 Research
 Open Access
 Published:
Network DEA based on DEAratio
Financial Innovation volume 7, Article number: 73 (2021)
Abstract
Data envelopment analysis (DEA) is a technique to measure the performance of decisionmaking units (DMUs). Conventional DEA treats DMUs as black boxes and the internal structure of DMUs is ignored. Twostage DEA models are special case network DEA models that explore the internal structures of DMUs. Most often, one output cannot be produced by certain input data and/or the data may be expressed as ratio output/input. In these cases, traditional twostage DEA models can no longer be used. To deal with these situations, we applied DEARatio (DEAR) to evaluate twostage DMUs instead of traditional DEA. To this end, we developed two novel DEAR models, namely, range directional DEAR (RDDR) and (weighted) Tchebycheff norm DEAR (TNDR). The validity and reliability of our proposed approaches are shown by some examples. The Taiwanese nonlife insurance companies are revisited using these proposed approaches and the results from the proposed methods are compared with those from some other methods.
Introduction
Data envelopment analysis (DEA) is an approach used to measure the relative efficiency of a set of DMUs with multiple inputs and outputs, first introduced into the operations research and management science literature by Charnes et al. (1978). DEA has been used in various environments and numerous applications (Kou et al. 2021; Zha et al. 2020; Castelli et al. 2004 and references therein). The traditional DEA considers each DMUs as black boxes consuming some inputs to produce some outputs without regarding the internal structure. In realworld problems, DMUs may have a network or internal structures; see, for example Färe and Grosskopf (1996) (who introduced the concept of network DEA for the first time), Castelli et al. (2004), Tone and Tsutsui (2009) and Guo et al. (2017). Kao (2014) presented a review of network DEA. In some cases, DMUs may consist of twostage network structures where the outputs of the first stage, known as intermediate measures/ outputs/ products, are inputs to the second stage. Many authors have studied twostage network DEA. Kao and Hwang (2008) proposed a multiplicative efficiency aggregation approach in which the overall efficiency of the twostage process is expressed as the product of the efficiency of two individual stages. Chen et al. (2009) revealed that Kao and Hwang’s (2008) twostage DEA model assumed constant returns to scale (CRS) and did not apply the variable returns to scale (VRS) assumption. So, they developed an additive efficiency decomposition approach in which the overall efficiency is expressed as a weighted average of the efficiency of the individual stages under VRS technology. Despotis et al. (2016) showed that the additive decomposition approach proposed by Chen et al. (2009) is biased toward the second stage and presents a composition approach to estimate unbiased efficiency scores for the individual stages.
Chen and Zhu (2004) offered a twostage DEA model and indicated that the units with individual stages were efficient overall. However, using an example, they pointed out that overall efficient units do not necessarily indicate efficient performance in the two stages. Wang and Chin (2010) defined the overall efficiency of twostage DMU under evaluation as the weighted harmonic mean of the efficiencies of the twostage DMU in each stage. They also generalized the additive efficiency decomposition model of Chen et al. (2009) to taking into consideration the relative importance weights of two individual stages (see models (24), (26) and (28) of Wang and Chin 2010). Zha and Liang (2010) presented a method for studying a twostage production process in which initial inputs of a twostage DMU are freely allocated in both stages. However, their method does not allow for the existence of shared intermediate products or additional direct inputs to be used in the second stage. Yu and Shi (2014) improved the method of Zha and Liang (2010) and proposed a twostage DEA parametric model where part of the outputs of the first stage are used as inputs in the second stage and additional inputs are allocated in the second stage but, does not allow for final outputs to be produced directly in the first stage or shared inputs. To overcome the aforementioned problems, Izadikhah and Farzipoor (2016) considered a twostage DEA model in which initial inputs of a twostage DMU are freely allocated in both stages and additional direct inputs are used in the second stage. Lozano et al. (2013) proposed a directional distance approach to deal with undesirable outputs. Lozano (2015a, 2015b) proposed slackbased measures (SBM) models for general network structures in which the exogenous inputs and outputs are considered at the system level instead of the process level. This model also relaxed the constraints for both the fixedlink and the freelink cases, thus enhanced the discriminating power of the model.
DEA models usually deal with data as absolute numeric values, while in the real world there are cases where data are ratios, for example, in efficiency measurement of financial institutions where financial ratios are included as output variables. There are two types of ratio data. In the first category, the DMU_{j} is considered as follows: \({\textit{DMU}}_j=(X_j^V, X_j^R, Y_j^V, Y_j^R)\) in which \(X_j^V\) and \(Y_j^V\) are the input and output components with absolute numeric values and are nonratio, respectively. \(X_j^R\) and \(Y_j^R\) are the input and output components with ratio scores, respectively. Therefore, in the first category, some input and output components are ratios and others are nonratio measures. Emrouznejad and Amin (2009), Olesen et al. (2015), Olesen et al. (2017) and HatamiMarbini and Toloo (2019) are examples of the first category. In the second category, the inputs and outputs of the \({\textit{DMU}}_j\) are as follows: \({\textit{DMU}}_j=(X_j^V, Y_j^V)\) where \(X_j^V\) and \(Y_j^V\) are nonratio input and output components which have absolute values. But, the ratio \(\frac{Y_j^V}{X_j^V}\) or \(\frac{X_j^V}{ Y_j^V}\) are defined. Despic et al. (2007), Liu et al. (2011), Gerami et al. (2020) and Gerami et al. (2020) are examples of the second category.
Emrouznejad and Amin (2009) proposed DEA models for dealing with ratio data. In turn, Olesen et al. (2015) showed that the use of ratio inputs and outputs in the variable returnstoscale (VRS) and constant returnstoscale (CRS) models generally violates the stated production assumptions. They developed new RatioCRS (RCRS) and RatioVRS (RVRS) models that allow the incorporation of ratio inputs and outputs. In the following, Olesen et al. (2017) developed radial and nonradial models under variable and constant returnstoscale technologies. HatamiMarbini and Toloo (2019) showed several problems in both solutions extended by Emrouznejad and Amin (2009) and introduced modified envelopment and multiplier DEA models for measuring performance to avoid the problems associated with Emrouznejad and Amin (2009). Despic et al. (2007) combined DEA methodology and Ratio Analysis (FernandezCastro and Smith 1994) and introduced the DEARatio (DEAR) model. In DEAR all possible ratios “output/input” are treated as outputs within the standard DEA model. The DEAR is an approach to apply expert’s opinions in performance evaluationn of DMUs. For example, if a certain output cannot be produced by a certain input, then, the corresponding ratio “output/input” can be deleted from the model. Also, if the ratio data is important to managers, traditional twostage DEA models are not applicable. Tohidnia and Tohidi (2019) applied DEAR for evaluating the efficiency and productivity change of DMUs over time. For more discussions on the advantage of DEAR see Mozaffari et al. (2020), Ostovan et al. (2020), Kamyab et al. (2021), Sotiros et al. (2019), Sexton and Lewis (2003), Zhang et al. (2015) and Gerami and Mozaffari (2013).
