The loss baseline and the impact baseline that together price and measure the instrument: loss emergence curves, loss drivers, guarantee structure, and the impact metrics that must be evidenced.
How would we price this instrument, and how would we prove it worked?
Everything below is the evidence for those three sentences. You can read only this box and still follow the argument. Underlined terms open a plain definition — the full glossary has them all.
EMBI has to satisfy a rating committee and a set of foundation guarantors at the same time. Those are different evidentiary burdens and they need different baselines. This document establishes both from data in hand, and names precisely what is missing.
For the rating agency, the structuring bank and the guarantors underwriting the credit enhancement.
The Moody's SME asset-backed methodology in the data room asks for static-pool vintage cumulative loss data, a loss timing curve for extrapolating unseasoned pools, and evidence on underwriting and servicing practice. The SBA FOIA register can produce all three directly, because it records approval date, approval amount, charge-off date and charge-off amount for every loan.
What follows is that analysis: eighteen 7(a) vintages and nineteen 504 vintages, each tracked from origination to terminal outcome.
For the mission investors, the foundation guarantors and the initiative's own reporting obligation.
"Tracking regional economic velocity and small business growth" is listed as a key activity in the business model. That is not yet a measurement framework. A framework requires named indicators, defined denominators, identified sources, a measured value today, and a cadence.
Section 04 proposes sixteen indicators across four families and assigns each a baseline value where the data supports one — and a status flag where it does not.
The 7(a) records available here end at fiscal year 2009. There is no post-crisis 7(a) vintage evidence in this dataset at all. Every loss curve below is anchored on vintages originated before the 2010 credit-box reforms, before SBA One, before the current secondary-market structure, and before the 2021–2023 origination surge.
A rating committee will ask for recent vintage performance and it is not here. Obtaining the SBA's FY2010–present 7(a) FOIA file is the highest-priority data acquisition of the research period; nothing else in this document is as load-bearing.
Each cohort is treated as a static pool: total dollars approved in a fiscal year, tracked forward, with cumulative charge-off expressed as a percentage of the original balance. Cancelled and never-funded approvals are excluded — they never entered a pool.
Each line traces how much of one year's loans was eventually lost to , as the loans aged.
Lines spread far apart means loss is unpredictable from year to year, which matters more than the average.
Each line is one origination cohort. Loss is charged-off principal as a share of original approved dollars. The dispersion between cohorts — not the average — is the underwriting question.
A bar for each loan cohort showing the total share of its balance eventually lost to .
Taller bars are worse cohorts; look for how far the tallest bars sit above the average line.
Terminal cumulative loss as a share of original approved balance.
How quickly losses show up after a loan is made, year by year, on average.
A steep early rise means losses appear fast; a newer, unseasoned pool can be projected forward using this shape.
Share of a cohort's ultimate loss realised by each year of seasoning, averaged across fully seasoned vintages. This is the extrapolation anchor for pricing a pool that has not run off.
| Program | Vintages | Mean | Median | SD | Min | Max | P90 | CV |
|---|---|---|---|---|---|---|---|---|
| 7(a) | 18 | 9.50% | 7.71% | 5.76% | 3.45% | 23.93% | 18.98% | 0.606 |
| 504 | 19 | 8.10% | 6.31% | 6.44% | 1.70% | 22.88% | 17.71% | 0.796 |
Seasoned vintages are those with at least fifteen years of observation as of 31 December 2025. Coefficient of variation is the standard deviation divided by the mean — a direct measure of vintage risk, and the input that drives the gap between expected and stressed loss in a Monte Carlo portfolio simulation.
Mean ultimate gross loss across eighteen 7(a) vintages is 9.50% of original balance. The best cohort lost 3.45%. The worst lost 23.93%. A pool underwritten to the average would have been wrong by a factor of seven between its best and worst year.
Vintage explains the most variance, and EMBI cannot choose its vintage. But the next three drivers — origination channel, loan size and term — are all underwriting choices, and the spread between them is wide enough to move a rating.
