thesisPrototype

Government data

SBA Disaster Charge-Off Rate FY2026

What will the SBA first print for FY2026 Disaster / Disaster charge-off rates as a percent of unpaid principal balance in Table 9 of Small Business Administration Loan Program Performance?

Forecast

1.6%
0.72%–3.25%
80% prediction interval
gpt-5.5Run record ↗

Cumulative probability

Chance that the outcome is at or below each value.

0%25%50%75%100%-0.6%1.6%5.72%Forecast value

Shaded band: 80% interval (0.72%–3.25%). Dashed line: point estimate (1.6%).

Based on probabilities reported by the forecasting agent.

Analysis

§SBA Disaster / Disaster FY2026 charge-off rate forecast

Framing: the target is the first official FY2026 SBA Loan Program Performance Table 9 print for the Disaster program row labeled Disaster, not the separate COVID EIDL row. The resolution date byte-echoes the Thesis resolve-by-bound of 2028-12-31; this is an outer bound, not an inferred SBA release date. The resolutionSourceUrl byte-echoes the registered methodology-announcement URL, and the required official announcement fetch returned HTTP 200 for 37099 bytes with response SHA-256 5a77a6bb8e74afdefcffd588fb37cab831ca69385b69d6911b900a99efaede64.

▸ Reported tool use: official.lookupmodel report
Fetched the SBA Loan Program Performance page and FY25Q3 zip, then extracted Table 9 Charge Off Rates from WDS_ChargeOffRates_Report_20250630.pdf.
↳ For Disaster / Disaster, fetched charge-off rates were 2016 1.46%, 2017 1.39%, 2018 1.09%, 2019 1.98%, 2020 1.42%, 2021 0.20%, 2022 1.97%, 2023 3.44%, 2024 3.06%, and 2025 Q3 0.90%.
▸ Reported tool use: official.lookupmodel report
Fetched the SBA FY25Q3 UPB table WDS_UPB_Report_20250630.pdf for the same Disaster / Disaster row.
↳ For Disaster / Disaster, fetched UPB was $9,155,822,926 in 2022, $9,379,479,606 in 2023, $9,789,692,235 in 2024, and $11,976,493,088 as of 2025-06-30.
▸ Reported tool use: official.lookupmodel report
Fetched the SBA FY25Q3 Charge Off Amount table WDS_ChargeOffAmount_Report_20250630.pdf for the same Disaster / Disaster row.
↳ For Disaster / Disaster, fetched charge-off amounts were $180,342,594 in 2022, $322,632,623 in 2023, $299,971,326 in 2024, and $107,714,599 as of 2025-06-30.
▸ Reported tool use: public.contextmodel report
Checked CRS disaster-loan program context for forward-looking credit risk in the disaster account.
↳ CRS table values for FY2025 disaster-loan assumptions included a 3.16% borrower interest rate, 29.39% default rate, and 29.17% post-default recovery rate; FY2024 values were 2.93%, 28.22%, and 27.76% respectively.

Reference class and base rate: the current official SBA Table 9 vintage through 2025-06-30 is the admissible base rate because no separate revised/backcast official print was found. The 2016-2024 full-year Disaster / Disaster history has a median of 1.46%, with a recent elevated cluster at 1.97%, 3.44%, and 3.06% in FY2022-FY2024, while the FY2025 partial-year print is much lower at 0.90%.

Variant control: all numeric anchors above are the SBA Table 9 charge-off rate as a percent of UPB for Disaster / Disaster. I excluded COVID EIDL values even though they appear under Disaster, because COVID EIDL is a separate row and the target field is Disaster / Disaster.

Prior/update/interval: persistence prior is the 2016-2024 full-year reference class centered near the 1.46% median, updated downward from the FY2022-FY2024 high-rate cluster by the FY2025 Q3 0.90% rate and larger 2025 UPB denominator, then nudged upward for lagged disaster-loan credit stress; the CRS default and recovery assumptions are directional context, not direct inputs to the Table 9 charge-off-rate calculation. Interval method is the elicited threshold ladder anchored by the fetched 0.90%, 1.46%, 1.97%, 3.06%, and 3.44% values, with the announced methodology-transition/regime consideration handled by widening the upper tail rather than applying any fabricated revision adjustment.

