DMU to that of all other DMUs. This study applies a radial input-oriented
balance sheet. Moreover, total assets reflects the borrower’s fiscal DEA model under the assumption of VRS technology. This model re­
strength. This input consists of two kinds of assets: current and fixed. duces inputs to produce a certain amount of outputs and removes the
More total assets produce more revenue and income. This variable scale economy or diseconomy of a DMU from the efficiency score. This
can also be used as a firm size index. model is often used in empirical studies of DEA.
� Total cash and short-term investments: This variable reflects a kind The mathematical symbols used to illustrate production factors are
of current assets that are highly liquid; these can be used quickly if as follows:
the company needs quick money or cash. Higher cash and short-term
investments allow firms to hedge against unexpected future fluctu­ (a) Xj ¼ ðx1j ; x2j ; …; xmj ÞT > 0: a column vector of m inputs of the j-th
ations in profits and to enhance their liquidity ratios. DMU (j ¼ 1; …; n), and
� Total debt: This variable reflects the volume of loans and liabilities (b) Yj ¼ ðy1j ; y2j ; …; ysj ÞT > 0: a column vector of s outputs of the j-th
that the company uses to finance its investments, including long- and DMU (j ¼ 1; …; n), where superscript “T” indicates a vector
short-term liabilities. Higher total debt to total assets increases the transpose. The inequality (>) implies that the relationship is
degree of leverage and financial risk. applied to all components of the two column vectors.

[Three outputs]. In addition to the above production factors, which are given as an
observed dataset, this study uses the following symbols (unknown),
� Total revenue: This reflects the total amount of money that the which are measured by applying the DEA:
company receives from the sales of products and services in a specific
(c) dxi � 0: unknown slack variable of the i-th input (i ¼ 1; …; m),
(d) dyr � 0: unknown slack variable of the r-th output (r ¼ 1; …; s),
7
See S&P CAPITAL IQ PLATFORM: [Link]
lligence/en/index.

6
A. Mohammed Atris Resources Policy 65 (2020) 101543

Table 2
Production Factors for the refineries’ Companies among four regions on Average.
Variable Outputs Inputs

Net Income Total Total Enterprise Employees Total Total Cash & ST Total Debt
Revenue Value Assets Investment

Unit Millions$ Millions$ Millions$ Employee Millions$ Millions$ Millions$

2008 Avg. U.S. and Canada 151.3 10711.2 1625.0 2760.7 2795.4 109.7 747.2
Europe 59.0 5674.2 1711.0 1472.2 2004.0 123.7 542.2
Asia- Pacific 418.3 14331.7 5691.9 5417.4 7922.0 603.7 2421.4
Africa and Middle 63.5 4705.3 948.1 937.9 1636.5 312.9 466.1
East
Total 267.2 11171.1 3629.9 3728.5 5193.3 399.4 1550.7
Avg.

2009 Avg. U.S. and Canada 38.2 2547.7 790.7 879.0 1039.9 59.3 289.6
Europe 225.5 9074.4 4586.6 6061.3 6417.5 262.7 1959.7
Asia-Pacific 67.6 7789.0 2391.0 3532.9 3438.1 471.8 1254.8
Africa and Middle 102.3 3168.8 990.3 922.5 1549.5 303.6 521.2
East
Total 85.6 5813.5 1983.2 2675.7 2838.7 324.9 966.3
Avg.

2010 Avg. U.S. and Canada 76.1 11611.1 2704.9 2796.0 4675.9 469.6 1022.6
Europe 230.0 10147.0 3789.2 4978.8 5943.8 311.6 1542.8
Asia-Pacific 161.1 8453.5 3033.1 3332.6 4565.6 438.0 1435.4
Africa and Middle 94.9 5218.5 3637.4 1505.3 4068.3 544.7 1673.8
East
Total 141.4 8815.1 3167.3 3138.7 4694.1 445.2 1402.2
Avg.

2011 Avg. U.S. and Canada 355.5 16002.6 2553.0 3551.2 5360.8 416.9 964.8
Europe 153.3 10258.3 2418.7 4132.8 4772.9 265.3 1306.2
Asia-Pacific 300.2 13587.5 3773.6 3699.6 7453.6 434.8 2492.8
Africa and Middle 83.9 4857.8 2151.6 1244.5 2840.5 170.0 1102.4
East
Total 265.6 12551.7 3047.6 3351.9 5897.3 370.9 1739.0
Avg.

2012 Avg. U.S. and Canada 638.9 24897.1 5532.8 4240.1 7712.6 791.8 1237.1
Europe 154.8 13141.4 3114.4 4350.0 5547.0 299.4 1346.9
Asia-Pacific 166.6 15580.4 4970.4 3884.5 8137.7 438.2 2788.6
Africa and Middle 79.9 4841.8 1900.0 980.6 2581.1 223.7 1134.9
East
Total 276.4 15718.6 4310.4 3459.2 6636.8 476.5 1880.3
Avg.

2013 Avg. U.S. and Canada 596.5 28504.8 7770.0 4489.2 9004.8 903.9 1314.4
Europe 152.0 12704.4 3521.2 5865.2 6249.7 327.2 1227.8
Asia-Pacific 165.5 14239.7 3725.8 3725.8 7027.6 366.5 2320.8
Africa and Middle 63.0 3486.2 1234.6 729.4 1720.5 197.6 592.4
East
Total 265.6 16254.1 4407.5 3618.2 6633.0 482.7 1671.1
Avg.