Restricted to fully seasoned, pre-boom vintages so that channel is not confounded with the crisis cohorts. Severity is charged-off principal as a share of the original balance of loans that charged off.
| Channel | Loans | Loss (amt) | Loss (count) | Avg loan | Severity |
|---|---|---|---|---|---|
| Community Express | 13,463 | 19.13% | 36.58% | $25,896 | 69.4% |
| International Trade Loans | 2,092 | 14.34% | 26.10% | $436,648 | 49.2% |
| SBA Express Program | 171,664 | 12.89% | 18.14% | $51,243 | 76.4% |
| Low Documentation Program | 124,986 | 10.42% | 15.81% | $69,326 | 60.4% |
| Preferred Lenders Program | 130,868 | 7.63% | 13.47% | $387,290 | 55.8% |
| 7a General | 86,532 | 7.59% | 13.59% | $314,556 | 54.7% |
| Certified Lenders Program | 33,162 | 5.32% | 10.04% | $300,938 | 51.9% |
| 7a with EWCP | 2,200 | 2.63% | 4.55% | $407,540 | 59.0% |
for groups of loans sorted by how large the loan was.
Longer bars mean that size group lost a bigger share of its money to .
for groups of loans sorted by how many years they were scheduled to run.
Longer bars mean loans of that repayment length lost a bigger share of their money.
for loans grouped by the borrower's industry.
Longer bars mean that industry lost a bigger share of its loan dollars to .
EMBI plans to originate through MDIs and CDFIs rather than through the high-throughput national Express channels. On this evidence that is a credit-positive design choice, not merely a mission one: Certified Lender Program originations lost 5.32% of original balance against 12.89% for SBA Express and 19.13% for Community Express, over identical vintages.
The corollary is that the channel advantage has to be preserved in the origination agreements. A CDFI network operating an Express-style credit box would inherit Express-style losses.
Two structural findings that change how the pool should be described to investors. The first concerns where credit risk actually sits. The second is a correction to the regional story.
Across the funded 7(a) book the weighted guarantee rate is 0.73% of approved dollars. When a loan charges off, SBA reimburses the guaranteed share, so the loss falls on the unguaranteed strip in proportion to its size.
The consequence is that the guaranteed portion carries United States government credit risk, not small-business credit risk. A pool of guaranteed strips has an expected loss close to zero and is priced off the sovereign, not off the loss curves above. A pool of unguaranteed strips carries the full 9.50% mean and 23.93% tail.
Which of the two EMBI holds is the single largest determinant of its rating, and the strategy documents do not yet specify it. This should be resolved before any loss modelling is commissioned.
| Approval size | Guarantee rate | Loans |
|---|---|---|
| <$150K | 0.69% | 616,032 |
| $150–350K | 0.74% | 139,092 |
| $350–700K | 0.76% | 79,249 |
| $700K–1.5M | 0.73% | 44,167 |
| $1.5M+ | 0.68% | 6,671 |
Yearly for the regional book compared with the national book, cohort by cohort.
Where the regional line sits above the national line the region did worse that year, and below means it did better.
Regional 7(a) performance depends on how it is measured. On a full-period count basis the region sits below the national rate — 0.2% against 0.2%. On a dollar-weighted basis it reverses: 0.1% against 0.1%.
Decomposed by vintage the picture is clearer and more useful. The region lost more than the nation in every cohort from FY1992 through FY2004, and substantially less in the FY2006–FY2008 crisis cohorts.
Kansas and Missouri are a lower-beta, higher-base-rate market: worse than national in calm conditions, materially better in a downturn. For a ten-year bond that is a defensible feature — but it cannot be sold as simple regional outperformance, because on dollars across the cycle it is not.
Fiserv small business revenue data, 91 months to July 2026, reaches the same conclusion from a different direction. At the April 2020 trough the national real sales index fell to —; Kansas bottomed at 89.3 and Missouri at 89.8, ranking among the shallowest of 50 states. The region absorbed roughly half the national drawdown — the same asymmetry the FY2006–FY2008 loan vintages show, thirty years apart and measured on revenue rather than losses.
Two independent series, two different crises, one structural characteristic. This is the strongest cross-validated finding in the analysis and it should be underwritten as a property of the collateral pool.
Impact claims fail diligence when they lack denominators. “Loans made” is an output; “loans per thousand jobs in the industry” is a measure. The Jackson County employment and survival file supplies the denominators for a real one.
What share of new local businesses are still open in each year after they started.
A steeper downward slope means more new businesses are closing sooner.