Ladder: P(X <= 0.4) = 0.04; P(X <= 0.6) = 0.07; P(X <= 0.8) = 0.12; P(X <= 1.0) = 0.20; P(X <= 1.2) = 0.30; P(X <= 1.4) = 0.40; P(X <= 1.6) = 0.50; P(X <= 1.8) = 0.60; P(X <= 2.1) = 0.70; P(X <= 2.5) = 0.78; P(X <= 3.0) = 0.87; P(X <= 3.5) = 0.93; P(X <= 4.2) = 0.97. Linear interpolation gives 10th percentile at 0.72, median at 1.60, and 90th percentile at 3.25.

Counter-considerations: upside risk for the rate is a delayed wave of default determinations on older disaster loans or a smaller-than-expected FY2026 UPB denominator, which would land above the interval if charge-offs resembled FY2023 while UPB stopped growing. Downside risk is continuation of FY2025's low run rate or unusually high recoverability, which could land below the interval if annual charge-offs stay near the 2025 Q3 pace against a large UPB base.

Review disposition: accepted both optional reviewer suggestions by naming the FY2025 Q3 vintage date more explicitly and clarifying that CRS default and recovery assumptions are directional context rather than direct Table 9 inputs; no required fixes were raised.

Key drivers

  • Current official FY2025 Q3 Disaster / Disaster charge-off rate is only 0.90 percent
  • Recent full-year charge-off rates were elevated in FY2022-FY2024
  • Disaster UPB rose to 11.98 billion dollars by June 30 2025, diluting a fixed charge-off dollar flow
  • COVID EIDL is a separate row and is not included in the resolver field

Tool evidence

This run has no captured tool responses. “Reported tool use” in the analysis is the model’s account, not an archived tool response.

Sources and resolution

Official source
Official SBA Loan Program Performance Table 9
Resolution date
December 31, 2028· outcome not recorded
Resolution rule
Resolve to the first official SBA Loan Program Performance print for FY2026, Table 9 - Charge Off Rates as a Percent of Unpaid Principal Balance (UPB) Amount by Program, field Disaster / Disaster. Use the percentage as printed by SBA, currently to 0.01 percentage point precision; do not use later revisions or same-day corrections unless they are the first official print itself.
Run details

The analysis is the model’s written report. Tool-use descriptions in that report are model claims; the activity artifacts contain the execution record.

thesis.analyst.ladder_v2 · gpt-5.5 · ladder_v2 · v2.5.7

pre-submit review · completed

Draft is publishable with a mostly complete resolver, prior, update, interval, and leakage discipline; only minor tightening would improve auditability.

  • info optional_suggestion: Name the exact SBA FY25Q3 file vintage/date in the historicalContext label or reasoning, since the sourceContext zip URL alone is less explicit than the extracted PDF names.
  • info optional_suggestion: Clarify that CRS default and recovery assumptions are directional context rather than direct inputs to the Table 9 charge-off-rate calculation.

disposition not applicable: Review disposition: accepted both optional reviewer suggestions by naming the FY2025 Q3 vintage date more explicitly and clarifying that CRS default and recovery assumptions are directional context rather than direct Table 9 inputs; no required fixes were raised.

disposition not applicable: Review disposition: accepted both optional reviewer suggestions by naming the FY2025 Q3 vintage date more explicitly and clarifying that CRS default and recovery assumptions are directional context rather than direct Table 9 inputs; no required fixes were raised.

Activity artifacts

Complete original trace

§SBA Disaster / Disaster FY2026 charge-off rate forecast

Framing: the target is the first official FY2026 SBA Loan Program Performance Table 9 print for the Disaster program row labeled Disaster, not the separate COVID EIDL row. The resolution date byte-echoes the Thesis resolve-by-bound of 2028-12-31; this is an outer bound, not an inferred SBA release date. The resolutionSourceUrl byte-echoes the registered methodology-announcement URL, and the required official announcement fetch returned HTTP 200 for 37099 bytes with response SHA-256 5a77a6bb8e74afdefcffd588fb37cab831ca69385b69d6911b900a99efaede64.