Considering the advantage of DEAR models over the CCRbased models, we evaluate the efficiency of twostage DMUs by DEAR models. To this end, we combine DEAR and twostage DEA methodologies and propose two DEA ratio models to evaluate overall ratio efficiency and individual stage ratio efficiencies of twostage DMUs. These two models are called Range Directional DEAR (RDDR) and (weighted) Tchebycheff norm DEAR (TNDR) models. We state and prove some facts about these two models. In the TNDR model, the Decision Maker can impose the preference of each stage over the other stages and project inefficient divided data on the Pareto front by selecting convenience weights. Additionally, RDDR model is translation invariant and unit invariant and projects all divided DMUs under evaluation on the strong frontier of production possibility set, under some conditions (see proposition 6. Moreover, we compare the RDDR and TNDR models with the proposed models in Chen et al. (2009), Despotis et al. (2016) (models (21) and (24) of Despotis et al. 2016) and Wang and Chin (2010) (models (24), (26) and (28) with (\(\lambda _1=\frac{2}{5}\), \(\lambda _2=\frac{3}{5}\))) by some examples in order to show the validity and reliability of the proposed RDDR and TNDR models and “twostage DEA based on DEAR.”
This paper is organized as follows. “Literature review” section presents the literature review on twostage DEA and ratio analysis. “Proposed methodology” section develops two models based on DEAR to evaluate overall ratio efficiency and individual stages ratio efficiency of each twostage DMUs under evaluation. Finally, “Numerical examples” section concludes the paper.
Literature review
Twostage DEA
Färe and Whittaker (1995) and Färe and Grosskopf (1996) used an input oriented twostage network DEA model to measure relative efficiencies in dairy production processes. Seiford and Zhu (1999) divided production process into independent subprocesses and calculated the efficiencies of the first stage, the second stage and the overall efficiency via three independent DEA models. Färe and Grosskopf (2000) presented a network DEA model for assessing Swedish Institute for Health Economics. Wang et al. (1997) and Noulas et al. (2001) proposed applications of twostage DEA to nonlife insurance policies and information technology, respectively. For an indepth review of the multistage DEA model, see Castelli et al. (2004). Figure 1 indicates a simple twostage production process where the first stage uses inputs \(x_i\), \((i=1,\ldots , m)\) to produce outputs \(z_d\), \((d=1,\ldots ,D)\) and then stage 2 uses these \(z_d\) as inputs to produce final outputs \(y_r\), \((r=1,\ldots ,s)\). In fact, the intermediate measures \(z_d\) are outputs of stage 1 and inputs of stage 2. The first and the second stage efficiencies of \({\textit{DMU}}_o, (o=1,\ldots , n)\), are defined as \(\theta _o^1=\frac{\sum _{d=1}^{D}w^1_d z_{do}}{\sum _{r=1}^{s}v_i x_{io}}\) and \(\theta _o^2=\frac{\sum _{r=1}^{s}u_r y_{ro}}{\sum _{d=1}^{D}w^2_d z_{do}}\), respectively, where \(v_i\) \((i=1,\ldots , m)\) and \(w^1_d\) \((d=1,\ldots , D)\) are the input and output weights in the first stage and \(u_r\) \((r=1,\ldots , s)\) and \(w^2_d\) \((d=1,\ldots , D)\) are the input and output weights in the second stage. Kao and Hwang (2008) defined the overall efficiency of \({\textit{DMU}}_o\) as the product of the two individual efficiencies, namely, \(\theta _o=\theta _o^1\times \theta _o^2\). Chen et al. (2009) found that the approach of Kao and Hwang (2008) cannot be extended to the VRS assumption (Banker et al. 1984) because, \(\theta _o=\left[ \frac{\sum _{d=1}^{D}w^1_d z_{do}w_o}{\sum _{r=1}^{s}v_i x_{io}}\right] \times \left[ \frac{\sum _{r=1}^{s}u_r y_{ro}u_o}{\sum _{d=1}^{D}w^2_d z_{do}}\right]\) could not be converted into a linear form under the condition of \(w^1_d=w^2_d\). They therefore stated the overall efficiency of \({\textit{DMU}}_o\) as a weighted sum of the efficiencies for the individual stages and proposed the following twostage DEA model under CRS assumption (model (11) of Chen et al. 2009):
Once the overall efficiency is obtained, the efficiency scores for the two individual stages of \({\textit{DMU}}_o\) can be determined. If the first stage efficiency is prior to the second stage, \(\theta _o^{1*}\) can be determined by solving the following LP model (model (18) of Chen et al. 2009):
The efficiency for the second stage is then calculated as \(\theta _o^2=\frac{\theta _o^{*}w_1^*.\theta _o^{1*}}{w_2^*}\), Also, by assuming preemptive priority for stage 2, \(\theta _o^{2*}\) can be determined by solving the following LP model (model (19) of Chen et al. 2009):
The efficiency for the first stage is then calculated as \(\theta _o^1=\frac{\theta _o^{*}w_2^*.\theta _o^{2*}}{w_1^*}\), where
by way of (1). If \(\theta _o^{1}=\theta _o^{1*}\) or \(\theta _o^{2}=\theta _o^{2*}\), then this indicates that we have a unique efficiency decomposition. Despotis et al. (2016) modeled the following singleobjective program under the constant returnstoscale (CRS) assumption; for assessments of the efficiencies of two individual stages and overall efficiency of the evaluated \({\textit{DMU}}_o\) (Model (21) of Despotis et al. 2016):
Once an optimal solution \((v^*, w^*, u^*)\) of the model (4) is obtained, the efficiency scores for \({\textit{DMU}}_o\) under assessments in the first and second stages are respectively:
Also, the overall efficiency of \({\textit{DMU}}_o\) is calculated as follows:
Moreover, they employed the unweighted Tchebycheff norm (\(L_\infty\) norm) to formulate following model (Model (24) in Despotis et al. 2016):
in which \(E_o^1\) and \(E_o^2\) indicate the efficiency scores of stage 1 and stage 2 and are calculated by model (20) of Despotis et al. (2016).
Once an optimal solution \((v^*, w^*, u^*)\) of the model (7) is obtained, the individual efficiency scores and overall efficiency for \({\textit{DMU}}_o\) under assessment are obtained by (5) and (6), respectively.
Wang and Chin (2010) generalized twostage DEA models proposed by Chen et al. (2009). They assigned specific weights \(\lambda _1>0\) and \(\lambda _2>0\), with \(\lambda _1+\lambda _2=1\), to each individual stage to reflect its relative importance in the whole process. They formulated the overall efficiency of \({\textit{DMU}}_o\) as follows (see model (24) of Wang and Chin 2010):
Once the overall efficiency \(\theta ^{*}_o\) is obtained, \(\theta ^{1*}_o\) and \(\theta ^{2*}_o\) can then be determined by solving following LP models (see models (26) and (28) of Wang and Chin 2010):
Ratio analysis
Consider a set of \({{\varvec{n}}}\) DMUs which is associated with \({{\varvec{m}}}\) inputs and \({{\varvec{s}}}\) outputs. Particularly, each \({\textit{DMU}}_{j}=(X_j,Y_j)\) \((j\in J=\{1,\ldots , n\})\) consumes amount \(x_{ij}(>0)\) of input \({{\varvec{i}}}\) and produces amount \(y_{rj}(>0)\) of output \({{\varvec{r}}}\). Corresponding to any \({\textit{DMU}}_j\), outputinput ratio vector \(V_j\) is defined as follows:
\(V_j\) can be considered as outputs vector i.e., the more \(\frac{y_{rj}}{x_{ij}}\), the better \(\frac{y_{rj}}{x_{ij}}\). Let \(V=\bigcup _{j=1}^{n}V_j\) be a group of outputinput ratio. Then, the smallest closed convex and freedisposal attainable set^{Footnote 1} that contains the observations can be expressed as follows: (see Liu et al. 2011 for detail)
