1. Introduction
The managers of any company must measure the financial performance of the firm they manage. Assessing the performance of modern construction companies is a complex issue from both an international and local perspective [
1]. The financial success of construction firms very much depends on the location [
2], proper management [
3], environmental characteristics [
4], selected technologies [
5], staff qualification [
6], and specific circumstances [
7,
8]. Besides, at the project level, the influence of various factors on the project success may be responsible for the production of differently significant outcomes [
9]. The managers can predict the data describing the projects accurately and determine them at certain intervals, or they should treat them as fuzzy data [
10,
11].
Management accounting information is mainly used for firm financial performance assessment. Since the early 1990s, management accounting has attracted studies from operations research that suggest frameworks or propose model building on methods, such as data envelopment analysis (DEA) (see Callen [
12] and Malmi [
13] for surveys). DEA [
14] is a multicriteria evaluation method that can screen the most desirable alternatives (i.e., decision-making units (DMUs), e.g., firms) among large sets by means of mathematical programming. DEA provides for each DMU a composite score, which is referred to as efficiency using actual input–output data for a sample of them and this facilitates the complexity of analysis by evaluating the multicriteria [
15]. DEA can be employed in a multiple input–multiple output setting and, in regard to model building, it avoids the prior assumption on weights of inputs and outputs; the weights are produced by the optimization process.
Firm assessment of financial performance using DEA deals with the analysis of financial statement data to distinguish a sample of firms into efficient and inefficient. DEA-based efficiency and inefficiency reflect good and bad financial conditions, respectively [
16].
The current paper aims to employ a constructive research approach [
17] by providing a novel construct that is based on DEA and demonstrate its practical applicability. Firm performance evaluation can be accomplished by means of a derived composite indicator (CI) using DEA [
18]. The approach used for the derivation of a CI distinguishes from the conventional DEA in that it looks at the one side of DEA (i.e., on outputs or inputs only) using a set of single (individual) indicators. The use of DEA to derive CIs at the firm-level is promising since DEA can aggregate multiple firm performance dimension, expressed as single financial ratios, into a consolidated performance metric.
Horta et al. [
18] recently employed DEA to evaluate construction enterprise performance by deriving firm CIs. The current study improves upon Horta et al. [
18] by firstly employing GRA for the selection of financial ratios and then using a modified range-adjusted measure (RAM) of efficiency [
19] suggested by Lozano and Gutierrez [
20] for the derivation of firm CIs. Recent studies integrate models based on grey theory with DEA [
21,
22], and, for the case of robustness, the DEA results are compared with those of GRA. In a second step, the determinants (i.e., drivers) of firm overall performance (i.e., DEA-based CIs) are explored using regression techniques (i.e., censored Tobit regression). The practical applicability of the proposed RAM-Tobit regression modeling is demonstrated at the project-level for a group of construction projects performance evaluation [
23]. In particular, this paper focuses on a sample of three upper-class contracting license (Classes 5–7) Greek construction firms with the aim to derive DEA-based performance scores (i.e., composite indicators (CIs)) using selected financial ratios.
GRA is based on grey theory and it is used to derive the relational degree of every attribute for a set of alternatives in a multicriteria decision-making (MCDM) problem where multiple variables and their interrelationships need to be taken into account. GRA produces for every alternative one single-attribute value by considering all multiple-attribute values to facilitate the whole process [
24]. It is worth noting that the utilization of GRA in construction is not very widespread in terms of multicriteria analysis [
25]. For other MCDM methods in construction, the interested reader is referred to the work of Jato-Espino et al. [
25].
In this paper, the composite indicator construction is modeled as a decision-making problem with multi-attributes. Financial ratios are viewed as attributes for construction firms, and firms as alternatives.
This paper contributes to the existing literature by outlining a procedure that can be used by firms and consultants to aggregate single firm financial ratios into one CI and identify the drivers of performance. Firstly, GRA ranks the level of importance of the financial ratios, and then, based on selected ratios, firm CIs are constructed by means of a no-input RAM by allowing firms to adjust in the direction of greater output-like values (i.e., single financial ratios) as much as it is feasible in the DEA context; DEA and GRA results are also compared in terms of firm ranking. In this sense, a useful performance companion is provided when the goal of firms is the maximization of selected financial ratios. The incremental contribution of the current research over Tsolas [
23] lies in the employment of a no-input RAM with the aim to pinpoint the drivers of firm performance in terms of DEA-based CIs and the integration of DEA with GRA. The identification of firm performance drivers will guide future studies for analyzing financial statement data of Greek construction firms.
