{"paper_id":"2610.05926","title":"Comparing two approaches for modelling the loss given default of credit cards: Run-off triangles vs regression","authors":["Arno Botha","Henko Crewe","Marcel Muller","Janette Larney"],"paper_date":"2026-10-05","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2610.05926","page_url":"https://thequant.space/flowcharts/comparing-two-approaches-for-modelling-the-loss-given-default-of-credit-cards/","scores":{"empirical_rigor":8,"math_complexity":6,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes two variants of a run‑off triangle (ROT) based LGD model and a classical two‑stage regression‑based LGD model for credit‑card data.","quote":"We propose two variants of a ROT-based LGD-model to serve as our baseline.","summary":"The use of run-off triangles (ROTs) is a common industry practice in estimating the loss given default (LGD) risk parameter when predicting credit losses in banking. We benchmark this industry practice using credit card data against a more sophisticated (though classical) regression-based approach…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The loss severity component follows a Gaussian distribution with constant variance.","type":"model","stated":"explicit","quote":"We assume that the response variable follows a Gaussian distribution with constant variance 𝜎2, i.e., 𝑌𝑖∼𝑁(𝜇𝑖, 𝜎2)."},{"assumption":"Development factors are stable across default cohorts and over time.","type":"model","stated":"explicit","quote":"The ROT-approach relies on the assumption that development factors are stable across default cohorts."},{"assumption":"A VIF‑threshold of 10 is used for variable selection.","type":"statistical","stated":"explicit","quote":"Accordingly, and given the relatively large size of our own dataset, we adopt a VIF‑threshold of 10 based on industry experience, and deselect variables accordingly."},{"assumption":"The logistic link function is used for the write‑off risk (binary) model.","type":"model","stated":"explicit","quote":"In this logit model, the inverse link function 𝑔−1 is the logistic (sigmoid) function (1 + 𝑒−𝑥)−1 of a given 𝑥."},{"assumption":"The dichotomisation step maximises the Generalised Youden Index to obtain an optimal cut‑off.","type":"model","stated":"explicit","quote":"This dichotomisation step substantially improves the ability of the models to recover the empirical LGD-distribution, as will be shown later. Moreover, this dichotomisation step is based on maximising the Generalised Youden Index 𝐽𝑎(𝑐) across possible cut‑off values 𝑐∈[0, 1]."}]},"datasets":{"status":"available","items":[{"name":"credit card portfolio data","provider":null,"kind":"other","access":"unknown","period":null,"universe":null,"quote":"We benchmark this industry practice using credit card data against a more sophisticated (though classical) regression‑based approach."}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"weekly-monthly"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"forecast-accuracy","findings":[{"finding":"ROT‑based aggregates diverge substantially from the empirical loss rate, while regression‑based aggregates closely follow empirical trends.","kind":"predictive-accuracy","quote":"The ROT-based aggregates diverge substantially from the empirical rate over most time periods, whilst the regression-based aggregates follow the empirical trends much closer."},{"finding":"The grand MAE for the ROT‑based LGD estimates is 64.37%, indicating large prediction error.","kind":"performance","quote":"The grand MAE of 64.37% suggests that the ROT-based LGD-estimates 𝐺𝑣,𝑚 systematically fail to approximate the realised loss rates across all (𝑣, 𝑚), with substantial average errors."},{"finding":"The regression‑based two‑stage model can recover the typical U‑shaped empirical LGD distribution, which the ROT approach cannot.","kind":"predictive-accuracy","quote":"This regression‑based approach can demonstrably recover the typical characteristics of the 'U‑shaped' empirical LGD‑distribution, which the ROT‑based approach cannot do."}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":19,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T17:02:54Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}