In the present era of complex environments, banks operate in a more dynamic environment, which in turn, affects their relative efficiency. Traditional Data envelopment analysis (DEA) models are widely used to measure efficiency. However, environmental/exogenous variables can significantly influence the DEA efficiency scores. Therefore, identifying the most important environmental variables is crucial in the evaluation of bank performances. This study introduces a three-stage DEA framework that employs a random forest as a powerful ensemble method for variable selection to search for the most influential environmental variables. The direction of influence of the selected environmental variables and their predictive power for predicting bank performances are investigated in the third stage, through a regression analysis. The proposed framework is tested with a sample of 110 banks in Middle East and North Africa countries, observed over a period of 3 years (2014 till 2016). Accordingly, a relevant set of environmental variables is identified and its effects on bank efficiency are studied. The findings indicate that the country where the bank operates has a significant effect on the bank’s efficiency. Results also show that the overall average efficiency score is stable (around 87%) for all banks. The study concludes with the limitations and suggested directions for further research.

中文翻译：

---

低效源追踪：来自数据包络分析和随机森林的证据

在当今环境复杂的时代，银行在更加动态的环境中运营，这反过来又会影响其相对效率。传统的数据包络分析 (DEA) 模型被广泛用于衡量效率。然而，环境/外生变量会显着影响 DEA 效率得分。因此，识别最重要的环境变量对于评估银行绩效至关重要。本研究引入了一个三阶段 DEA 框架，该框架采用随机森林作为变量选择的强大集成方法，以搜索最具影响力的环境变量。在第三阶段，通过回归分析，研究所选环境变量的影响方向及其对银行业绩的预测能力。提议的框架在中东和北非国家的 110 家银行样本中进行了测试，观察时间为 3 年（2014 年至 2016 年）。因此，确定了一组相关的环境变量，并研究了其对银行效率的影响。调查结果表明，银行经营所在的国家对银行的效率有显着影响。结果还显示，所有银行的整体平均效率得分稳定（约 87%）。该研究总结了局限性和进一步研究的建议方向。调查结果表明，银行经营所在的国家对银行的效率有显着影响。结果还显示，所有银行的整体平均效率得分稳定（约 87%）。该研究总结了局限性和进一步研究的建议方向。调查结果表明，银行经营所在的国家对银行的效率有显着影响。结果还显示，所有银行的整体平均效率得分稳定（约 87%）。该研究总结了局限性和进一步研究的建议方向。