FernandezCastro and Smith (1994) calculated the ratio efficiency of \({\textit{DMU}}_o\) by solving the following outputoriented model, for the first time:
or equivalently:
Proposition 1
At optimality of model (11), \(\varphi ^*_R\ge 1\).^{Footnote 2}
Wu and Liang (2005) used ratio analysis to develop an aggregated ratio model to evaluate \({\textit{DMU}}_o\) as follows:
Definition 1
Wu and Liang (2005) \({\textit{DMU}}_o\) is ratio efficient if and only if \(\psi ^*=1\).
Despic et al. (2007) referred to the model (12) as the DEAR model and showed that this model is equivalent to the outputoriented DEAR model (13):
Mozaffari et al. (2014) investigated the relationship between the DEAR models (13) and (12) and proved that \(\psi ^*=\frac{1}{\Delta ^*}\) on optimality of the models (12) and (13) (also see Despic et al. 2007). Moreover, they proved that the model (11) is dual of the model (13) and therefore, \(\varphi ^*_R=\Delta ^*\). So, we have the following property:
Proposition 2
At optimality of models (11) and (12), \(\varphi ^*_R=\frac{1}{\psi ^*}\).
Definition 2
Mozaffari et al. (2014) \({\textit{DMU}}_o\) is DEAR efficient if and only if the optimal objective function value of model (13), \(\Delta ^*=1\). In view of the above discussion, the ratio efficiency of \({\textit{DMU}}_o\) under evaluation can be obtained using the models (11) or (12) or (13).
The issue of underestimation of efficiency and pseudoinefficiency are two other problems that the traditional DEA models cannot overcome. Underestimation of efficiency occurs when input and output weights are computed as zero. Therefore, the importance of that variable is not considered in the efficiency score of the DMUs. In this case, the efficiency scores of DMUs are not calculated correctly. Different works have been reported in this field (see for instance Gerami et al. 2020, 2020). DEAR models prevent the underestimation of efficiency and pseudoinefficiency (Wei et al. 2011a, b, c). Therefore, they correctly calculate the efficiency score of DMUs. Moreover, in some cases, the available data are presented as ratio output/input. Also, it may occur that in producing an output only a subset of inputs is being used. In modeling such restrictions, traditional DEA is not applicable, and using the DEAR approach much easier than other proposed approaches (see Tohidnia and Tohidi 2019 for detail). The flexibility of DEAR in modeling such restrictions gave us the motivation to use DEAR for evaluating twostage DMUs. To this end, we combined DEAR and network DEA to propose two new models for evaluating twostage DMUs. In general, the proposed ratio network DEA models can be considered as a combination of the twostage DEA efficiency model and ratio analysis, and thus, they will be more preferred by experts who are familiar with ratio analysis.
Proposed methodology
As referred to in the “Introduction” section, DEAR is a suitable approach to incorporate expert”s opinions. Moreover, when just ratio data as “output/input” are available, DEAR is applicable. This shows the importance of applying DEAR to the twostage DEA. Because of these facts, we evaluated the efficiencies of twostage DMUs using the DEAR model (11) instead of conventional DEA models and proposed two models which we call (1)the Range Directional DEAR (RDDR) model and (2) (weighted) Tchebycheff norm DEAR (TNDR) model for measuring the individual stage ratio efficiencies and overall ratio efficiency of a twostage DMU under evaluation. Our proposed models obtain these ratio efficiencies by solving only one model. This is the other advantage of our proposed models in comparison with some other existing twostage models.
Let \(\left\{ {\textit{DMU}}_j=(X_j, Z_j, Y_j)j=1, \ldots , n\right\}\) be a group of twostage data; in which, \(X_j=(x_{1j}, \ldots , x_{mj})\), \(Z_j=(z_{1j}, \ldots , z_{Dj})\) and \(Y_j=(y_{1j}, \ldots , y_{sj})\), we define divided data \(\left\{ DMU'_j=\left( \frac{Z_j}{X_j}, \frac{Y_j}{Z_j}\right) j=1, \ldots , n\right\}\) where \(\frac{Z_j}{X_j}=\left( \frac{z_{1j}}{x_{1j}}, \ldots , \frac{z_{Dj}}{x_{1j}}, \ldots , \frac{z_{1j}}{x_{mj}}, \ldots , \frac{z_{Dj}}{x_{mj}}\right)\), \(\frac{Y_j}{Z_j}=\left( \frac{y_{1j}}{z_{1j}}, \ldots , \frac{y_{sj}}{z_{1j}},\ldots , \frac{y_{1j}}{z_{Dj}}, \ldots , \frac{y_{sj}}{z_{Dj}}\right)\), for each \(i=1,\ldots ,m\), \(d=1,\ldots , D\) and \(r=1,\ldots , s\).
Again, we assume that the more \(\frac{z_{dj}}{x_{ij}}\) and \(\frac{y_{rj}}{z_{dj}}\) are, the more they are considered as outputs. So, we have a set of DMUs with \(D\times m+s\times D\) outputs without any explicit inputs.
Similar to the attainable set P, the following attainable set defines a bounded closed convex and freedisposal set that contains the observations:
The inequalities of \(P'\) are corresponding to stage 1 efficiency and stage 2 efficiency, respectively. In the sequel, we develop two models to estimate the overall ratio efficiency and individual stage ratio efficiencies of twostage DMUs.
Tchebycheff norm DEAR (TNDR) model
The weighted Tchebycheff norm turned out to be very useful in generating nondominated solutions (Pareto optimal solutions) in multiple objective programs (see Kou et al. 2014; Kasimbeyli 2010). The distance between two points \(X=(x_1, x_2,\ldots ,x_n)\) and \(X'=(x'_1, x'_2,\ldots ,x'_n)\) with \(L_\infty\) norm is given by:
The evaluation of the efficiency value of DMU with \(L_\infty\)Norm (Tchebycheff norm) was introduced by Tavares and Antunes (2001). They presented the following model named TCH in DEA for evaluation of the performance of \({\textit{DMU}}_k\):
where \(U_o\in [0, +\infty ]\) is the efficiency value. Also, \({\textit{DMU}}_o\) is efficient in the Tavares model (14) if and only if in optimal solution \(U^*_o=0\), otherwise it is inefficient. With the introducing efficiency index as \(\theta ^*_o=\frac{1}{1+U^*_o}\), \(\theta ^*_o\in (0, 1]\), \({\textit{DMU}}_o\) is efficient in the Tavares model (14) if and only if in optimal solution \(\theta ^*_o=1\).
The objective function of model (14) minimizes the distance between \({\textit{DMU}}_o\) and its projected point on frontier efficiency. Using the weighted Tchebycheff norm, the model (14) can be written as the following weighted programming problem:
where \(\mathbf{W}_1\in {\mathbb{R}}_{\ge 0}^m\), \(\mathbf{W}_2\in {\mathbb{R}}_{\ge 0}^s\) are Decision Maker preference and \(\mathbf{1W}_1+\mathbf{1W}_2=1\). Now, motivated by the weighted model (15), we introduce (weighted) Tchebycheff norm DEAR (TNDR) models for evaluating twostage \({\textit{DMU}}_o=(X_o, Z_o, Y_o),\) \(o\in \{1, 2, \ldots , n\},\) as follows:
where \(\mathbf{W}_1\in {\mathbb{R}}_{\ge 0}^m\), \(\mathbf{W}_2\in {\mathbb{R}}_{\ge 0}^s\) with \(\mathbf{1W}_1+\mathbf{1W}_2=1\), are weights addressing total preference over the two stages and are determined by the Decision Maker.
Suppose that \((\alpha ^*_1, \alpha ^*_2)\) are the values of the objective function of model (16) at optimality. The stage ratio efficiencies and overall ratio efficiency scores are defined as \(F=\frac{1}{1+\alpha ^*_1}\), \(S=\frac{1}{1+\alpha ^*_2}\) and \(O=F\times S\), respectively.
Proposition 3
If divided DMU \(\left( \frac{Z_o}{X_o},\frac{Y_o}{Z_o}\right)\) lies on the strong frontier of the PPS \(P'\) then, \({\textit{DMU}}_o=(X_o, Z_o, Y_o)\) is ratioefficient overall and individual stages are ratioefficient and vice versa.
Definition 3
Suppose that \((\alpha ^*_1, \alpha ^*_2)\) are the values of the objective function of model (16) at optimality. Then,