The remainder of the paper is organized as follows:
Section 2 provides a review on the use of DEA in management accounting. Moreover, it also reviews firm DEA studies in the construction industry as well as the relation of DEA with multicriteria methods.
Section 3 describes the methods used for the analysis of data.
Section 4 presents the dataset.
Section 5 reports and discusses the results.
Section 6 provides policy implications.
Section 7 concludes.
2. Literature Review
DEA performance assessment in the field of management accounting research deals with the works of Turner [
26] on manufacturing maintenance performance; Banker et al. [
27] on nursing services; Deville [
28] on branch banking network assessment; Deville et al. [
29] on bank efficiency; Halkos and Salamouris [
30] on Greek commercial banks; and Rouse et al. [
31] on aircraft maintenance (see also Callen [
12] and Malmi [
13] for surveys on that subject). It is worth noting that DEA can be integrated with other MCDM methods for financial performance purposes [
32].
In the operations research literature, there are a few DEA works that are deemed as the first which make use of financial statement data [
33,
34,
35,
36]. Relevant are also the works by Feroz et al. [
37] who employed the DuPont model in oil and gas, pharmaceuticals and primary metals industries; Rodriguez-Perez et al. [
38] who provided a DEA-based performance assessment of insurance companies; and Demerjian et al. [
39] who used DEA to quantify managerial ability. A recent review can be found in Harrison and Rouse [
40].
In regard to relevant DEA studies in the construction sector, the works at the firm-level are classified as standard, two-stage, and series two-stage DEA studies. The standard studies focus on the relative efficiency of DMUs. The two-stage DEA studies aim not only to derive the efficiency metrics but also to identify the drivers of performance, whereas the series two-stage studies distinguish two stages with output from the first stage becoming input to the second stage. The works in the first and second strands can be classified further into studies that are based on production theory and studies that use DEA models to produce synthetic performance indicators.
In the first strand (i.e., standard DEA studies) lie a lot of relevant works. Pilateris and McCabe [
41] use conventional DEA (e.g., CCR [
14] and BCC [
42]) models to evaluate contractors. McCabe et al. [
43] and El-Mashaleh et al. [
44,
45] employ DEA to prequalify and evaluate contractors, respectively. Sueyoshi and Goto [
46] use DEA as a discriminant tool. Horta et al. [
47] employ DEA models with and without weight restrictions, and Horta and Camanho [
48] combine DEA with other data mining techniques.
In the second strand (i.e., two-stage DEA modeling) lies the work by Horta et al. [
18] who first derive DEA-based composite performance indicators by means of an equivalent to CCR model and then identify the drivers of performance.
In the third strand, i.e., series two-stage DEA modeling, firstly proposed by Seiford and Zhu [
49] lie the DEA studies by Tsolas [
50,
51] on the evaluation of listed Greek construction enterprises and Hu and Liu on the Australian [
52] and Chinese construction industry [
53].
As for the Greek construction industry, DEA firm-level studies are the works by Tsolas [
50,
51] and Christopoulos et al. [
54]. In the above studies [
34,
38] DEA is integrated with ratio analysis in a two-stage framework [
50] or employed to provide metrics that stem from models that use financial ratios and financial statement data separately [
54].
Using DEA, the derivation of composite performance metrics is based on the methodological relation between multicriteria decision analysis (MCDA) and DEA. The connection between the two approaches is that if all criteria in MCDA can be defined as benefit (i.e., maximizing) or cost (i.e., minimizing) criteria, outputs and inputs are equivalents of these in DEA terminology. Thus, if a criterion is defined as minimizing or maximizing, it can be considered in the DEA model as input or output. The basic function of DEA is to classify the units under evaluation in efficient and inefficient ones; in MCDA, these can be regarded as non-dominated and dominated alternatives, respectively. For more on the methodological relation between DEA and MCDA, the interested reader is referred to previous relevant studies [
55,
56,
57,
58].
