Journal of Risk Model Validation

Steve Satchell
Trinity College, University of Cambridge

This issue of The Journal of Risk Model Validation covers a wide range of topics. Those who, like myself, struggle to keep up with developments in artificial intelligence, machine learning, robotics and related phenomena will find some excellent applications outlined below. We are also delighted to include a paper that addresses risk validation in real estate.

“Exchange rate risk management for contractors within a hybrid payment scheme: a case study in Punta del Este, Uruguay” by Martín Egozcue, the first paper in this issue, investigates the strategies contractors can employ to mitigate the exchange rate risks present in hybrid payment systems. Contractors face both exchange rate risk, due to mismatches between their revenue and cost currencies, and property price risk, since they receive a portion of their revenue in the form of dwelling units. Egozcue compares the performance of three distinct risk models within the context of real estate development in Punta del Este, Uruguay. By evaluating these models against empirical data from a hypothetical project, his research provides valuable insights into the effectiveness of these models in managing exchange rate risk. This paper addresses the need to validate risk models in emerging real estate markets such as that in Punta del Este and more generally.

In our second paper, “Shapley values as an interpretability technique in credit scoring”, Hendrik Andries du Toit, Willem Daniël Schutte and Helgard Raubenheimer argue that the use of machine learning algorithms in credit scoring can be enhanced by an improved understanding of the reasoning behind model decisions. Although machine learning algorithms are widely regarded as highly accurate, their use in settings that require an explanation of model decisions has been limited due to a lack of transparency. In particular, the model risk frameworks of banks frequently require a significant level of model interpretability. In this paper, the Shapley value is evaluated as a machine learning interpretability technique in credit scoring. The Shapley value is calculated by averaging the marginal contributions that quantify the contribution of each feature in the prediction of a specific observation. The usefulness of this technique is evaluated by du Toit et al on various simulated data sets with covariates from different underlying distributions that are linearly and nonlinearly related to the outcome. Traditional models (eg, logistic and linear regression) and machine learning algorithms are trained on the data, and Shapley values are generated. The authors show that Shapley values are related to “weights of evidence” (a well-known measure in the scorecard literature), which can be used to explain the direction of relationships between explanatory variables and the outcome.

The issue’s third paper is titled “Online attention and directors’ and officers’ liability insurance: evidence from Chinese listed firms”. Its authors, Can Lin and Huobao Xie, investigate the effects of online attention on corporate purchases of D&O liability insurance. Their theoretical analysis and empirical tests using data for A-share Chinese listed firms from 2011 to 2021 show that online attention leads to an increase in D&O liability insurance purchases. The analysis and tests of the mediating mechanism show that online attention enhances purchases by improving investor protection and exacerbating managerial career concerns, and the authors demonstrate that investors’ website visits weaken the promotional effect of online attention on purchases, while corporate tax aggressiveness enhances this effect. The authors’ heterogeneity analysis and tests indicate that the increase in purchases driven by online attention is more statistically significant in firms that have lower investor confidence, in those that have involuntary disclosure of social responsibility, in those that are state owned, and in those that face greater financing constraints. The findings of this paper should aid the improvement of risk management systems in China’s capital markets.

The final paper in this issue is “Forecasting the default risk of Chinese listed companies using a gradient-boosted decision tree based on the undersampling technique” by Shanshan Wang, Guotai Chi, Ying Zhou and Li Chen. Default prediction is of interest to the creditors, customers and suppliers of any firm, as well as to policymakers and current and potential investors. Imbalanced classification of default prediction is considered to be a crucial issue. Therefore, this study proposes a default risk prediction model using a gradient-boosted decision tree (GBDT) based on a random undersampling (RUS) technique. The authors build default prediction models based on 29 indicators and five different time windows. The model has two main steps. First, the proposed RUS-GBDT model adopts an undersampling approach to generate different training samples based on the imbalance ratio of training data; and second, the parameters of GBDT are adaptively tuned with respect to the receiver operating characteristic curve (area under the curve) of the predictive model for the selected training sample. Wang et al analyze the optimal imbalance ratio of different training samples, and their experiment compares prediction performance with several classification methods, including logistic regression and a support vector machine. Their experimental results demonstrate that the model they propose performs better than other classifiers at predicting and classifying the default status of listed companies in China.