Mixed network DEA with shared resources: A case of measuring performance for banking industry

As a bank manager, not only the overall performance of the bank is considered, but also the performances of all divisions needed to be evaluated. Though there have been many works investigating the performance of banking by Data Envelopment Analysis (DEA) since the 1990s, the internal structure and operational processes of a banking system are never such simple as represented in existing DEA models. We propose a new DEA modelling technique to solve a mixed network structure, which consists of serial and parallel processes with shared input resources. Relational decomposition and additive aggregation network models are developed to measure the overall efficiency as well as the efficiencies of all components within the model. The empirical results obtained for the Taiwan banking industry acknowledges the discriminating power of our structure models compared to the ones proposed in existing literature. We also find evidence that traditional banking (lending) has been gradually outperformed by low-risk businesses (investment or non-interest-based services). Some managerial implications are then discussed to improve the system's efficiency.