Default, Fraud, and Fraud Classification in UK COVID-19 Loan Schemes: Evidence from One Million Guaranteed Loans

We study default outcomes and fraud classification in UK COVID-19 government-guaranteed lending programmes using loan-level data of 1,006,579 limited-company facilities under the Bounce Back Loan Scheme (BBLS) and Coronavirus Business Interruption Loan Scheme (CBILS). We estimate logit models of loan outcomes, Cox proportional hazard models of time-to-default, and Accelerated Failure Time (AFT) models that quantify timing effects under four parametric survival distributions. We distinguish all-cause default, insolvency-related default, and defaults that lenders classify as fraud. Three findings stand out. First, scheme design dominates risk: BBLS loans—issued under self-certification with minimal verification—show fraud odds approximately 160 times those of CBILS recipients, with Cox hazard ratios reaching 170. AFT models under the preferred generalised gamma specification show BBLS borrowers default 37 per cent sooner than CBILS recipients, with loan-to-sales ratio being the strongest predictor of accelerated failure, compressing survival time by 65 per cent per unit increase. Second, conventional credit risk indicators maintained predictive power despite pandemic disruption: probability of default scores, loan leverage, director experience, and board composition discriminate between outcomes across logit, Cox, and AFT specifications. Director experience and board size protect against accelerated failure, with governance effects compounding over time. Third, we document systematic under-classification of fraud at challenger banks and alternative finance/fintechs. Given default and controlling for borrower characteristics, these lenders classify fraud at only 40–50 per cent of the rate at main banks. AFT timing evidence reinforces this: challenger bank defaults cluster 30 per cent earlier than main bank defaults, consistent with rapid enforcement without fraud investigation. Propensity score analysis indicates approximately 2,500 fraud cases—representing £70 million—went undetected in challenger portfolios. We identified 1,904 cases (£76m in loans) as unambiguously fraudulent based on Companies House filings—loans to companies dissolved pre-pandemic or incorporated after scheme announcement—yet lenders rarely flagged these. Main banks screened out 28 per cent of such applicants at origination and classified 14 per cent of defaults as fraud; challenger banks screened out only 22 per cent and classified just 1.4 per cent as fraud. These findings reveal trade-offs in crisis lending design and highlight an overlooked issue in delegated guaranteed schemes: the state relies on heterogeneous