Benchmarking bank efficiency under alternative measurement specifications: intellectual capital and ownership structure in Vietnam

Purpose This paper benchmarks Vietnamese banks’ technical efficiency and tests whether the benchmarks, and their association with intellectual capital, are robust to the choice of data envelopment analysis (DEA) specification. Design/methodology/approach Using an unbalanced panel of 624 bank-years from 42 banks (2002–2022), efficiency is estimated under the intermediation and production approaches on an identical common sample. The scores are related to the Modified Value Added Intellectual Coefficient (MVAIC) through fractional probit regression, with system GMM and Simar–Wilson estimation as robustness checks. Findings MVAIC is positively and significantly associated with efficiency under the intermediation approach but only marginally under the production approach. Specification choice changes which banks are identified as benchmarks: most bank-years efficient under one approach are not efficient under the other. State-owned banks are far more consistently robust benchmarks than joint-stock banks. We find no evidence that the association differs by ownership or across systemic shock periods. Practical implications Banks efficient under both specifications are the safest benchmarks; single-specification benchmarks suit only the activity that specification measures, and disagreement between specifications indicates which activity warrants closer examination. Originality/value This is among the first studies to show directly, through rank correlations, frontier transitions and bank-level benchmark identification that a bank’s benchmark status depends on which of two standard DEA specifications is used, with implications for benchmarking institutions with heterogeneous business models.