2014 Avg. U.S. and Canada 564.1 20873.1 6388.7 4094.1 7427.7 625.8 1488.6
Europe 30.9 5967.3 2979.6 1879.7 3065.0 117.4 848.2
Asia-Pacific 108.9 11925.9 3171.5 2931.8 6296.1 495.9 2277.5
Africa and Middle 75.6 2720.6 1239.3 1041.8 1361.5 190.9 448.2
East
Total 271.7 13570.7 4067.4 3022.7 5795.3 479.8 1626.5
Avg.

2015 Avg. U.S. and Canada 838.1 19835.8 9706.8 6012.2 10721.8 770.2 2307.8
Europe 252.3 8602.6 3386.8 4412.9 4707.9 592.3 1245.2
Asia-Pacific 49.1 3452.7 1468.3 1235.5 1620.4 146.5 493.3
Africa and Middle 97.5 1784.2 1239.4 885.9 1125.6 126.8 392.5
East
Total 280.5 7891.0 3726.4 2763.7 4192.4 354.6 1022.7
Avg.

2016 Avg. U.S. and Canada 379.9 17426.9 10694.0 5916.4 11748.1 900.1 2817.8
Europe 392.4 7132.1 3433.9 4291.3 4655.8 502.3 1002.7
Asia-Pacific 71.0 2877.4 1262.9 1262.5 1606.1 161.4 424.3
Africa and Middle 78.6 1779.9 1191.1 972.8 1572.0 198.6 553.4
East
Total 182.3 6442.6 3614.0 2630.6 4235.1 375.5 1051.3
Avg.

2017 Avg. U.S. and Canada 1026.1 22248.6 14951.9 6611.3 14869.3 997.7 3627.1
Europe 467.7 9330.4 4646.2 4126.4 5316.7 603.9 911.6
Asia-Pacific 311.5 8054.9 5993.2 6069.6 7436.7 629.5 2232.6
(continued on next page)

7
A. Mohammed Atris Resources Policy 65 (2020) 101543

Table 2 (continued )
Variable Outputs Inputs

Net Income Total Total Enterprise Employees Total Total Cash & ST Total Debt
Revenue Value Assets Investment

Unit Millions$ Millions$ Millions$ Employee Millions$ Millions$ Millions$

Africa and Middle 116.2 2215.3 1358.8 940.3 1649.4 241.8 587.4
East
Total 435.8 9934.7 6738.3 4986.8 7571.4 627.3 2040.3
Avg.

Total Avg. All 3364.81 439.03 1516.01 5474.22 253.64 10962.30 3960.15
Min 1.00 0.04 0.00 2.08 0.12 0.38 1.61
Max 140483 10474 30342 109937 5106 165960 100833
S.D. 8244.02 917.41 3416.99 11648.72 659.89 22079.99 8916.45

