Do Commercial Banks Benefited From the Belt and Road Initiative? A Three-Stage DEA-Tobit-NN Analysis

Akber Aman Shah, Desheng Dash Wu, Vladimir Korotkov, Gul JabeenIEEE Access 2019 · OpenReview

Abstract

Traditional Data Envelopment Analysis (DEA) often treats decision-making units (DMUs) as "black boxes," failing to account for internal structures or provide actionable paths for improvement. This paper introduces a novel three-stage hybrid model—combining DEA, Tobit regression, and Neural Networks (NN)—to evaluate and enhance the efficiency of commercial banks within the Belt and Road Initiative (BRI). By applying this model to 1,343 banks, the study demonstrates how targeted adjustments to internal determinants can unlock efficiency gains across various risk categories, transforming descriptive analysis into prescriptive action.

The "Black Box" Problem in Efficiency Measurement

In operations research, measuring the efficiency of complex entities like banks is a persistent challenge. While DEA is a standard tool for identifying efficient benchmarks, it typically focuses on the relationship between aggregate inputs and outputs without peering into the internal transformation mechanisms. Existing "two-stage" models attempt to address this but often encounter mathematical conflicts and remain purely descriptive. They can identify that a bank is inefficient, but they cannot provide a specific roadmap for how to improve.

Methodology: A Three-Stage Prescriptive Framework

To bridge the gap between measurement and improvement, the authors developed a three-stage framework designed to move beyond the "black box" limitation.

Conceptual flow of the three-stage DEA-Tobit-NN model.
  1. Stage 1: Efficiency Evaluation (DEA): The model uses a BCC output-oriented DEA approach to calculate efficiency scores, distinguishing between benchmark banks and those lagging behind.
  2. Stage 2: Identifying Determinants (Tobit Regression): Since efficiency scores are censored (ranging from 0 to 1), the authors employ Tobit regression to identify the internal (e.g., assets, equity) and external (e.g., GDP growth) factors driving performance.
  3. Stage 3: Improvement Prediction (Neural Network): This stage introduces a Multi-Layer Perceptron (MLP) neural network. By training the NN on data from efficient banks, the model can predict the optimal "improved determinants" for inefficient ones.
Detailed algorithmic logic for efficiency enhancement using Neural Networks.

As shown in the logic flow, the model iteratively adjusts parameters for "non-benefited" groups, using the NN as a virtual consultant to suggest specific operational changes required to reach the efficiency frontier.

Data and Context: The Belt and Road Initiative

The study analyzed 1,343 commercial banks across BRI countries from 2010 to 2017. Banks were categorized into six groups based on their Economist Intelligence Unit (EIU) risk ratings (from low-risk Group A to high-risk Group CCC). This allowed the researchers to account for the diverse economic environments within the BRI.

Key Insights and Results

The analysis revealed that the BRI did not provide a universal "rising tide" for all banks. Initially, only "Group B" banks (including countries like Russia and Vietnam) saw a statistically significant natural boost in efficiency, rising from 79% to 84%.

However, the prescriptive power of the NN stage showed that all groups could potentially benefit:

  • Group A (Low risk): Predicted efficiency increase of 2.25%.
  • Group BB: Predicted efficiency increase of 1.25%.
  • Group CCC (High risk): Predicted efficiency increase of 1.75%.

By simulating improvements in internal determinants, the model demonstrated that even banks in high-risk environments could achieve significant gains if they optimized their internal structures according to the NN's recommendations.

Conclusion and Future Directions

The three-stage DEA-Tobit-NN model represents a significant shift from descriptive to prescriptive analytics in banking. It provides a practical tool for bank managers to navigate large-scale geopolitical shifts like the BRI. Future research may refine these insights by incorporating deep learning architectures or more granular data on specific banking sub-sectors to handle increasingly complex, non-linear economic relationships.

Summary updated

Abstract

The data envelopment analysis (DEA) treats decision-making units (DMUs) as black boxes: there is an unknown internal structure and transformation mechanism of input to output. Two-stage models have been proposed to resolve this problem by considering the internal structure of DMUs. However, each DMU has a different structure, and in two-stage models, the poor estimation of sub-models causes conflicts in the intermediate layer. Therefore, it is necessary to use additional tools to extract insight into opportunities to enhance the performance of DMUs. This paper presents a three-stage model employing DEA to evaluate efficiency, a Tobit regression model to identify the determinants, and a neural network (NN) to improve those determinants. Improvement in the determinants of a DMU enhances its efficiency. The developed model is applied to the empirical dataset of commercial banks from the countries that have joined the belt and road initiative (BRI), grouping them based on their economist intelligence unit (EIU) rating. The results provide valuable information on the efficiency enhancement process for banks to benefit from the BRI.
BibTeX
@inproceedings{shah2019do,
  author = {Akber Aman Shah and Desheng Dash Wu and Vladimir Korotkov and Gul Jabeen},
  title = {{Do Commercial Banks Benefited From the Belt and Road Initiative? A Three-Stage DEA-Tobit-NN Analysis}},
  booktitle = {IEEE Access},
  year = {2019},
  url = {https://openreview.net/forum?id=9EnWLYZqAA},
}
Figure