Share of establishments founded in the preceding five years still operating. Census Business Dynamics Statistics.
Each dot is an industry, plotting how many people it employs against how much lending it has received.
Dots low on lending but high on employment are industries getting relatively little SBA credit.
2025 employment against cumulative SBA lending, rolled to two-digit NAICS so that code revisions between 2002 and 2022 do not break the join. Bubble size is the sector's location quotient.
| Sector | Employees | LQ | Growth | 7(a) loans | Loans/1k jobs |
|---|---|---|---|---|---|
| Health Care | 56,074 | 1.13 | 2.5% | 153 | 2.73 |
| Accommodation & Food | 35,609 | 1.07 | -1.4% | 156 | 4.38 |
| Professional Services | 34,446 | 1.57 | -1.9% | 185 | 5.37 |
| Retail Trade | 29,448 | 0.96 | -1.4% | 303 | 10.29 |
| Construction | 23,221 | 1.29 | 4.7% | 197 | 8.48 |
| Finance & Insurance | 16,090 | 1.53 | -2.8% | 27 | 1.68 |
| Administrative Services | 14,194 | 0.77 | 1.7% | 104 | 7.33 |
| Manufacturing | 11,111 | 2.02 | -0.7% | 147 | 13.23 |
| Other Services | 10,692 | 1.23 | -0.8% | 207 | 19.36 |
| Wholesale Trade | 9,059 | 1 | -2.0% | 109 | 12.03 |
| Management of Companies | 7,133 | 1.14 | 1.8% | 1 | 0.14 |
| Arts & Recreation | 6,664 | 2.27 | -2.9% | 30 | 4.5 |
| Educational Services | 6,291 | 1.08 | 1.6% | 18 | 2.86 |
| Transportation | 5,525 | 0.68 | -7.1% | 46 | 8.33 |
| Information | 5,311 | 1.51 | -0.2% | 33 | 6.21 |
| Real Estate & Rental | 5,302 | 1.08 | 0.4% | 36 | 6.79 |
| Agriculture | 618 | 1.52 | 11.3% | 1 | 1.62 |
Loans per 1,000 jobs uses cumulative 7(a) approvals FY1992–FY2009 against 2025 employment, so it measures historical credit intensity against a current base — directionally valid, not a rate. Management of companies and similar sectors show near-zero penetration because they are largely outside SBA size and eligibility standards, not because they are underserved.
Geography. Jackson County only. It cannot speak to the Johnson–Wyandotte access gap, which is the sharpest equity finding in the regional analysis and sits entirely outside this county.
Suppression. BLS omitted every industry under 500 employees and all non-private employment, leaving 276,788 jobs across 107 industries. The smallest firms — precisely EMBI's borrower profile — are systematically excluded from the denominator.
NAICS vintage drift. SBA codes each loan to the NAICS edition in force at approval. At four digits the join fails badly. Two-digit rollup is stable; four-digit analysis requires a 2002–2022 crosswalk that does not yet exist in this workspace.
FY2020–FY2026 7(a) records, as of 30 June 2026, are now in the analysis: 317,155 funded loans and $171B. They resolve several open items in this framework and leave the most important one open.
Current market structure. Origination is now measurable: who lends, at what size, through which channel, in which industries. The 7(a) lender list in the targets workbook is a call list rather than a qualification list.
Pricing. Initial interest rate is populated on essentially all modern records against almost none of the legacy file. Median 9% across the period, moving from 6% in FY2020 to 9.5% in FY2026.
Borrower stage. A dimension the legacy file did not carry, and it discriminates: change-of-ownership loans run 4.5% against 13.8% for businesses under two years old.
Mission-lender channel. Community Advantage is visible for the first time — it did not exist when the legacy file ends.
The seasoned post-crisis vintage. FY2010–FY2019 7(a) is not in the dataset. There is now a ten-year hole between FY2009 and FY2020, and it contains the only cohorts that are both post-reform and fully seasoned.
Loss curves. Only 23.7% of the modern book has resolved. At a 4.2-year median time to charge-off, and with 44% of ultimate loss emerging by year five, the FY2020–FY2026 cohorts cannot anchor a loss model. Cumulative charge-off stands at 0.52% of balance and will rise by a large multiple.