(e) λ ¼ ðλ1 ; …; λn ÞT : unknown column vector of “intensity” or m

P
All the DMUs are classified by a discriminant function ð vi xij þ
“structural” variables, i¼1
(f) ε: a small number to be prescribed by the DEA researcher, in this s
P
wr yrj þ σÞ. Unknown weights (vi for i ¼ 1;:::; m;wr for r ¼ 1;:::;s)
study ε ¼ 0:0001. r¼1
indicate the slope of the discriminant function. (Note that vi and wr are
The input-oriented radial DEA model used in this study is as follows: dual variables (multipliers) in the DEA but are weights in the DEA–DA.)
Pm Ps
hX X i The constraints ( vi þ wr ¼ 1; vi � εζi ; i ¼ 1;:::;m; wr � εζr ; r ¼
m s
Minimizeθ þ ε dxi þ dyr i¼1 r¼1
i¼1 r¼1
1; :::; sÞ) indicate that all the weights are positive, so that the discrimi­
X nant function is a full model. The sum of the unknown weights is unity.
n
s:t: xij λj þ dxi ¼ θxij ði ¼ 1; …; mÞ;
j¼1 This restriction is often referred to as “normalization.” Model (2) in­
corporates binary variables (ζi : binary for all i, and ζr : binary for all r) to
X
n
(1)
yij λj dyr ¼ yrj ðr ¼ 1; …; sÞ; count the number of positive weight estimates. It is possible to change
j¼1 the number of binary variables to set the weights at a number lower than
X
n
the actual number (i.e., it is possible to reduce the number of binary
λj ¼ 1; variables according to the importance of each factor and depending on
j¼1
whether you need to deal with zero and/or negative data). Such a
λj � 0ðj ¼ 1; …; nÞ; θ : URS; dxi � 0ði ¼ 1; …; mÞ; dyr � 0ðr ¼ 1; …; sÞ; change depends upon the degree of freedom between the number of
observations (DMUs) and the number of weights. After applying Model
where URS is unrestricted.
(2) to the dataset, we obtain an optimal solution and compute the
following score for the j-th DMU:
3.2.2. DEA-DA
The DEA method is a management science technique, and DA is a X
m X
s
ρj ¼ v�i xij þ w�r yrj þ σ� for all j ¼ 1; :::; n: (3)
statistical methodology. The DEA-DA is a research tool that combines
i¼1 r¼1
managerial and statistical approaches to predict group membership in
sampled data. This study applies a DEA-DA model and classifies all Using the ρj score, the adjusted efficiency score is computed for the
DMUs into efficient (E) and inefficient (IE) groups as formulated in refineries through the following procedure:
Model (2):
X X (a) Find the maximum and minimum values of ρ by maxj ρj and
Min: M Zj þ Zj minj ρj .
(b) Find the range between them by
j2E j2IE
Xm X
s
st: vi xij þ wr yrj þ σ þ Mzj � 0; j 2 E; (b-1) range (A) ¼ maxj ρj min � j ρj �if minj is non-negative and
i¼1 r¼1 (b-2) range (B) ¼ maxj ρj þ �minj � if minj ρj is negative.
X X (c) The adjusted efficiency score for the j-th DMU is measured by
m s
vi xij þ wr yrj þ σ Mzj � ε; j 2 IE;
i¼1 r¼1 (2) (c-1) Efficiency ¼ ½ρj minj �= ½range ðAÞ� if minj ρj is non-
X
m X
s negative and
vi þ wr ¼ 1; vi � εζi ; i ¼ 1; :::; m; wr � εζr ; r ¼ 1; :::; s; (c-2) Efficiency ¼ ½ρj þminj �=½range ðBÞ� if minj ρj is negative.
i¼1 r¼1
Xm X
s
ζi ¼ m; ζr ¼ s; σ : URS; vi � 0 for all i; wr � 0 for all r; 3.2.3. Rank sum tests
i¼1 r¼1 This study utilizes the Kruskal–Wallis rank sum test (H) to examine
Zj : binary for all j; ζi : binary for all i; and ζr : binary for all r:
the null hypotheses regarding whether the different groups and periods
have the same distribution. This study reorders all DMUs in ascending
here, M is a prescribed large number, and ε is a prescribed small number.
order based on their average adjusted efficiency scores to compute the
It is necessary to specify the two numbers before solving Model (2). The
test (Sueyoshi and Goto, 2012). Let Rjt denote the rank of the j-th DMU in
objective function minimizes the total number of incorrectly classified
the t-th group (or period). The rank sum of all DMUs in the t-th group (or
DMUs by counting a binary variable (Zj ). In this classification, the n
P
inefficient group (IE) has less priority than the efficient group (E). period) is Rt ¼ Rjt where n stands for the number of companies at the
j¼1
Therefore, we add M to the efficient group in the objective of Model (2).
The discriminant score is expressed by σ ðj 2 EÞ and σ ε ðj 2 IEÞ, t-th group (or period). The KW test is mathematically expressed as
respectively. The small number (ε) is incorporated into Model (2) in follows:
order to avoid a case where an observation exists on an estimated
discriminant function.

8
A. Mohammed Atris Resources Policy 65 (2020) 101543

12 XT
R2t the sample, respectively.
H¼ 3ðN þ 1Þ: (4) Table 4 shows a comparison of average efficiency scores using the
NðN þ 1Þ t¼1 n
conventional DEA and average adjusted efficiency scores using DEA-DA.
here, the number (N) refers to the total number of DMUs in all groups (or To clarify the difference, for 2014, the average efficiency scores and
periods). The statistic (H) follows the x2 distribution with a degree of average adjusted efficiency scores are 0.3200 and 0.5270 for Europe
freedom (df ¼ T-1). Hollander and Wolfe (1999) have described the 0.7103 and 0.6543 for the U.S. and Canada, respectively. The DEA-DA
Kruskal–Wallis rank sum test in detail. results show that applying this discriminant analysis to an industry-
When multiple DMUs have the same rank, the H statistic needs to be wide evaluation is important for determining the real efficiency per­
adjusted as follows: formance in each region.
Fig. 2 depicts the average adjusted efficiency scores of the four re­
� ��� P �
12 XT
R2t q gions from 2008 to 2017. U.S. and Canada region shows the best per­
Hc ¼ 3ðN þ 1Þ 1 : (5)
NðN þ 1Þ t¼1 n N3 N formance, followed by the Asia-Pacific, Africa and the Middle East, and

