Spatial dependence in the efficiency of Indonesian regional development banks: A two-stage DEA-SAR Tobit analysis

Abstract

This study analyzes the technical efficiency and spatial dependence of the 27 Regional Development Banks (RDBs) in Indonesia from 2013 to 2022. A two-stage quantitative design was used. First, using Data Envelopment Analysis under the intermediation approach, the inputs used to estimate annual efficiency were third-party funds, total costs, and fixed assets, while the outputs were loans, interest income, and non-interest income. Second, Moran’s I was used to examine spatial autocorrelation in the DEA scores, while pooled SAR and SEM Tobit models were used to estimate spatial dependence and the relationships between regional and bank-specific characteristics and latent efficiency. A row-standardized first-order Queen contiguity matrix was used to represent spatial links; units with no contiguous neighbors had zero rows. Mean efficiency was 0.970, declined to 0.946 in 2020, and recovered to 0.984 in 2022. In 2018, 2020, and 2021, significant positive values of spatial autocorrelation were found. The SAR Tobit model also performed better than the non-spatial pooled Tobit model (LR = 8.944, p = 0.0028), and the spatial-lag coefficient was positive and significant (ρ = 0.025, p = 0.003). There was a negative relationship between non-performing loans and latent efficiency and a positive relationship between the loan-to-deposit ratio and latent efficiency. At the 10% level, GRDP per capita, capital adequacy, and the cost-to-income ratio exhibited weaker relationships. The results indicate that a dual strategy is needed: bank-level risk management and intermediation enhancement, along with coordinated benchmarking, digital collaboration, and peer learning through OJK, ASBANDA, provincial shareholders, and RDB management.