Every vintage curve in Section 01 still terminates on the FY2007 peak. That has not changed.
The programme has moved decisively toward Preferred Lenders and away from low-documentation channels. On the FY1994–FY2005 evidence in Section 02 that is credit-positive — but the modern mix has not been tested through a downturn.
| Channel | Modern share | Legacy share |
|---|---|---|
| Preferred Lenders Program | 55.2% | 21.6% |
| SBA Express Program | 37.0% | 38.6% |
| 7a General | 6.2% | 14.6% |
| Community Advantage Initiative | 0.7% | — |
| Working Capital CAPLine | 0.3% | — |
| International Trade Loans | 0.3% | 0.4% |
The channel closest to EMBI's intended origination model, and the worst performing of the modern era.
| Value | |
|---|---|
| Loans | 2,240 |
| Gross approved | $355M |
| Average loan | $158,357 |
| Charge-off rate | 28.3% |
| Share resolved | 36.1% |
| Lenders in region | 86 |
Charge-off is measured against a largely unseasoned book — 36% resolved — so this is a directional signal, not a terminal loss rate.
The guarantee rate has risen from 72.75% of approved dollars in the legacy book to 75.54% today. On a pool of guaranteed strips that raises the sovereign-risk share further; on unguaranteed strips it thins the piece that absorbs the loss. The strip question in Section 03 is now more consequential, not less.
Sixteen indicators across four families. Each has a definition, a source, a cadence and — where the data supports it — a measured baseline value as of this research period. Indicators without a baseline are not aspirational; they are work items.
| Indicator | Definition | Source | Baseline | Cadence | Status |
|---|---|---|---|---|---|
| ORIGINATION & ACCESS | |||||
| Median loan size | Median gross approval | SBA FOIA | Region $230,550 · US $200,000 | Quarterly | Baselined |
| Share collateralised | Loans with collateral pledged ÷ total | SBA FOIA | 91.0% region · 81.9% US | Quarterly | Baselined |
| COVERAGE & CONCENTRATION | |||||
| Counties with zero approvals | Count of county-equivalents with no loan | SBA FOIA | 504: 353 US · 25 in KS/MO | Annual | Baselined |
| Gini across counties | Gini of loan counts, zeros included | SBA FOIA | 7(a) 0.813 · 504 0.843 | Annual | Baselined |
| Top-decile county share | Share of loans in top 10% of counties | SBA FOIA | 7(a) 73.7% · 504 76.3% | Annual | Baselined |
| Lending-to-formation correlation | Corr of loans per capita with NEB formation | SBA FOIA + Kauffman NEB | 0.55 (FY05–09) → 0.209 (FY20–24) | Annual | Baselined |
| MDI/CDFI origination share | Approvals via mission lenders ÷ total | SBA FOIA + NBA + CDFI list | 0.78% regional | Annual | Partial |
| CREDIT PERFORMANCE | |||||
| Vintage cumulative gross loss | Charged-off principal ÷ original balance | SBA FOIA | 7(a) mean 9.50% · σ 5.76 | Annual | Partial |
| Loss timing curve | Share of ultimate loss by seasoning year | SBA FOIA | 44% by yr 5 · 84% by yr 10 | Annual | Baselined |
| Channel loss differential | Gross loss by origination channel | SBA FOIA | CLP 5.32% vs Express 12.89% | Annual | Baselined |
| Community Advantage benchmark | Charge-off on the mission-lender channel | SBA FOIA | 28.3% on 36% resolved | Quarterly | Partial |
| Revenue-to-default leading indicator | Lagged corr, revenue index vs charge-offs | Fiserv + SBA FOIA | Tested; no usable signal | Monthly | Not yet |
| ENTERPRISE OUTCOMES | |||||
| Rate of new entrepreneurs | Adults starting a business monthly | Kauffman KESE | KS 0.220% · MO 0.312% · US 0.373% | Annual | Baselined |
| Opportunity share | New entrepreneurs not previously unemployed | Kauffman KESE | KS 80.0% · MO 81.2% | Annual | Baselined |
| Startup early survival | New establishments active after one year | Kauffman KESE | KS 75.5% · MO 74.3% · US 78.5% | Annual | Baselined |
| Application-to-employer conversion | Applications becoming employer businesses | Kauffman NEB | KS 10.27% · US 8.39% | Annual | Baselined |
| Young-firm pay vs national | EJI compensation, firms aged 0–1 | Kauffman EJI | 53% (KC) vs 65% US | Annual | Baselined |
| Small business real sales | Index of POS revenue, 2019 = 100 | Fiserv | KS 113.4 · MO 113.1 · US 109.5 | Monthly | Baselined |
| Demographic composition of borrowers | Race, sex, veteran status of approved borrowers | — | — | — | Not yet |
Measured value in hand from data already analysed. Can be reported to guarantors at close and re-measured on the stated cadence.