here, q ¼ z3 z where z denotes the number of observations on the same

Table 4
rank.
Average of efficiency and adjusted efficiency scores.
Region Year Average of Average of Rank
3.2.4. Wilcoxon rank sum test
efficiency score adjusted
This study applies a Wilcoxon rank sum test (also known as the efficiency
“Mann–Whitney test”) to examine the null hypotheses regarding
U.S. and Avg. 2008 0.6761 0.5589 21
whether there is no significant statistical difference between any two
Canada 2009 0.7069 0.6589 39
regions (i.e., whether the medians of the two regions are equal; Mann 2010 0.5805 0.5825 31
and Whitney, 1947). The rankings of the DMUs are calculated according 2011 0.6281 0.6184 35
to their average adjusted efficiency scores. As both samples are larger 2012 0.7087 0.6578 38
2013 0.7497 0.6661 40
than 10, the Wilcoxon rank sum test is mathematically treated as
2014 0.7103 0.6543 37
follows: 2015 0.6694 0.6416 36
2016 0.5654 0.5310 13
nA ðnA þ nB þ 1Þ
μA ¼ (6) 2017 0.6599 0.6040 34
2
Europe Avg. 2008 0.4243 0.6015 33
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2009 0.4650 0.5205 8
nA nB ðnA þ nB þ 1Þ
σA ¼ (7) 2010 0.4753 0.5612 24
12 2011 0.4301 0.5508 17
2012 0.4447 0.5096 3
RA μA 2013 0.4315 0.5100 4
z¼ (8)
σA 2014 0.3200 0.5270 10
2015 0.4376 0.5305 11
here, μA is the mean for group A (i.e., the expected value for group A), σ A 2016 0.4899 0.5413 14
2017 0.5848 0.5771 28
is the variance of group A, RA is the sum of the rankings of group A, and Z
is the critical value. Asia-Pacific Avg. 2008 0.5454 0.5583 20
2009 0.4584 0.5112 5
2010 0.4815 0.5647 25
4. Empirical results 2011 0.4942 0.5600 22
2012 0.4978 0.5076 2
The number of companies and their efficiency results obtained from 2013 0.5123 0.5308 12
model 1 are summarized in Table 3. The total number of refineries in 2014 0.3876 0.4874 1
2015 0.4949 0.5807 29
each region is 178 in the U.S. and Canada, 66 in Europe, 328 in the Asia-
2016 0.4581 0.5456 15
Pacific region, and 124 in Africa and the Middle East. These results show 2017 0.5145 0.5689 26
that the highest efficiency score is for the U.S. and Canada, where 46
Africa and Avg. 2008 0.6263 0.5559 19
firms achieved an efficiency score of 1, and 61 refineries received scores Middle East 2009 0.6694 0.5820 30
of less than 0.5; these represent 26% and 34%, respectively, of the U.S. 2010 0.6358 0.5712 27
and Canada sample. Europe has the lowest number of firms because of 2011 0.5393 0.5529 18
the wave of refinery shutdowns over the last decade Four refineries in 2012 0.6202 0.5859 32
2013 0.5527 0.5177 6
Europe are fully efficient, with an efficiency score of 1, and 42 com­
2014 0.5823 0.5244 9
panies have efficiency scores less than 0.5; these firms represent 6% and 2015 0.5633 0.5610 23
64% of the European sample, respectively. The Asia-Pacific region has 2016 0.5145 0.5202 7
the highest number of refineries in the sample. Only 40 of the region’s 2017 0.4870 0.5486 16
refineries are efficient, and 208 obtained efficiency scores less than 0.5; Note: the DEA-DA Model obtains the adjusted efficiency score. This study pro­
these refineries represent 12% and 63% of the Asia-Pacific sample, posed M ¼ 10000 and ε ¼ 0.0001 for the computation of the DEA-DA model. The
respectively. Africa and the Middle East region has 11 efficient refineries optimal solution of DEA-DA is v1 ¼ 0.0001, v2 ¼ 0.1307, v3 ¼ 0.1535,
and 56 firms with scores less than 0.5; these represent 9% and 45% of v4 ¼ 0.2287, w1 ¼ 0.1727, w2 ¼ 0.0174, w3 ¼ 0.2968 and σ ¼ 0.008.

Table 3
Summarized efficiency results of model 1.
Efficiency score 1 0.9 0 .8 0 .7 0 .6 0 .5 less than 0.5 Number of companies

U.S. and Canda 46 (26%) 8 (4%) 14 (8%) 17 (10%) 16 (9%) 16 (9%) 61 (34%) 178
Europe 4 (6%) 1 (2%) 1 (2%) 2 (3%) 7 (11%) 9 (14%) 42 (64%) 66
Asia-Pacific 40 (12%) 11 (3%) 10 (3%) 15 (5%) 26 (8%) 18 (5%) 208 (63%) 328
Africa and Middle East 11 (9%) 4 (3%) 11 (9%) 12 (10%) 13 (10%) 17 (14%) 56 (45%) 124

9
A. Mohammed Atris Resources Policy 65 (2020) 101543

Fig. 2. Average adjusted efficiency scores of the four regions.