(a1)
if \(O=1\), \({\textit{DMU}}_o\) is said to be overall ratio efficient.

(a2)
if \(F=1\), \({\textit{DMU}}_o\) is said to be ratio efficient in stage 1.

(a3)
if \(S=1\), \({\textit{DMU}}_o\) is said to be ratio efficient in stage 2.

(a4)
if \(O<1\), \({\textit{DMU}}_o\) is said to be overall ratio inefficient.

(a5)
if \(F<1\), \({\textit{DMU}}_o\) is said to be ratio inefficient in stage 1.

(a6)
if \(S<1\), \({\textit{DMU}}_o\) is said to be ratio inefficient in stage 2.
Range directional DEAR (RDDR) model
Silva Portela et al. (2004) proposed range directional measure for handling the negative data based on the directional distance function approach which provides an efficiency score that results from the comparison of the DMU under evaluation with the socalled ideal point. Motivated by their approach, this paper defines Range Directional DEAR (RDDR) model as follows. Let \(\{Y_j j=1,\ldots ,n\}\), in which \(Y_j=(y_{1j},\ldots ,let y_{sj})\), be a group of data in \({\mathbb{R}}^s\). The range directional model without explicit inputs for assessing \({\textit{DMU}}_o=Y_o\) can be defined as follows:
or equivalently:
where \(\mathbf{R}_o^+=(R_{1o}^{+},\ldots , R_{so}^{+})^t\) is the vector between ideal point \(Y_M=\left( \underset{j=1,\ldots , n}{\max }\{y_{1j}\},\ldots , \underset{j=1,\ldots , n}{\max }\{y_{sj}\}\right) ^t\) and \({\textit{DMU}}_o=Y_o=(y_{1o}, \ldots , y_{so})\) under evaluation, that is, \(\mathbf{R}_o^+=Y_MY_o\) and \(R_{ro}^{+}=\underset{j=1,\ldots , n}{\max }\{y_{rj}\}y_{ro}\), \(r=1,\ldots , s\).
Proposition 4
Model (17) is always bounded, that is, \(0\le \beta ^*_o\le 1\). \({\textit{DMU}}_o\) is efficient if \(\beta ^*_o=0\). Therefore, the RDDefficiency measure of \({\textit{DMU}}_o\) is given by \(\rho ^*=1\beta ^*_o\).
By model (17), \({\textit{DMU}}_o=(y_{1o}, y_{2o},\ldots , y_{so})\) is projected onto the frontier of P along the direction of the vector \(\mathbf{R}_{o}^{+}\) at\(Y_o+\beta ^*_o R_{ro}^{+}\).
Model (17) also has following important characteristic that can be used to handle negative data, that is:
Proposition 5
Model (17) is translation invariant and unit invariant.
Proof
The proof is trivial. \(\hfill\square\)
The following proposition discusses the conditions that \({\textit{DMU}}_o\) under evaluation is projected onto the strong frontier of the attainable set P via Range Directional model (17).
Proposition 6
Suppose that the ideal point \(Y_M\) is constructed by s(=the dimension of \(R^s_+\)) extreme DMUs. If the vector of \(Y_M\) along with the vectors of these extreme efficient DMUs constitute an affine independent set then, \({\textit{DMU}}_o\) is projected onto the strong frontier (hyperplane) of attainable set P via model (17).
Proof
It is enough to show that the first r constraints of model (17) are as equality at optimality. Without loss of generality, suppose that \(Y_M\) is constructed by s extreme DMUs \(Y_1, Y_2, \ldots , Y_s\). By affinity independence of the vectors \(Y_1, Y_2, \ldots , Y_s\) and \(Y_M\), the set \(\{Y_1Y_M, Y_2Y_M, \ldots , Y_sY_M\}\) constitutes a basis for vector space \(R^s_+\). Therefore, there exist the constants \({\bar{\lambda }}_1, {\bar{\lambda }}_2,\ldots , {\bar{\lambda }}_s\) so that, \(\sum _{j=1}^{s}{\bar{\lambda }}_{j}(Y_MY_j)=Y_MY_o\), \({\bar{\lambda }}_{j}\ge 0\). (Note that defining \(Y_M\), the vector \(Y_MY_o\) lies in the positive linear combination of the vectors \(Y_MY_1, Y_MY_2,\ldots , Y_MY_s\)). Therefore, \(\sum _{j=1}^{s}{\bar{\lambda }}_{j}(Y_MY_j)=Y_MY_0, {\bar{\lambda }}_{j}\ge 0, j=1,\ldots , s\). This implies \(\sum _{j=1}^{s}{\bar{\mu }}_{j}Y_j=Y_o+\beta _oR_o^+(Y_MY_o)\), \(\sum _{j=1}^{s}{\bar{\mu }}_{j}=1\), where \(\beta _o=1\frac{1}{\sum _{j=1}^{s}{\bar{\lambda }}_{j}}\), \({\bar{\mu }}_{j}=\frac{{\bar{\lambda }}_{j}}{\sum _{j=1}^{s}{\bar{\lambda }}_{j}}\) and \(R_o^+=Y_MY_o\). So,
thus, \((\beta _o, {\bar{\mu }}_{j}, j=1,\ldots , s)\) is an optimal solution of (17). This completes the proof. \(\hfill\square\)
Note: By proposition 6, if attainable set P contains strong hyperplane(s) then, \({\textit{DMU}}_o\) is projected onto them, by model (17).
Now, inspired by model (17), Range Directional DEAR (RDDR) models for evaluation twostage \({\textit{DMU}}_o=(X_o, Z_o, Y_o)\) is introduced as follows:
where \(\mathbf{R}^{1+}_o=\frac{Z_M}{X_M}\frac{Z_o}{X_o}\) and \(\mathbf{R}^{2+}_o=\frac{Y_M}{Z_M}\frac{Y_o}{Z_o}\),
for each \(i=1,\ldots ,m\), \(d=1,\ldots , D\) and \(r=1,\ldots , s\). The notation \(\Vert R\Vert\) denotes the norm of vector R.
The RDDR model (18) projects the vector \(\frac{Z_o}{X_o}\) and \(\frac{Y_o}{Z_o}\) on the frontier of attainable set \(P'\) in direction of \(\mathbf{R}^{1+}_o\) and \(\mathbf{R}^{2+}_o\) and measure the ratioefficiency of stage 1 and stage 2 of the twostage DMU \({\textit{DMU}}_o\), respectively. Model (18) can be written as following single model:
where \(R^{1+}_{dio}=\frac{z_{dM}}{x_{iM}}\frac{z_{do}}{x_{io}}\) and \(R^{2+}_{rdo}=\frac{y_{rM}}{z_{dM}}\frac{y_{ro}}{z_{do}}\).
Proposition 7
The RDDR model (19)is always bounded that is, \(0\le \beta ^*_1\le 1\) and \(0\le \beta ^*_2\le 1\).^{Footnote 3}
Twostage \({\textit{DMU}}_o=(X_o, Y_o, Z_o)\) is ratioefficient if \(\beta ^*=0\). Therefore, the RDDratio efficiency measure of \({\textit{DMU}}_o\) is given by \(\rho ^*=1\beta ^*\).
So, \(\rho _1^*=1\beta ^*_1\), \(\rho _2^*=1\beta ^*_2\) and \(\rho _o^*=\rho _1^*\rho _2^*\) is said to be the RDDratio efficiency score of stage 1, RDDratio efficiency score of stage 2 and overall ratioefficiency score of the twostage \({\textit{DMU}}_o\), respectively.
Definition 4
Suppose that (\(\beta ^*_1\), \(\beta ^*_2\)) is the optimal solution of the model (19). Then,

(a1)
if \(\rho ^*_o=1\), \({\textit{DMU}}_o\) is said to be overall ratio efficient.

(a2)
if \(\rho ^*_1=1\), \({\textit{DMU}}_o\) is said to be ratio efficient in stage 1.

(a3)
if \(\rho ^*_2=1\), \({\textit{DMU}}_o\) is said to be ratio efficient in stage 2.

(a4)
if \(\rho ^*_o<1\), \({\textit{DMU}}_o\) is said to be overall ratio inefficient.

(a5)
if \(\rho ^*_1<1\), \({\textit{DMU}}_o\) is said to be ratio inefficient in stage 1.

