Fraudulent Financial Reporting and Technological Capability in the Information Technology Sector: A Resource-Based Perspective

Abstract

Motivated by the disproportionately high incidence of fraudulent financial reporting in the IT sector where technological capability is a major source of competitive advantage, this study investigates the possible relationship between technological capability and fraud probability in the IT sector. Technological capability is measured by a firm’s technical efficiency relative to peers in transforming cumulative R&D resources into innovative output, which is a source of competitive advantage, according to the resource-based view (RBV) of the firm. Technical efficiency is estimated via data envelopment analysis. A sample of fraud firms taken from Accounting and Auditing Enforcement Releases is matched with control samples of non-fraud firms. Consistent with the RBV, technological capability is found to have a negative and economically significant effect on fraud probability. Moreover, fraud probability is insignificantly associated with the scale efficiency of innovative activities, as investment in R&D resources per se is not a source of sustainable competitive advantage.

Appendix

DEA compares the relative efficiencies of “decision-making units (DMUs)” (e.g., firms) in using similar resources to generate similar output. The efficiency score of each DMU ranges from 0 to 1. The most efficient DMUs have an efficiency score of 1 which is the benchmark of “best practice” (i.e., the frontier) among peers. The lower a DMU’s efficiency score is below 1 (i.e., below the frontier), the more inefficient the DMU relative to the best practice.

Based on the work of Banker et al. (1984), Fare et al. (1994), and Ruggiero (1996, 1998), this study specifies an output-oriented DEA model controlled for exogenous technological progress as the following linear programming problem:

$$\begin{array}{*{20}l} {{\text{Max }}\theta_{h} } \hfill & {} \hfill \\ {{\text{Subject to}}:} \hfill & {Y\lambda \ge \theta_{h} PAT_{h} } \hfill \\ {} \hfill & {X\lambda \le RND_{h} } \hfill \\ {} \hfill & {\lambda j = \, 0{\text{ if }}t_{j} > t_{h} {\text{ for all }}j \ne h} \hfill \\ {} \hfill & {I_{N\prime } \lambda = \, 1} \hfill \\ {} \hfill & {\lambda \ge \, 0} \hfill \\ \end{array}$$

where 1 ≤ θ_h ≤ ∞; Y = (PAT₁,…, PAT_N); X = (RND₁,…, RND_N); t₁,…, t_N is the time trend capturing exogenous technological progress; λ is a N × 1 vector of weights; and I_N is a N×1 vector of ones. By imposing the constraint of λj = 0 if t_j > t_h for all j≠h, this model excludes observations with more advanced technologies (i.e., a more favorable environment) from the reference set (Ruggiero 1996, 1998). I_N′λ = 1 imposes variable returns to scale (VRS) on the solution (Banker et al. 1984).

The interpretation of Yλ ≥ θ_hPAT_h and Xλ ≤ RND_h is as follows. Choose a weighted combination of all input observations (Xλ) that uses at most the input observation under evaluation (RND_h) to produce the largest possible multiple of the output observation under evaluation (θ_hPAT_h). The input–output observation under evaluation is efficient if its output is best produced using its own input, i.e., one cannot find any λ that generates θ_h > 1. This efficient observation with θ_h = 1 defines a point on the frontier because its efficiency cannot be further improved relative to the other observations. If θ_h > 1, θ_h − 1 is the proportional increase in PAT_h without increasing RND_h. 1/θ_h therefore defines an efficiency score varying between 0 and 1.

The value of θ for each input–output observation can be obtained by solving the preceding linear programming problem N times. To separate scale efficiency from technical efficiency, the former can be calculated as the difference between θ and θ′, where θ′ is the solution without the VRS constraint.