Europe. The high performance in North America is due to the shale It is working on maintaining its capacity and enhancing its refineries’
boom, which changed the position of the U.S. in the global oil market, throughput via joint ventures. For instance, Saudi Aramco closed the
particularly in the petrochemicals industry. Petrochemical companies Jeddah refinery in 2017 to enhance its domestic business. The company
and gas investors in the U.S. and Canada have invested greatly in re­ is working to expand its downstream business overseas through joint
fineries and natural gas plants, as well as the extraction and processing ventures in emerging markets in Asia (e.g., China, Malaysia, and
of natural gas liquids (e.g., ethane, pentane, propane, butane). More­ Indonesia); it is also expanding in the U.S. by consolidating operations at
over, the low cost of ethane in the U.S. enhances the efficiency of U.S. the Motiva plant (600 kb/d) in a joint venture with Shell. Saudi Aramco
refineries and gives U.S. producers a competitive advantage. currently refines only 30% of its crude oil production; it aims to double
Further, U.S. refineries exceeded 17 md/d in 2017, with a 92% ca­ its refining capacity and increase its crude productivity. To this end,
pacity utilization rate. Most of this capacity comes from the Gulf Coast Saudi Aramco and SABIC petrochemical company signed an agreement
region, and it is expected to promote U.S. capacity over the next few to construct a 400 kb/d crude-producing chemical complex by 2025.
years. The increase in U.S. crude oil output, along with the inflows of Other Middle Eastern countries, such as Bahrain, Kuwait, and Oman, are
discounted Canadian crude oil, are the main drivers of the change in the also pursuing overseas expansion. The UEA has been less active in
U.S. position in the global oil market. expanding its capacity. However, ADNOC is starting main upgrades to
The Asia-Pacific region is endeavoring to boost the productivity and refine sour crude through its Ruwais refinery.
efficiency of its O&G refineries and reserves to meet increasing energy The Middle East is expected to attain the world’s highest crude
demand. China, India, and some parts of Thailand, Indonesia, and productivity by 2023. Africa has a low average utilization rate, of under
Malaysia have mature O&G reserves. Emerging markets such as the 70%. However, it is working to increase its refinery capacity. Nigeria
Philippines and Myanmar have begun developing new oil fields. The will drive capacity growth in Africa and the Middle East by adding
main O&G industry players in the Asia-Pacific region are divided into 1.933 md/d by 2022.9 The worst performance is seen in Europe due to
national oil companies (NOCs) and international oil companies (IOCs). the changes in their market patterns caused by the switchover to electric
NOCs include the China National Offshore Oil Corporation (CNOOC), vehicles and buses and because Europe is moving toward a low-carbon
India’s Oil and Natural Gas Corporation (ONGC), Malaysia’s Petronas, economy and the use of fuel-mix energy.
Thailand’s PTT, and Vietnam’s PetroVietnam. NOCs are leading the Refineries’ margins and fuel oil production are susceptible to oil
investment and development of the regional O&G industry. Meanwhile, price fluctuations, as shown in Fig. 2. Average adjusted efficiency in all
IOCs such as Exxon Mobil, BP, Chevron, Shell, and Murphy are helping regions increased in 2015 due to declining oil prices in 2014, which
to expand the Asia-Pacific O&G market through many joint ventures. affected all refineries’ margins and profits due to the increased demand
For instance, Papua New Guinea’s InterOil was acquired by Exxon for their products. However, in 2016, all regions recorded a drastic
Mobil, BP plans to expand its Tangguh LNG project in Indonesia, and decrease in average adjusted efficiency due to OPEC’s decision to cut
Exxon Mobil and Chevron are investing in oil ventures in Kazakhstan to their oil production to increase oil prices.
transport crude oil to China. Other IOCs and independents are also Table 5 presents the Kruskal–Wallis rank sum test results for the
participating in O&G market activities, including Reliance (one of In­ average adjusted efficiency scores. Test 1 investigates the null hypoth­
dia’s largest private sector companies), Murphy, and Shell. esis (H0) that the average adjusted efficiency measures are uniformly
Africa and the Middle East region (particularly the Middle East) is distributed among the four regions, while test 2 examines the null hy­
increasing its refinery capacity and investing more in technology to pothesis (H0) that the average adjusted efficiency scores are uniformly
integrate its upstream and downstream sectors.8 Saudi Arabia is a sig­ distributed over the 10 years analyzed in this study. The test statistics (H
nificant oil-producing nation not only in the region but also in the world. statistics) are 14.847 for test 1 and 3.902 for test 2, with 3 and 9 degrees
of freedom (df) respectively. The critical values for the statistics are

8
Sueyoshi and Wang (2014), and Atris and Goto (2019) discussed the role of
9
supply chain and vertical structure in enhancing the overall efficiency in the See [Link]
O&G industry. a-will-propel-refining-capacity-growth-in-the-middle-east-and-africa/.