(a6)
if \(\rho ^*_2<1\), \({\textit{DMU}}_o\) is said to be ratio inefficient in stages 2.
The following theorem states that individual stage efficiencies in the TNDR model (16) are equivalent to individual stage efficiencies in the RDDR model (19):
Theorem 1
At optimality of models (16) and (19), \(\alpha _1^*=0\) (\(\alpha _2^*=0\)) if and only if \(\beta _1^*=0\) (\(\beta _2^*=0\)).
Proof
The proof is straightforward.
The insurance industry extends the productivities and services by providing safety and confidence. These companies have positive effects on the growth of the economy of a country. Fecher et al. (1993) was the first to conduct a study that applies DEA in evaluating the performance of insurance firms. Thereafter, many researchers have studied the insurance business by using the DEA techniques (see Eling and Jia 2019; Tone et al. 2019; An et al. 2020, for instance). Kaffash et al. (2019) found that from 1992 to 2018 there were 132 studies on the insurance sector. Studies reviewed in their paper applied DEA to calculate the efficiency of firms from various backgrounds, with multiple inputs and outputs. There are other studies in the literature (Borges et al. 2008; Cummins and Weiss 2013; Ilyas Ashiq 2019; Mandal 2014), which have calculated the efficiency of insurance companies by adopting DEA as a tool for efficiency measurement. Hwang and Kao (2006) were the first to employ NetworkDEA to evaluate the performance of nonlife insurance firms in Taiwan. Thereafter, many authors have investigated the performance of insurance firms using the twostage DEA (see Kao 2014; Chen et al. 2009; Guo et al. 2017; An et al. 2020 and references therein). Most of the prior research used more traditional DEA models. A few recent studies have adopted new DEA models such as the NetworkDEA. However, to the best of our knowledge, there are only two studies that used the NetworkDEA and DEAR simultaneously in the insurance industry (see Gerami et al. 2020; Ostovan et al. 2020). In the next section, we apply the proposed methods to evaluate the performance of nonlife insurance firms in Taiwan also studied in Kao and Hwang (2008). \(\hfill\square\)
Numerical examples
Example 1
In this example we evaluate twostage DMUs that have been previously evaluated by Izadikhah et al. (2018). Consider 3 twostage DMUs with each DMU consuming a single input (X) to produce a single output (Z), in stage 1 and a single input (Z) to produce a single output (Y), in stage 2. Data set for these DMUs are given in Table 1. The last column of Table 1 represents divided DMUs corresponding to the twostage DMUs. Evidently, \((W_1, W_2)=\left( \frac{1}{2}, \frac{1}{2}\right)\).
Figure 2 depicts efficient frontier of PPS constructed by divided DMUs. Ratioefficiency scores of the proposed TNDR and RDDR models and efficiency scores of the Izadikhah’s method (Izadikhah et al. 2018) are reported in Table 2. Given Fig. 2 and the obtained efficiency scores, only DMU B is individual stages ratioefficient and overall ratioefficient by our proposed methods and individual stages efficient and overall efficient by Izadikhah’s method (Izadikhah et al. 2018). More importantly, these results show that proposed methods evaluate twostage DMUs in constant return to scale technology.
In view of Fig. 2 and the results of Table 2, the proposed TNDR and RDDR models are capable to compare the first (second) stage of each twostage DMUs with that of others and ranking them. For example, in the first stage, twostage DMU \(A=(2, 1.5, 1.5)\) produces 1.5 output using 2 input and twostage DMU \(B=(4, 4, 5)\) produces 4 output using 4 input. So, ratioefficiency of stage 1 of twostage DMU B must be better (more) than that of twostage DMU A. Table 2 confirms this fact. Also, the proposed models are capable to compare the overall efficiencies of twostage DMUs and ranking them correctly. For example, twostage DMU \(A=(2, 1.5, 1.5)\) produces 1.5 output using 2 input and twostage DMU \(B=(4, 4, 5)\) produces 5 output using 4 input. It is reasonable to expected that overall efficiency of DMU B is better (more) than that of DMU A. Table 2 confirms this fact.
Example 2
Consider 10 twostage DMUs with each DMUs consuming two inputs \((X_1, X_2)\) to produce a two output \((Z_1, Z_2)\) in stage 1 and consuming inputs \((Z_1, Z_2)\) to produce two outputs \((Y_1, Y_2)\) in stage 2. Data set for these DMUs are given in Table 3. Applying TNDR model (16) (with equal weights), RDDR model (19) and model (21) in Despotis et al. (2016) to each divided \({\textit{DMU}}_k\), \(k=1,\ldots , 10\), produces the results reported in Tables 4 and 5.
Similar to the example 1, this example shows the validity of proposed approaches. For example, twostage DMU6 and DMU7 produce outputs (3, 3) and (4, 4), respectively, using the same inputs (2, 2) in stage 1. It is reasonable to expect that the rank of DMU7 must be better than that of DMU6. Table 4 confirms this fact. The ranking obtained for individual stages and overall system of all DMUs by TNDR, RDDR models and Despotis et al. (2016) model (21) is shown in the 5th–7th columns and 11th–13th columns of Table 4 and 5th–7th columns of Table 5, respectively. More interestingly, the ranking results of our proposed methods and model (21) of Despotis et al. (2016) are very similar to each other. Especially, the ranking results of stages 1 and 2 of the TNDR and RDDR models and model (21) of Despotis et al. (2016) are almost the same. These results confirm the soundness of our model and its capacity of reliably solving the twostage DEA problem.