▸ Reported tool use: official.lookupmodel report
Fetched the SBA Loan Program Performance page and FY25Q3 zip, then extracted Table 9 Charge Off Rates from WDS_ChargeOffRates_Report_20250630.pdf.
↳ For Disaster / Disaster, fetched charge-off rates were 2016 1.46%, 2017 1.39%, 2018 1.09%, 2019 1.98%, 2020 1.42%, 2021 0.20%, 2022 1.97%, 2023 3.44%, 2024 3.06%, and 2025 Q3 0.90%.
▸ Reported tool use: official.lookupmodel report
Fetched the SBA FY25Q3 UPB table WDS_UPB_Report_20250630.pdf for the same Disaster / Disaster row.
↳ For Disaster / Disaster, fetched UPB was $9,155,822,926 in 2022, $9,379,479,606 in 2023, $9,789,692,235 in 2024, and $11,976,493,088 as of 2025-06-30.
▸ Reported tool use: official.lookupmodel report
Fetched the SBA FY25Q3 Charge Off Amount table WDS_ChargeOffAmount_Report_20250630.pdf for the same Disaster / Disaster row.
↳ For Disaster / Disaster, fetched charge-off amounts were $180,342,594 in 2022, $322,632,623 in 2023, $299,971,326 in 2024, and $107,714,599 as of 2025-06-30.
▸ Reported tool use: public.contextmodel report
Checked CRS disaster-loan program context for forward-looking credit risk in the disaster account.
↳ CRS table values for FY2025 disaster-loan assumptions included a 3.16% borrower interest rate, 29.39% default rate, and 29.17% post-default recovery rate; FY2024 values were 2.93%, 28.22%, and 27.76% respectively.

Reference class and base rate: the current official SBA Table 9 vintage through 2025-06-30 is the admissible base rate because no separate revised/backcast official print was found. The 2016-2024 full-year Disaster / Disaster history has a median of 1.46%, with a recent elevated cluster at 1.97%, 3.44%, and 3.06% in FY2022-FY2024, while the FY2025 partial-year print is much lower at 0.90%.

Variant control: all numeric anchors above are the SBA Table 9 charge-off rate as a percent of UPB for Disaster / Disaster. I excluded COVID EIDL values even though they appear under Disaster, because COVID EIDL is a separate row and the target field is Disaster / Disaster.

Prior/update/interval: persistence prior is the 2016-2024 full-year reference class centered near the 1.46% median, updated downward from the FY2022-FY2024 high-rate cluster by the FY2025 Q3 0.90% rate and larger 2025 UPB denominator, then nudged upward for lagged disaster-loan credit stress; the CRS default and recovery assumptions are directional context, not direct inputs to the Table 9 charge-off-rate calculation. Interval method is the elicited threshold ladder anchored by the fetched 0.90%, 1.46%, 1.97%, 3.06%, and 3.44% values, with the announced methodology-transition/regime consideration handled by widening the upper tail rather than applying any fabricated revision adjustment.

Ladder: P(X <= 0.4) = 0.04; P(X <= 0.6) = 0.07; P(X <= 0.8) = 0.12; P(X <= 1.0) = 0.20; P(X <= 1.2) = 0.30; P(X <= 1.4) = 0.40; P(X <= 1.6) = 0.50; P(X <= 1.8) = 0.60; P(X <= 2.1) = 0.70; P(X <= 2.5) = 0.78; P(X <= 3.0) = 0.87; P(X <= 3.5) = 0.93; P(X <= 4.2) = 0.97. Linear interpolation gives 10th percentile at 0.72, median at 1.60, and 90th percentile at 3.25.

Counter-considerations: upside risk for the rate is a delayed wave of default determinations on older disaster loans or a smaller-than-expected FY2026 UPB denominator, which would land above the interval if charge-offs resembled FY2023 while UPB stopped growing. Downside risk is continuation of FY2025's low run rate or unusually high recoverability, which could land below the interval if annual charge-offs stay near the 2025 Q3 pace against a large UPB base.

Review disposition: accepted both optional reviewer suggestions by naming the FY2025 Q3 vintage date more explicitly and clarifying that CRS default and recovery assumptions are directional context rather than direct Table 9 inputs; no required fixes were raised.

calibrated forecast · 80% CI
1.6%[0.72% · 3.25%]
Target metadata

Data point: sba.disaster.loan_program.charge_off_rate_upb.2026.first_print

More government data forecasts