10
A. Mohammed Atris Resources Policy 65 (2020) 101543

Table 5 Table 7
Kruskal-wallis rank sum test. Wilcoxon rank sum test of Europe.
df H statistics Critical value P value Region W-Test The Asia- Africa and the Middle U.S. and
Pacific East Canada
Test 1 3 14.340 7.815 0.002
Test 2 9 6.918 16.919 0.646 Europe Z 0.1510 0.7560 3.0990
P. 0.8798 0.4497 0.0019
Value
7.815 and 16.919, respectively, at the 5% significance level. The null
hypotheses are thus rejected for test 1 but not for test 2. In other words,
the average adjusted efficiency did not uniformly distribute among the Table 8
four regions (the adjusted efficiencies were, on average, different among Wilcoxon rank sum test of Africa & the middle east.
the four regions). Contrariwise, the average adjusted efficiency was
Region W-Test The Asia- Europe U.S. and
uniformly distributed over the 10 years from 2008 to 2017 (there was no Pacific Canada
difference across the 10 years).
Africa and the Middle Z 0.7560 0.7560 2.8730
Table 6 presents the Wilcoxon rank sum test results, separately, be­ East P. 0.4497 0.4497 0.0041
tween the U.S. and Canada and the other three regions. The test in­ value
vestigates the null hypothesis (H0) that the mean of the average adjusted
efficiency level between the U.S. and Canada and the other regions is
equal. The results indicate that the Z statistics between the U.S. and the Asia-Pacific region on one hand and Africa and the Middle East, the
Canada on one hand and the Asia-Pacific region, Europe, and Africa and EU, and the U.S. and Canada on the other are 0.7560, 0.1510, and
the Middle East on the other are 3.0240, 2.8730, and 3.0990, 3.0240, respectively, and their p-values are 0.4497, 0.8798, and
respectively, with p-values of 0.0025, 0.0041, and 0.0019, respectively, 0.0025, respectively, at a 5% significance level. Thus, we do not reject
at the 5% significance level. These results reject the null hypothesis (H0) the null hypothesis (H0) between the Asia-Pacific region and Africa and
that there is no statistically significant difference in the average adjusted the Middle East and Europe. This indicates that there is no statistically
efficiency mean between the U.S. and Canada and the other three significant difference in average adjusted efficiency mean between the
regions. Asia-Pacific region and Africa and the Middle East and Europe. There­
Table 7 shows the Wilcoxon rank sum test statistics between Europe fore, we reject (H0) between the Asia-Pacific region and the U.S. and
and the other three regions. The test investigates the null hypothesis Canada (see Table 6).
(H0) that there is no difference in means average adjusted efficiency
levels between any two regions. The critical values (Z) between Europe 5. Conclusion
and the Asia-Pacific region, Africa and the Middle East, and the U.S. and
Canada are 0.1510, 0.7560, and 3.0990, respectively, and their p- This study examined the efficiency of O&G refineries in four global
values are 0.8798, 0.4497, and 0.0019, respectively, at the 5% signifi­ regions—the U.S. and Canada, the Asia-Pacific region, Africa and the
cance level. Consequently, we do not reject the null hypothesis (H0) that Middle East, and Europe—from 2008 to 2017 using a unique unbalanced
there is no statistically significant difference in average adjusted effi­ dataset comprised of 696 refineries operating in the four regions. The
ciency mean between Europe on one hand and the Asia-Pacific region study utilized the averages of the refineries’ adjusted efficiencies to
and Africa and the Middle East on the other. Therefore, we reject the null simplify the discussion because the dataset was unbalanced and the
hypothesis regarding the EU region and U.S. and Canada, as shown in sample was large. The study applied a combination of DEA and DEA-DA
Table 6. to provide an efficiency-based ranking of the O&G refineries.
Table 8 shows the Wilcoxon rank sum results of the test between A Kruskal–Wallis rank sum test was used to test whether the average
Africa and the Middle East and the other three regions. The test in­ adjusted efficiencies measures differed among the four regions and if
vestigates the null hypothesis (H0) that there is no difference in mean they changed over time. Furthermore, a Wilcoxon rank sum test was
average adjusted efficiency levels between any two regions. The critical utilized to investigate whether the average adjusted efficiency levels
values (Z) between Africa and the Middle on one hand and the Asia- differed between any of the two regions over the study’s 10-year sample
Pacific region, Europe, and the U.S. and Canada on the other are period.
0.7560, 0.7560, and 2.8730, respectively, and their p-values are The results indicate that the U.S. and Canada outperform the other
0.4497, 0.4497, and 0.0041, respectively, at the 5% level of significance. three regions; this is followed by the Asia-Pacific region, Africa and the
As a result, we do not reject the null hypothesis (H0) between Africa and Middle East, and Europe, in that order. The study also found a statisti­
the Middle East and the Asia-Pacific region and Europe. This indicates cally significant difference among the four regions. Moreover, the
that there is no statistically significant difference in average adjusted average adjusted efficiency ranks were invariant from 2008 to 2017.
efficiency mean between Africa and the Middle East and the Asia–Pacific This result might be due to the differences between the refineries’ types,
region and Europe. Therefore, we reject (H0) between Africa and the technological complexity levels, and capacities: The U.S. has many
Middle East and the U.S. and Canada (see Table 6). advanced technologies available with which to extract light tight oil
Table 9 shows the Wilcoxon rank sum results for the test between the (LTO) as well as restrictive environmental regulations, which give U.S.
Asia-Pacific region and the other three regions. The test investigates the refineries a competitive advantage in adapting to global market changes
null hypothesis (H0) that there is no difference in mean adjusted effi­ and requirements.
ciency level between any two regions. The critical values (Z) between Furthermore, the operations of integrated U.S. oil companies (such
as Majors) boost the region’s global position through the many joint

Table 6 Table 9
Wilcoxon rank sum test of U.S. and Canada region. Wilcoxon rank sum test of the Asia-Pacific.
Region W-Test The Asia- Africa and the Middle Europe Region W-Test Africa and the Middle Europe U.S. and
Pacific East East Canada

U.S. and Z 3.0240 2.8730 3.0990 The Asia- Z 0.7560 0.1510 3.0240
Canada P. 0.0025 0.0041 0.0019 Pacific P. 0.4497 0.8798 0.0025
Value Value