Example 3
We here apply our new approaches to the 24 Taiwanese nonlife insurance companies studied in Kao and Hwang (2008) where the inputs of stage 1 are Operation expenses (X1) and Insurance expenses (X2) and the output of stage 2 are Underwriting profit (Y1) and Investment profit (Y2). Also, the Direct written premiums (Z1) and Reinsurance premiums (Z2) are two intermediate measures, that is, the outputs of the stage 1 and the inputs to the stage 2. The data set appear in Table 6. In this example \(m=s=D=2\). Therefore, each divided data \({\textit{DMU}}_j=\left( \frac{Z_j}{X_j}, \frac{Y_j}{Z_j}\right) \in {\mathbb{R}}^4\times {\mathbb{R}}^4\). For such a twostage structure and the data set, the efficiency results from RDDR model (19) and TNDR model (16) are reported in Tables 7 and 8, respectively. As referred in Despic et al. (2007), DEAR is comparable with conventional DEA; we compare the results obtained by our proposed RDDR model (19) and TNDR model (16) and those by proposed methods in Chen et al. (2009) (models (18)–(19)), Despotis et al. (2016) (models (21) and (24) of Despotis et al. 2016) and Wang and Chin (2010) (models (19)–(21) and models (25), (27) and (29) with (\(\lambda _1=\frac{2}{5}\), \(\lambda _2=\frac{3}{5}\)) of Wang and Chin 2010) and show there are some interesting relationship between them. The ranking results obtained from RDDR model (19) for all the DMUs is shown in the 6th–8th columns of Table 7. Also, the ranking obtained from TNDR model (16) is shown in Table 8. For simplicity, all chosen Decision Maker’s preference (weights) \(W_j=(w_{1j}, w_{2j}, w_{3j}, w_{4j})\), \(j=1, 2\) in TNDR model (16) are given in Table 9. Table 10 shows the results obtained from Chen et al. (2009) (columns 2–5), Despotis et al. (2016) (columns 6–11) and Wang and Chin (2010) (columns 12–17). It is interesting to note that twostage DMUs 9, 12, 15, 19, 24 are ratioefficient in the first stage by our proposed models (for all weights in TNDR model) and all models in Table 10 (except for stage 1 of DMU 15 in Despotis et al. (2016) model (24)). Moreover, twostage DMUs 3, 5, 17, 22 are ratioefficient in the second stage by our proposed models (for all weights in TNDR model) and all models in Table (10). The ranking results for individual stages of all twostage DMUs are almost the same by our proposed models and all models in Table (10). But, there are some significant difference between the ranking results of our proposed models and those of the proposed models in overall in Table (10). Especially, DMU 24 is ranked as 24th DMU by Despotis et al. (2016) model (21) and 4th DMU by TNDR model (16). It may be due to roundoff error problems. Comparison of the results obtained from the traditional CCR model (Kao and Hwang 2008) from one side and our proposed models from the other side, shows the similarity between them. DMUs 9, 12, 19 and 24 are stage 1 efficient in the traditional CCR, TNDR and RDDR models. Also, DMUs 3, 5, 17 and 22 are stage 2 efficient in the CCR, TNDR and RDDR models. The above discussion shows the validity and reliability of the proposed RDDR and TNDR models.
Conclusion
In network DEA, if all data are expressed as ratio output/input or if in the production process an input is not used for producing a particular output or if data ratio is important to managers, traditional twostage DEA models can no longer be used. Flexibility of the DEAR in incorporating the mentioned restrictions and expert opinions to the DEA models gave us the motivation to apply DEAR for the evaluation of twostage DMUs. Also, the main contribution of this paper is to evaluate the twostage DMUs using DEAR to overcome the mentioned restrictions and shortfalls. To this end, we propose two novel models namely, RDDR and TNDR to calculate individual ratio efficiencies and overall ratio efficiency of twostage DMUs. Moreover, a decisionmaker can consider the preference of each stage over other stages by selecting weights in the TNDR model. Also, the RDDR model projects all divided \({\textit{DMU}}_o\) under evaluation on the strong frontier of production possibility set (see proposition 6). We compared our proposed methods with some existing methodologies to show validity and reliability of the of proposed approaches and “twostage DEA based on DEAR.” The results show that our approaches are similar to the existing approaches. We evaluated twostage DMUs using DEAR where all the outputs from the first stage are the only inputs to the second stage. Initial studies had shown that our approach can be applied with freelink, and fixedlink assumptions. Moreover, it can be extended to the more complicated twostage DEA models. For example, it can be extended to twostage DEA models with a shared resources and feedback and in case there are external inputs. This, however, needs further research and a deeper analysis.
Availability of data and material
Data used in this paper were extracted from Kao and Hwang, “Efficiency decomposition in twostage data envelopment analysis: an application to nonlife insurance companies in Taiwan,” European Journal of Operational Research; 2008 85(1), 418–429.
Notes
See Liu et al. (2011) for definition of attainable set.
Superscript ‘*’ indicates optimality.
(*) is used for the optimal solution.
Abbreviations
 CCR::

