Data-Driven Credit Risk and SME Portfolio Optimization: Evidence from U.S. SBA 7(a) Lending Data (Iconic Research and Engineering Journals)

Summary

• An empirical analysis of 463,775 resolved SBA 7(a) loans approved between fiscal years 2010 and 2021 shows an overall observed charge-off rate of 7.17%.p.1
• New or startup businesses record an 8.20% charge-off rate compared to 6.18% for established operating firms.p.1
• Unsecured exposures show an observed charge-off rate of 9.62%, whereas secured loans achieve 6.21%.p.1
• Loans with maturities of 60 months or less exhibit a 17.59% charge-off rate, compared to 1.26% for facilities with maturities of 241-360 months.p.1
• Among high-volume states, Florida records an 11.21% charge-off rate, while Minnesota demonstrates greater resilience at 4.97%.p.7
• Specialized processing methods like Community Express show a 17.55% charge-off rate, compared to 7.08% for the core 7a General program.p.7
• An out-of-time logistic regression model evaluated on 2019-2021 test data achieves an AUC of 0.846, with the top two predicted risk deciles capturing 68.23% of observed defaults.p.1

Key insights

1. Small businesses employ 62.3 million people and account for 43.5% of gross domestic product in the United States.p.1
2. The resolved-loan sample exhibits a mean gross approval amount of $362,445 and a median of $105,000.p.4
3. Loans sized up to $150,000 show an 8.23% charge-off rate, whereas exposures between $1 million and $5 million record 3.87%.p.5
4. Sectoral risk varies significantly, with transportation and warehousing recording an 8.74% charge-off rate while health care and social assistance reaches 4.67%.p.7
5. The highest predicted risk decile alone captures 43.92% of all observed default cases in the test sample.p.9
6. Combining interpretable credit analytics with relationship-manager judgement and explicit override protocols strengthens loan governance.p.10