11
A. Mohammed Atris Resources Policy 65 (2020) 101543

ventures being conducted with other countries. Vertical integration is a Acknowledgment

crucial factor in enhancing refinery efficiency because it reduces risks
and increases profitability at every stage, from the wellhead to the The author gratefully thanks prof. Mika Goto for her support.
gasoline station. Vertical integration and joint ventures also help re­
fineries achieve a balance between conducting operations and protect­ References
ing themselves against market instability.
A 2018 oil market report10 stated that the U.S. was expected to be a Al-Najjar, S.M., Al-Jaybajy, M.A., 2012. Application of data envelopment analysis to
measure the technical efficiency of oil refineries: a case study. Int. J. Bus. Adm. 3 (5),
significant player in massive integrated petrochemical projects for pro­ 64–77.
ducing ethylene. This will help the U.S. to not only cover the growth in Atris, A.M., Goto, M., 2019. Vertical structure and efficiency assessment of the US oil and
its domestic market but also export to other markets. Further, the U.S. gas companies. Resour. Policy 36, 101437.
Azadeh, A., Roohani, A., Motevali Haghighi, S., 2015. Performance optimization of gas
and Canada will become the leading suppliers of crude oil over the next refineries by ANN and DEA based on financial and operational factors. World. J. Eng.
five years. The U.S. will export LTO to Europe to address its deficit of 12 (2), 109–134.
African light oil and also to the Asia-Pacific region to satisfy the market Azadeh, A., Salehi, V., Mirzayi, M., Roudi, E., 2017. Combinatorial optimization of
resilience engineering and organizational factors in a gas refinery by a unique
demand for petrochemical raw material. This illustrates the vital role of mathematical programming approach. Hum. Factors Ergon. Manuf. 27, 53–65.
vertical integration and resource proximity in refinery efficiency. Bandyopadhyay, R., Alkilde, O.F., Menjon, I., Meyland, L.H., Sahlertz, I.V., 2019.
The Asia-Pacific region shows the highest O&G capacity and activity Statistical analysis of variation of economic parameters affecting different
configurations of diesel hydrotreating unit. Energy 183, 702–715.
levels, as well as increasing adjusted efficiency trends. As a result, the
Bevilacqua, M., Braglia, M., 2002. Environmental efficiency analysis for ENI oil
Asia-Pacific region has become an attractive market for major oil com­ refineries. J. Clean. Prod. 10, 85–92.
panies, which are increasing their investments by engaging in many Cerd� a, J., Pautasso, P.C., Diego, C.C., 2018. Optimization approaches for efficient crude
joint ventures, particularly in petrochemical plants. blending in large oil refineries. Ind. Eng. Chem. Res. 57, 8484–8501.
Chan Oh, S., Hildreth, J.A., 2014. Estimating the technical improvement of energy
Africa and the Middle East have also started to integrate their re­ efficiency in the automotive industry—stochastic and deterministic frontier
fineries through many joint ventures in order to generate more complex benchmarking approaches. Energies 7, 6196–6222.
refining and petrochemical products domestically. That implies the Chowdhury, H., Zelenyuk, V., Laporte, A., Wodchis, W.P., 2014. Analysis of productivity,
efficiency and technological changes in hospital services in Ontario: how does case-
importance of petrochemical and complex refining products, which are mix matter? Int. J. Prod. Econ. 150, 74–82.
considered key to gaining more global market share, thereby enhancing F€are, R., Grosskopf, S., Hernandez-Sancho, F., 2004. Environmental performance: an
refining efficiency. index number approach. Resour. Energy Econ. 26, 343–352.
Feng, C., Huang, J.B., Wang, M., 2018. Analysis of green total-factor productivity in
Despite the expected refinery shutdown waves in Europe due to the China’s regional metal industry: T A meta-frontier approach. Resour. Policy 58,
IMO regulation in 2020, the demand for refinery products slightly 219–229.
increased in 2017 because of the global trend toward high-consuming Feng, C., Huang, J.B., Wang, M., 2019. The sustainability of China’s metal industries:
features, challenges and T future focuses. Resour. Policy 60, 215–224.
vehicles (SUVs). This increase suggests that changes in demand pat­ Francisco, C., Almeida, M., Silva, D., 2012. Efficiency in Brazilian refineries under
terns affect refinery products and that Europe should invest in tech­ different DEA technologies regular paper. Int. J. Eng. Bus. Manag. 4 (35), 1–11.
nology in order to overcome their resource limitations. Glover, F., Sueyoshi, T., 2009. Contributions of professor William W. Cooper in
operations research and management science. Eur. J. Oper. Res. 197, 1–16.
Based upon the results, various management/policy recommenda­
Hollander, M., Wolfe, D.A., 1999. Nonparametric Statistical Methods, second ed. John
tions were proposed to enhance the refineries’ operational efficiency in Wiley & Sons, Inc., New York.
each region. These include (1) Processing of incremental amounts of Hosseinzadeh, A., Smyth, R., Valadkhani, A., Le, V., 2016. Analyzing the efficiency
domestic LTO by U.S. refiners is likely to have a positive impact on re­ performance of major Australian mining companies using bootstrap data
envelopment analysis. Econ. Modell. 57, 26–35.
finery efficiency. However, the U.S. needs to import different qualities of Hsu, Y.C., 2013. The efficiency of government spending on health: evidence from Europe
crude oil to maximize its throughput, considering its mix of refining and Central Asia. Soc. Sci. J. 50, 665–673.
capacity. Hence, it would be uneconomic to run refineries only with Korotin, V., Popov, V., Tolokonsky, A., Islamov, R., Ulchenkov, A., 2017. A multi-criteria
approach to selecting an optimal portfolio of refinery upgrade projects under margin
light crude oil; (2) Adjustment of refinery structure and control the scale and tax regime uncertainty. Omega 72, 50–58.
of industry is required to improve the scale efficiency of refined and Li, H., Dong, K., Sun, R., Yu, J., Xu, J., 2017. Sustainability assessment of refining
petrochemical products, for instance, the Asia-Pacific region has to enterprises using a DEA-based model. Sustainability 9 (620), 1–15.
Li, Y., Chiu, Y.H., Lin, T.Y., 2019. Coal production efficiency and land destruction in
adjust their refineries capacity to adapt rising in petrochemical demand China’s coal mining industry. Resour. Policy 63, 101449.
leads by China, India and Southeast Asia, particularly, in chemical Ma, D., Fei, R., Yu, Y., 2019. How government regulation impacts on energy and CO2
products; (3) The future for oil producers lies in developing a value- emissions performance in China’s mining industry. Resour. Policy 62, 652–663.
Managi, S., Opaluch, J.J., Jin, D., Grigalunas, T.A., 2004. Technological change and
added sector of refining and petrochemicals. Therefore, the Middle depletion in offshore oil and gas. J. Environ. Econ. Manag. 47, 388–409.
East region has to engage in many joint ventures through intensive Mann, H.B., Whitney, D.R., 1947. On a test of whether one of two random variables is
capital investment to expand into the downstream sector and to refine its stochastically larger than the other. Ann. Math. Stat. 18, 50–60.
Mardani, A., Streimikiene, D., Balezentis, T., Saman, M., Nor, K., Khoshnava, S., 2018.
crude oil domestically, which creates more value from their crude oil
Data envelopment analysis in energy and environmental economics: an overview of
throughput and diversify the region’s economies; (4) The Asia-Pacific the state-of-the-art and recent development trends. Energies 11, 1–21, 2002.
has to accelerate the application of new technologies to the refineries. Mekaroonreung, M., Johnson, A.L., 2010. Estimating the efficiency of American
Whereas, turning heavy oil into high-quality products also requires more petroleum refineries under varying assumptions of the disposability of bad outputs.
Int. J. Energy Sect. Manag. 4 (3), 356–398, 2010.
advanced molecular processing than is possible with simple refining or !"count(.//sb:host[1]//child::*//sb:date)">Roman, L., Zuzana, H., Henrieta, P., .
distillation; (5) European refineries have to launch joint ventures or Efficiency of mining and quarrying industry of V4 countries: the impact of
cooperate with electric vehicle (EV) producers to offset their losses investments and selected indicators. Acta Montan. Slovaca 22 (2), 136–145.
Simar, L., Wilson, P.W., 1998. Sensitivity analysis of efficiency scores: how to bootstrap
incurred when converting to other energy resources; (6) Europe region in nonparametric frontier models. Manag. Sci. 44 (11), 49–61.
should not add more costs on the refining sector and stimulates R&D to Simar, L., Wilson, P.W., 2000. Statistical inference in nonparametric frontier models: the
unlock innovation to develop the low-carbon technologies for refineries state of the art. J. Prod. Anal. 13 (1), 49–78.
Song, J., Zhang, Z., 2009. Oil Refining Enterprise Performance Evaluation Based on DEA
and their products. and SVM. IEEE, Second International Workshop on Knowledge Discovery and Data
One limitation of this study is that it does not examine the effects of Mining, pp. 401–404.
M&A policies on the global refining market. This issue will be discussed Song, Y., Huang, J.B., Feng, C., 2018. Decomposition of energy-related CO2 emissions in
China’s iron and steel T industry: a comprehensive decomposition framework.
in future research as an extension of this study. Resour. Policy 59, 103–116.
Sueyoshi, T., 2000. Stochastic DEA for restructure strategy: an application to a Japanese
petroleum company. Omega 28, 385–398.
10
See the International Energy Agency ([Link]
ort-series-oil-2018).