Charnes, Cooper and Rhodes
 DMU::

Decision making unit
 DEA::

Data envelopment analysis
 DEAR::

DEAratio
 RDDR::

Range directional DEAR
 TNDR::

Tchebycheff norm DEAR
 CRS::

Constant returns to scale
 VRS::

Variable returns to scale
 SBM::

Slackbased measure
 RCRS::

RatioCRS
 RVRS::

RatioVRS
 LP::

Linear programming
References
Allen R, Athanassopoulos A, Dyson RG, Thanassoulis E (1997) Weights restrictions and value judgments in data envelopment analysis: evolution development and future directions. Ann Oper Res 73:13–34
An Q, Ping W, Emrouznejad A, Junhua H (2020) Fixed cost allocation based on the principle of efficiency invariance in twostage systems. Eur J Oper Res 283(2):662–675
Ang S, Chen CM (2016) Pitfalls of decomposition weights in the additive multistage DEA model. Omega 58:139–153
Banker RD, Charnes A, Cooper WW (1984) Some models for the estimation of technical and scale inefficiencies in data envelopment analysis. Manage Sci 4:1078–1092
Borges MR, Nektarios M, Barros CP (2008) Analysing the efficiency of the Greek life insurance industry. Eur Res Stud 11(3):35–52
Castelli L, Pesenti R, Ukovich W (2004) DEAlike models for the efficiency evaluation of hierarchically structured units. Eur J Oper Res 154(2):465–476
Charnes A, Cooper WW, Rhodes E (1978) Measuring the efficiency of decisionmaking units. Eur J Oper Res 2:429–444
Charnes A, Cooper WW, Golany B, Seiford LM, Stutz J (1985) Foundations of data envelopment analysis for ParetoKoopmans efficient empirical production functions. J Econ 30:91–107
Chen Y, Cook WD, Li N, Zhu J (2009) Additive efficiency decomposition in twostage DEA. Eur J Oper Res 196:1170–1176
Chen Y, Zhu J (2004) Measuring information technology’s indirect impact on firm performance. Inf Technol Manage 5(1–2):9–22
Chen Y, Liang L, Zhu J (2009) Equivalence in twostage DEA approaches. Eur J Oper Res 193:600–604
Cummins JD, Weiss MA (2013) Analyzing firm performance in the insurance industry using frontier efficiency and productivity methods. Handbook of Insurance, pp 795–861
Cook WD, Zhu J, Bi G, Yang F (2010) Network DEA: additive efficiency decomposition. Eur J Oper Res 207:1122–1129
Despic O, Despic M, Paradi JC (2007) DEAR: ratiobased comparative efficiency model, its mathematical relation to DEA and its use in applications. J Prod Anal 28:33–44
Despotis DK, Koronakos G, Dimitris S (2016) Composition versus decomposition in twostage network DEA: a reverse approach. J Prod Anal 45(1):71–87
Emrouznejad A, Amin GR (2009) DEA models for ratio data: convexity consideration. Appl Math Model 33(1):486–498
Fecher F, Kessler D, Perelman S, Pestieau P (1993) Productive performance of the French insurance industry. J Product Anal 4:77–93
FernandezCastro A, Smith P (1994) To wards a general nonparametric model of corporate performance. Omega 22(3):237–49
Färe R, Grosskopf S (1996) Productivity and intermediate products: a frontier approach. Econ Lett 50(1):65–70
Färe R, Grosskopf S (2000) Network DEA. SocioEcon Plan Sci 34(1):34–49
Färe R, Whittaker G (1995) An intermediate input model of dairy production using complex survey data. J Agric Econ 46(2):201–213
Geoffrion AM (1964) Proper efficiency and theory of vector maximization. J Math Anal Appl 22:618–630
Gerami J, Mozaffari MR (2013) Measuring performance of network structure by DEAR model. Adv Environ Biol 7(7):1224–1232
Gerami J, Mozaffari MR, Wanke PF (2020) A multicriteria ratiobased approach for twostage data envelopment analysis. Expert Syst Appl https://doi.org/10.1016/j.eswa.2020.113508
Gerami J, Kiani Mavi R, Frzipoor Saen R, Kiani Mavi N (2020) A novel network DEA model for evaluating hospital service supply chain performance. Ann Oper Res 295:843–880. https://doi.org/10.1007/s1047902003744z
Guo Ch, Wei F, Ding T, Zhang L, Liang L (2017) Multistage network DEA: decomposition and aggregation weights of component performance. Comput Ind Eng 133:64–74
Guo C, Abbasi SR, Foroughi AA, Zhu J (2017) Decomposition weights and overall efficiency in twostage additive network DEA. Eur J Oper Res 257:896–906
Hwang SN, Kao C (2006) Measuring managerial efficiency in nonlife insurance companies: an application of twostage data envelopment analysis. Int J Manag 23(3):699–720
HatamiMarbini A, Toloo M (2019) Data envelopment analysis models with ratio data: a revisit. Comput Ind Eng 133:331–338
Izadikhah M, Farzipoor Saen R (2016) Evaluating sustainability of supply chains by twostage range directional measure in the presence of negative data. Transp Res Part D 49:110–126
Izadikhah M, Tavana M, Di Caprio D, SantosArteaga FJ (2018) A novel twostage DEA production model with freely distributed initial inputs and shared intermediate outputs. Expert Syst Appl 99:213–230
Khalili M, Camanho AS, Portela MCAS, Alirezaee MR (2010) The measurement of relative efficiency using data envelopment analysis with assurance regions that link inputs and outputs. Eur J Oper Res 203:761–770
Ostovan S, Mozaffari MR, Jamshidi A, Gerami J (2020) Evaluation of twostage networks based on average efficiency using DEA and DEAR with Fuzzy data. Int J Fuzzy Syst 22(1):1665–1678
Kao C, Hwang SN (2008) Efficiency decomposition in twostage data envelopment analysis: an application to nonlife insurance companies in Taiwan. Eur J Oper Res 185(1):418–429
Kao C (2014) Network data envelopment analysis: a review. Eur J Oper Res 239:1–16
Kaffash S, Azizi R, Huang Y, Zhu J (2019) A survey of data envelopment analysis applications in the insurance industry 1993–2018. Eur J Oper Res 284(3):801–813
Kamyab P, Mozaffari MR, Gerami J, Wanke PF (2021) Twostage incentives system for commercial banks based on centralized resource allocation model in DEAR. Int J Product Perform Manag 70(2):427–458
Lim S, Zhu J (2016) A note on twostage network DEA model: Frontier projection and duality. Eur J Oper Res 248:342–346
