{"paper_id":"2409.00296","title":"Credit Scores: Performance and Equity","authors":["Stefania Albanesi","Domonkos F. Vamossy"],"paper_date":"2024-08-30","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2409.00296v1","page_url":"https://thequant.space/flowcharts/credit-scores-performance-and-equity/","scores":{"empirical_rigor":7.5,"math_complexity":6.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a machine‑learning based credit‑scoring model that ranks consumers by predicted default probability as an alternative to conventional credit scores.","quote":"We develop a model to predict consumer default based on machine learning. We then use the ordinal ranking of consumers based on their predicted probability of default as a credit score alternative and compare the performance of conventional credit scores to our model.","summary":"Credit scores are critical for allocating consumer debt in the United States, yet little evidence is available on their performance. We benchmark a widely used credit score against a machine learning model of consumer default and find significant misclassification of borrowers, especially those…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Out‑of‑sample performance is evaluated with an 8‑quarter forward testing horizon, training on quarter t and testing on quarter t+8.","type":"execution","stated":"explicit","quote":"Given the 8‑quarter horizon for the prediction problem, we allow for fully out‑of‑sample performance evaluation by separating our training/validation and testing data by 8 quarters."},{"assumption":"Credit‑scoring companies are prohibited by law from using protected attributes such as race, gender, etc.","type":"other","stated":"explicit","quote":"Credit scoring models are also restricted by law from using information on race, color, gender, religion, marital status, salary, occupation, title, employer, employment history, nationality."},{"assumption":"The Experian data are nationally representative and randomly drawn from the universe of borrowers with an Experian credit report.","type":"data","stated":"explicit","quote":"The data comprises over 200 variables for a nationally representative panel of 1 million households, constructed with a random draw from the universe of borrowers with an Experian credit repo…"},{"assumption":"The hybrid deep neural network/gradient boosting model is the best‑performing model for the given prediction problem and data.","type":"model","stated":"explicit","quote":"making this combination the best‑performing model in this class for our specific prediction problem and data."}]},"datasets":{"status":"available","items":[{"name":"Experian credit file data","provider":"Experian","kind":"alternative","access":"commercial","period":"2004Q1‑2015Q4","universe":"nationally representative panel of 1 million households","quote":"We use anonymized credit file data from the Experian credit bureau. The data is quarterly. It starts in 2004Q1 and ends in 2015Q4. The data comprises over 200 variables for a nationally representative panel of 1 million households, constructed with a random draw from the universe of borrowers with an Experian credit re"}],"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":"Conventional credit scores misclassify borrowers, with Deep Subprime, Subprime and Near‑Prime default rates of 68%, 44% and 22%, showing inability to differentiate risk.","kind":"performance","quote":"The realized default rates for Deep Subprime, Subprime, and Near Prime borrowers are 68%, 44%, and 22%, respectively. This suggests that the credit score is unable to differentiate between borrowers with substantially different default risks."},{"finding":"The machine‑learning model’s predicted default rates deviate from realized rates by no more than five percentage points within any risk category.","kind":"predictive-accuracy","quote":"By contrast, the discrepancy between predicted and realized default rates for our model is never more than five percentage points within any risk category."},{"finding":"The ML model places more weight on factors that improve prediction for thin‑file, low‑income, young and minority borrowers, leading to equity gains.","kind":"other","quote":"better ability to deliver accurate predictions when confronted with low quality data, such as thin files, low credit mix and a history of default, attributes that are prevalent among these traditionally marginalized populations."},{"finding":"Payment history and amounts owed explain 35% and 30% of credit‑score variation respectively, while the ML model relies on a different factor weighting.","kind":"statistical","quote":"Payment history explains 35% of the credit score variation and amounts owed 30%. Length of credit history explains 15% of the credit score variation, while credit mix and new credit each explain 10% of the variation."},{"finding":"Using the ML‑based ranking improves predictive performance especially for borrowers at the bottom of the credit‑score distribution.","kind":"performance","quote":"We have shown that ranking consumers by predicted default risk based on our model delivers significant performance improvements, particularly for borrowers at the bottom of the credit score distribution."}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":28,"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":"2026-11-14T18:03:14Z"},"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/"}