Developing an Impairment Loss Given Default Model Using Weighted Logistic Regression Illustrated on a Secured Retail Bank Portfolio

Abstract

This paper proposes a new method to model loss given default (LGD) for IFRS 9 purposes. We develop two models for the purposes of this paper—LGD1 and LGD2. The LGD1 model is applied to the non-default (performing) accounts and its empirical value based on a specified reference period using a lookup table. We also segment this across the most important variables to obtain a more granular estimate. The LGD2 model is applied to defaulted accounts and we estimate the model by means of an exposure weighted logistic regression. This newly developed LGD model is tested on a secured retail portfolio from a bank. We compare this weighted logistic regression (WLR) (under the assumption of independence) with generalised estimating equations (GEEs) to test the effects of disregarding the dependence among the repeated observations per account. When disregarding this dependence in the application of WLR, the standard errors of the parameter estimates are underestimated. However, the practical effect of this implementation in terms of model accuracy is found to be negligible. The main advantage of the newly developed methodology is the simplicity of this well-known approach, namely logistic regression of binned variables, resulting in a scorecard format.

Strengthen the Security Management of Customer Information in the Virtual Banks of Hong Kong through Business Continuity Management to Maintain Its Business Sustainability

“…To support our choice of classification tool, Daniel Jurafsky and James H.Martin (2009) [25] proposed that not only is LR extraordinarily fitted to figure out the connections or cues between the given outcomes and some special features, but also it performs the binary decision extremely well. Moreover, some of the previous published studies have also applied LR in analyzing the bank's risk factors and the related topics, such as the examples shown in Table 1, the bank's reliability calculation (Ravi, V., and Madhav, V., 2021) [26], the estimate of the bank risks (Breed, D. G et al, 2019) [27], the evaluation of the consumers' credit risk (Abid, L et al, 2018) [28], the assessment of the natural risk factors (Davis, L., and Harden, C., 2014) [29], and the prediction of the risk indexes for the banking failure (Taha Zaghdoudi, 2013) [30]. Apart from that, many researchers have utilized LR in the analysis of risk factors in other areas, such as the fields of weather and climate.…”

Development of an Impairment Point in Time Probability of Default Model for Revolving Retail Credit Products: South African Case Study

“…The methodology's strengths may be summarised as follows: The model methodology utilises well-understood methods (e.g., Lorenz curve calibration, scorecards, term structure modelling) used in the banking industry. The inclusion of re-default events in the proposed IFRS 9 PD methodology will simplify the development of the accompanying IFRS 9 LGD model due to the reduced complexity for the modelling of cure cases (Breed et al 2019). Attrition effects are naturally included in the PD term structures and do no longer require a separate model.…”

“…IFRS 9 LGD model due to the reduced complexity for the modelling of cure cases (Breed et al 2019). Attrition effects are naturally included in the PD term structures and do no longer require a separate model.…”

“…It is one of the most popular retail credit scoring measures (Baesens et al 2016) and has the added advantage that it is a single value (Tevet 2013). For detailed Gini calculations, the reader can consult various literature sources, e.g., SAS Institute (2017) and Breed et al (2019).…”

The Changing Landscape of Financial Credit Risk Models

“…The credit crisis of 2007/2008 provoked concerns about the use of models in finance (De Jongh et al 2017). Many examples can be found in the literature on the effect that this credit crisis had on financial credit risk predictive models (Breed et al 2019), (Dendramis et al 2018), and (Skoglund 2014). Generally, financial credit risk predictive models break down in a financial crisis (Saliba et al 2023).…”