Measurable for part of the geography, part of the period, or with a known bias that must be disclosed alongside the number.
Requires a data acquisition named in Section 06. Should not appear in investor materials until sourced.
Ordered by how much each unblocks. Items already in hand are marked; the first two remain prerequisites for a credible rating conversation.
The FY2020–FY2026 file is in hand and analysed in Section 07. FY2010–FY2019 is not, leaving a ten-year hole between FY2009 and FY2020 that contains the only cohorts which are both post-reform and fully seasoned. Every vintage loss curve in Section 01 still terminates on FY2007.
This remains the single highest-value acquisition outstanding. It is the prerequisite for a seasoned post-crisis loss curve, for a current lender-performance ranking, and for the revenue-to-default lead-lag test that failed on the unseasoned modern book.
Not a dataset — a decision, and open. Expected loss on a pool of guaranteed strips is near zero and prices off the sovereign; on unguaranteed strips it carries the full 9.50% mean and 23.93% tail documented here. The strategy materials describe the guarantee as a risk mitigant without stating which side of it the bond holds.
The modern book raises the weighted guarantee to 75.5% from 72.75%, which thins the unguaranteed strip further and makes the specification more consequential, not less. It should precede any commissioned loss model.
All three series are in the analysis. NEB supplies the formation funnel and the tightest available denominator for 7(a); KESE supplies the rate of new entrepreneurs, opportunity share and startup survival, with demographic breakdowns at national level; EJI supplies county and metro young-firm employment, job quality and pay.
Together they establish that Kansas and Missouri convert entrepreneurial activity into SBA credit at or above the national rate on three independent denominators, and that the binding regional constraint sits upstream of credit at business formation.
The National Bankers Association MDI network has been joined on FDIC certificate number — an exact join, documented in the companion mission-lender baseline. A CDFI candidate list has been assembled from a secondary commercial source and three regional depositories verified by name, state and institution type.
What is still missing is the authoritative CDFI Fund certified list, which carries identifiers permitting an exact join for depositories. The current source omits Kansas entirely and carries no certification dates, so any figure derived from it is provisional.
Jackson County employment and firm-survival data is in hand. Johnson, Wyandotte, Clay, Platte and Cass are not. EJI reaches county level and has partially closed this gap — it resolves the Johnson–Wyandotte access question — but it measures employment and job quality rather than formation or survival.
Adding the remaining counties would give a metro-wide industry credit-penetration baseline rather than a single-county illustration, and would extend the five-year survival series beyond Jackson.
Ninety-one months of state and sector revenue data, benchmarked to 2019, seasonally adjusted and CPI-deflated. It supplies the only revenue measure in the framework and independently confirms the low-beta regional finding through the COVID trough.
It does not yet function as a leading indicator. Tested against thousands of charge-off events the correlations run the wrong direction, because the modern book is only 23.7% resolved and grew rapidly across the window. Re-test once FY2010–FY2019 is in hand.
Still outstanding. SBA codes each loan to the NAICS edition in force at approval; QCEW and EJI report on current editions. At two digits this is harmless and all sector analysis here uses two-digit rollups for that reason. At four digits it silently destroys the join.
Census publishes concordance tables for each revision. Building this once makes four-digit industry analysis reliable rather than approximately right.
Still outstanding, and now lower priority. Kauffman NEB supplies a formation denominator at state level and EJI supplies an employment denominator at county level, which covers most of what CBP was wanted for.
CBP would still add a stock measure — the count of existing establishments — enabling a penetration rate against the standing business population rather than against annual formation.
This is document 3 of 5 in the reading path. Back to the Research index.