Liu WB, Zhang DQ, Meng W, Li XX (2011) A study of DEA models without explicit inputs. Omega 39:472–480
Lozano S, Gutiérrez E, Moreno P (2013) Network DEA approach to airports performance assessment considering undesirable outputs. Appl Math Model 37:1665–1676
Lozano S (2015) Alternative SBM model for network DEA. Comput Ind Eng 82:33–40
Lozano S (2015) A jointinputs network DEA approach to production and pollution generating technologies. Expert Syst Appl 42(21):7960–7968
Ilyas AM (2019) An empirical investigation of efficiency and productivity in the Indian nonlife insurance market. Benchmark Int J 26(7):2343–2371
Mozaffari MR, Gerami J, Jablonsky J (2014) Relationship between DEA models without explicit inputs and DEAR models. CEJOR 22(1):1–12
Mozaffari MR, Dadkhah F, Jablonsky J, Wanke PF (2020) Finding efficient surfaces in DEAR models. Appl Math Comput 386:125497
Noulas AG, Hatzigayios T, Lazaridis J, Lyroudi K (2001) Nonparametric production frontier approach to the study of efficiency of nonlife insurance companies in Greece. J Financ Manag Anal Int Rev Finance 14(1):19–26
Olesen OB, Petersen NC, Podinovski VV (2015) Efficiency analysis with ratio measures. Eur J Oper Res 245(2):446–462
Olesen OB, Petersen NC, Podinovski VV (2017) Efficiency measures and computational approaches for data envelopment analysis models with ratio inputs and outputs. Eur J Oper Res 261:640–655
Podinovski VV, Athanassopoulos A (1998) Assessing the relative efficiency of decision making units using DEA models with weight restrictions. J Oper Res Soc 49:500–508
Seiford LM, Zhu J (1999) Profitability and marketability of the top 55 US commercial banks. Manag Sci 45:1270–1288
Seiford LM, Zhu J (1999) Profitability and marketability of the top 55 U.S. commercial banks. Manag Sci 45(9):1270–1288. https://doi.org/10.1287/mnsc.45.9.1270
Silva Portela MCA, Thanassoulis E, Simpson G (2004) Negative data in DEA: a directional distance function approach applied to bank branches. J Oper Res Soc 55:1111–1121
Song M, Wu J, Wang Y (2011) An extended aggregated ratio analysis in DEA. J Syst Sci Syst Eng 20(2):249–256
Sotiros D, Koronakos G, Despotis DK (2019) Dominance at the divisional efficiencies level in network DEA: the case of twostage processes. Omega 85:144–155
Sexton T, Lewis H (2003) Twostage DEA: an application to major league baseball. J Prod Anal 19(2–3):227–249
Tavares G, Antunes CH (2001) A Tchebycheff DEA model. In: Rutcor Research Report
Tohidi G, Razavyan S, Tohidnia S (2014) Profit malmquist index and its global form in the presence of the negative data in DEA. J Appl Math 1–8
Tohidnia S, Tohidi G (2019) Measuring productivity change in DEAR: a ratiobased profit efficiency model. J Oper Res Soc 70(9):1511–1521
Tone K, Tsutsui M (2009) Network DEA: a slacksbased measure approach. Eur J Oper Res 197:243–252
Wang C, Gopal R, Zionts S (1997) Use of data envelopment analysis in assessing information technology impact on firm performance. Ann Oper Res 73:191–213. https://doi.org/10.1023/A:1018977111455
Wang YM, Chin KS (2010) Some alternative DEA models for twostage process. Expert Syst Appl 37:8799–8808
Wei CK, Chen LC, Li RK, Tsai CH (2011) A study of developing an input oriented ratiobased comparative efficiency model. Expert Syst Appl 38:2473–2477
Wei CK, Chen LC, Li RK, Tsai CH (2011) Exploration of efficiency underestimation of CCR model: based on medical sectors with DEAR model. Expert Syst Appl 38:3155–3160
Wei CK, Chen LC, Li RK, Tsai CH (2011) Using the DEAR model in the hospital industry to study the pseudoinefficiency problem. Expert Syst Appl 38:2172–2176
Wu D, Liang L (2005) Aggregated ratio analysis in DEA. Int J Inf Technol Decis Mak 4(3):369–384
Yu Y, Shi Q (2014) Twostage DEA model with additional input in the second stage and part of intermediate products as final output. Expert Syst Appl 41(15):6570–6574
Zha Y, Liang L (2010) Twostage cooperation model with input freely distributed among the stages. Eur J Oper Res 205(2):332–338
Zhang L, Hu H, Zhang D (2015) A credit risk assessment model based on SVM for small and medium enterprises in supply chain finance. Financ Innov 1(14):1–21
Tone K, Kweh QL, Lu W, Ting IWK (2019) Modeling investments in the dynamic network performance of insurance companies. Omega 88:237–247
Eling M, Jia R (2019) Efficiency and profitability in the global insurance industry. Pac Basin Finance J 57:101190
Kou G, Peng Y, Wang G (2014) Evaluation of clustering algorithms for financial risk analysis using MCDM methods. Inf Sci 275:1–12
Kasimbeyli R (2010) A nonlinear cone separation theorem and scalarization in nonconvex vector optimization. SIAM J Optim 20(3):1591–1619
Kou G, Olgu Akdeniz Ö, Dinçer H, Yäksel S (2021) Fintech investments in European banks: a hybrid IT2 fuzzy multidimensional decisionmaking approach. Financ Innov 7:39
Zha Q, Kou G, Zhang H et al (2020) Opinion dynamics in finance and business: a literature review and research opportunities. Financ Innov 6:44
Acknowledgements
The author would like to thanks the four anonymous reviewers and the editor for their insightful comments and suggestions.
Funding
This research did not receive any specific grant from funding agencies in the public, commercial, or notforprofit sectors.
Author information
Authors and Affiliations
Contributions
The author read and approved the final manuscript.
Corresponding author
Ethics declarations
Competing interests
The author declares that he has no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Akbarian, D. Network DEA based on DEAratio. Financ Innov 7, 73 (2021). https://doi.org/10.1186/s40854021002786
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s40854021002786
Keywords
 Data envelopment analysis
 DEAR
 Twostage DEA