GRA studies in construction are scarce [
25]. There are mainly GRA studies at the project level on pre-evaluation of engineering design [
59] and bid evaluation [
60].
Two-stage DEA is performed by firstly calculating DEA efficiency ratings and then regressing them on explanatory variables using Tobit or ordinary least squares (OLS) regression. Simar and Wilson [
61] suggest the use of an integrated with bootstrapping truncated regression rather than Tobit regression. They bootstrap the efficiency scores to produce bias-corrected efficiency ratings and then they regress these corrected ratings on explanatory variables using bootstrapped truncated regression. As to which regression method is the most appropriate to employ in the second stage of analysis is a subject that is the interest in a lot of studies. Banker and Natarajan [
62] argue that the above methods are all appropriate. McDonald [
63] supports the use of OLS instead of Tobit regression. Moreover, new methods have been proposed such as fractional regression [
64], partial least squares regression [
65], or other efficiency estimators [
66]. For a recent review the interested reader is referred to Liu et al. [
66].
Banker et al. [
67] argue that Simar and Wilson’s [
61] procedure precludes statistical noise and it is inconsistent with the modeling of production functions. In the light of the simulation results of their study [
67] the DEA-OLS and DEA-Tobit procedure dominate the DEA-bootstrapped truncated regression. The results hold whether explanatory and other variables are independent or correlated, and whether there is no statistical noise.
The DEA-bootstrapped truncated regression cannot be applied to slack-based models, such as RAM. To the best of the author’s knowledge, there has not been developed yet a bootstrapping procedure for slacks-based models along the lines of Simar and Wilson [
61]. Therefore, in the current paper the DEA-Tobit procedure will be employed. Moreover, for the case of robustness the results of DEA-OLS approach will be also discussed.
Another problem with two-stage DEA modeling is relative to the separation of the space of the input/output variables used in first stage DEA and the space of the explanatory factors used as independent variables in the second stage of analysis. Simar and Wilson [
61] found that regression techniques such as Tobit and OLS were inappropriate in the second stage and proposed bootstrapped truncated regression, despite understanding that this technique may suffer from the same problem. According to Daraio et al. [
68], if the separability condition does not hold, the second stage results would have drawbacks [
69].
The separability assumption of Simar and Wilson [
61] is strong [
67,
70] and it is unlikely to be satisfied in real applications [
67,
71]. Despite the strong nature of this assumption the tests presented in a proceeding research by Daraio et al. [
68,
71,
72] may be employed to confirm the separability condition. In cases where the separability assumption is rejected, models of conditional efficiency may be used. For a recent survey on that issue the reader is referred to Henriques et al. [
69]. Due to unavailability of a test code, Benito et al. [
73] propose the comparison of outcomes based on the bootstrapped truncated regression [
61] procedure, as well as on the conditional efficiency measures [
71]; results produced by the two methods are considered robust if there is consistency of outcomes, otherwise it may be possible that the separability condition does not hold. Most of the previous two-stage DEA studies, although may comment on the reality of the above separability condition, take its fulfillment for granted [
73]. Therefore, in the current paper as in most of the previous relevant works separability appears as a taken-for-granted assumption.
The contribution of the current research is manifold. First, this paper distinguishes from previous construction industry DEA works that use the same two-stage framework (e.g., [
23]) in that it focuses on financial firm performance. Second, it documents how to derive CIs for financial firm performance by integrating GRA with a no-input RAM of efficiency. The employment of conventional DEA models, such as the CCR and the BCC model, may be problematic when financial statement-based ratios are used due to the presence of negative values, and therefore, by employing the RAM for the derivation of construction firm Cis, the current study improves upon Horta et al. [
18]. Moreover, the GRA and DEA results, in terms of firm ranking, are also compared. Third, the two-stage modeling adopted herein involves a DEA model (i.e., RAM) that firstly uses a set of single financial ratios to produce firm CIs and then the firm CIs produced are regressed on a set of explanatory variables such as size, geographical location and class of the firm. The latter variable reflects the complexity of firm projects and characterizes the structure of the Greek construction industry [
51], whereas the other two variables have also been used by other researchers [
18]. As regards the class of the firm, the current research uses a sample that represents about 88 percent of the total population and therefore this sample feature aids to generalize results.