12
A. Mohammed Atris Resources Policy 65 (2020) 101543

Sueyoshi, T., Goto, M., 2012. Efficiency-based rank assessment for electric power Yang, W., Li, L., 2017. Efficiency evaluation and policy analysis of industrial wastewater
industry: a combined use of Data Envelopment Analysis (DEA) and DEA- control in China. Energies 10 (1201), 1–18.
Discriminant Analysis (DA). Energy Econ. 34, 634–644. Zhang, J., Zhu, X.X., Towler, G.P., 2001. A simultaneous optimization strategy for overall
Sueyoshi, T., Wang, D., 2014. Sustainability development for supply chain management integration in refinery planning. Ind. Eng. Chem. Res. 40 (12), 2640–2653.
in U.S. petroleum industry by DEA environmental assessment. Energy Econ. 46, Zhao, H., Ierapetritou, M.G., Shah, N.K., Rong, G., 2017. Integrated model of refining and
360–374. petrochemical plant for enterprise-wide optimization. Comput. Chem. Eng. 97,
Sueyoshi, T., Yuan, Y., Goto, M., 2017. A literature study for DEA applied to energy and 194–207.
environment. Energy Econ. 62, 104–124. Zhu, X., Chen, Y., Feng, C., 2018. Green total factor productivity of China’s mining and
Wang, W., Zhang, C., 2018. Evaluation of relative technological innovation capability: quarrying industry: a global data envelopment analysis. Resour. Policy 57, 1–9.
model and case study for China’s coal mine. Resour. Policy 58, 144–149.

13