Incorporating causal modeling into data envelopment analysis for performance evaluation
本文在动态网络DEA中引入贷款损失准备金作为理想投入，并用因果建模验证模型，发现2013-2020年中国银行业效率波动大，国有银行效率最低，不同行业信贷风险对效率影响各异。
Abstract The risk factors in banking have been considered an undesirable carryover variable by the literature. Methodologically, we consider the risk factor using loan loss reserves as a desirable carryover input with dynamic characteristics, which provides a new framework in the dynamic network Data Envelopment Analysis (DEA) modelling. We substantiate our formulation and results using novel techniques for causal modelling to ensure that our dynamic network model admits a causal interpretation. Finally, we empirically examine the impact of risk from various economic sectors on efficiency. Our results show that the inefficiencies were volatile in Chinese banking over the period 2013–2020, and we further find that the state-owned banks experienced the highest levels of inefficiency and volatility. The findings report that credit risk derived from the agricultural sector and the Water Conservancy, Environment and Public Facilities management sector decreases bank efficiency, while credit risk derived from the wholesale and retail sector improves bank efficiency. The results of our innovative causal modelling show that our pioneering modelling on the role of loan loss reserves is valid. In addition, from an empirical perspective, our second-stage analysis regarding the impact of risk derived from different economic sectors on bank efficiency can be applied to other banking systems worldwide because of our successful validation from causal modelling. Our attempt to incorporate causal inference into DEA can be generalized to future studies of using DEA for performance evaluation.