Reading Rejection From the Loan Outcomes: Where SBA Approvals Cancel, Where Loans Fail, and What the File Cannot Say

A search for which SBA loans cancel or charge off by asset class is often a search for something the public loan record cannot supply: a list of the projects lenders turn down. The record cannot show rejection at all. It holds approvals only, so an application that a lender, a CDC or SBA declined leaves no row, and the reasons behind every decision, a refusal or an approval, stay out of public view ^{[1] [2]}. What the record can show is how each approval ended up: cancelled before any money was disbursed, disbursed and later charged off, paid in full, or still open. Read class by class, the first two of those outcomes show where approvals have stopped short of funding and where funded loans have been written off. They do not show why either happened, and neither of them is a measure of what lenders rejected.

This article reads those two outcomes for the asset classes the Loan Analytics series covers, in both the 7(a) and the 504 programs. It sets out what the file can and cannot show about rejection, prints the cancellation and charge-off floors of each class beside the floor of its program, measures how far each class sits from that floor, and turns the result into working rules for a feasibility study. It does not say why any class sits where it does. The file carries no reason for any outcome, the dataset's own report offers no explanation of the class patterns, and so this article offers none either.

A declined application leaves no row in the file

SBA releases a great deal about the loans it guarantees, and its disclosure rules say plainly what stays back. Under the appendix to SBA's FOIA regulations, the names and business addresses of recipients of approved loans are generally disclosed ^[3], and so are loan amounts, terms and maturities, along with statistics on lending and on defaults ^[4]. On the other side of the same appendix, non-statistical information on pending, declined, withdrawn or canceled applications is generally exempt ^[1]. The public loan file follows that line exactly: it opens at approval. An application a lender turned down, one a CDC never sent forward, and one SBA itself refused all leave nothing behind in it.

The reasons are held back along with the applicants. The same appendix places an applicant's financial statements, credit reports, business plans, marketing strategy and fiscal projections among the records generally exempt ^[2], and it reaches portions of loan applications and of loan officer's reports as well ^[5]. Most of what a credit decision turns on, and the substance of what a feasibility study puts in front of a lender, sits on the withheld side. Even for an approval that does appear, the analysis behind it does not. The series has a separate piece on the limits of the FOIA file that maps those gaps field by field. The point here is narrower: nobody can rank asset classes from this file by how often lenders refuse them, because refusals are never counted.

What the rules reveal about declines the data never shows

The regulations make clear that declines happen at more than one point, and that a declined borrower can come back. A CDC in the Premier Certified Lenders Program may process a 504 application that another CDC operating in the same Area of Operations declined or rejected, provided the applicant is located in that area and SBA has not itself declined that applicant's 504 application before ^[6]. An approval that reaches the file by that route carries no trace of the first refusal. No field records it, so the approvals in a class can include projects that one CDC turned away and another took on.

The standard a lender applies is written down, even though its results are not published. Under 13 CFR 120.150, creditworthiness is required of the applicant and of any Operating Company, and the rule asks for loans that are "so sound as to reasonably assure repayment" ^[7]. Three groups of criteria may be weighed, as applicable, when a loan is approved: the credit record of the applicant and of its Associates and guarantors, the applicant's earnings or cash flow, and whatever equity or collateral the applicant brings ^[8]. Lenders and CDCs must run that analysis with prudent, generally accepted commercial credit methods, consistent with the ones they apply to their own similarly sized loans that carry no SBA guarantee ^[9]. Each criterion is a ground on which an application can fail. None of them is a field in the public file.

A further rule decides who can appear in the file at all. The guarantee is reserved, by 13 CFR 120.101, for applicants who could not obtain the credit they seek on reasonable terms from any source outside federal, state and local government ^[10]. The lender or CDC certifies that, or shows it in another way, after weighing factors of conventional lending practice, and the rule's list of them includes the industry the applicant is in, a business age of two years or less, and how adequate its collateral is ^[11]. Every loan in a class cell has been through that test. The outcomes of a class therefore describe the businesses in that industry that came to SBA once credit on reasonable terms was not available to them elsewhere. They do not describe the industry as a whole, and a study that cites a class floor should not dress it up as an industry default rate.

Is there any public reading of how often lenders say no? None of the public sources this article draws on gives one for SBA loans by asset class. The nearest regular source is the Senior Loan Officer Opinion Survey that the Federal Reserve runs, whose October 2026 round covered changes in the standards, terms and demand for bank loans to businesses and households over the prior three months, which roughly means the third quarter of 2026 ^[12]. The survey treats lending standards as banks' policies for approving applications in a loan category, and lending terms as the conditions written into the loans it approves ^[13]. In that round banks reported standards on commercial and industrial loans basically unchanged for large and middle-market firms and tighter for small ones ^[14], and tighter standards, on net, in every commercial real estate category, from construction and land development to nonfarm nonresidential and multifamily ^[15]. The survey's small firms are those with annual sales below $50 million ^[16], a line the survey draws for its own purposes. What it reports is how banks describe the direction of their approval policies over the quarter ^[13]. That is not a count of declined applications, and it covers bank lending in general rather than SBA loans in a given asset class. A study may cite it as the climate in which an application is read. It is never a rejection rate for a class.

The two endings the file does record

With refusals out of reach, the file records two endings, other than repayment, that an approval can reach. The 7(a) status codes are spelled out in the data dictionary SBA publishes with the files, with CANCLD meaning cancelled, CHGOFF charged off and COMMIT undisbursed ^[17], and the same dictionary explains that the status of a loan that was disbursed but not cancelled, paid in full or charged off is exempt from disclosure and shown as EXEMPT ^[18]. A cancellation is an approval that ended before any disbursement. A charge-off is a loan that disbursed, defaulted and was written off. The series explains in its piece on cancellation against charge-off why the two can never be added together, and in its piece on approval-cohort floors why any rate built from this file can only understate. This article takes both points as settled.

The dataset measures each ending on its own cohort. The cancellation floor divides the approvals with a cancelled status by all approvals made in FY2017 to FY2024, which were 2 to 10 years old when the data was cut on 2026-09-30, and its ceiling adds every approval in that cohort still waiting on a first disbursement. Keeping cancelled approvals in the denominator follows SBA's own practice, since the agency's gross approval figures include loans cancelled in the year of approval and in later years ^[19]. The charge-off floor divides the loans charged off to date by the loans disbursed from FY2017 to FY2021 approvals, 5 to 10 years old at the same date. A second charge-off rate counts only the loans written off within five years of approval, which holds the horizon fixed for loans of different ages. SBA refreshes the FOIA data every quarter ^[20], and every rate here comes from the release dated September 30, 2026 ^[21].

Across each program as a whole, the 7(a) cancellation floor is 11.24 percent ^[22] under a ceiling of 12.24 percent ^[23], and the 504 floor is 13.60 percent ^[24] under a ceiling of 16.90 percent ^[25]. The 7(a) charge-off floor is 6.33 percent ^[26], with 3.72 percent ^[27] written off within five years of approval, and the 504 charge-off floor is 0.35 percent ^[28], with 0.17 percent ^[29] within five years. All of these are approval-cohort floors. They are the frame for every class figure that follows: a class is high or low only against the floor of its own program, measured on the same cohort.

These are not SBA's own loss statistics, which run on another basis. SBA's performance tables report charge-off rates by program as a percent of the unpaid principal balance outstanding when each fiscal year closes ^[30], and guarantee purchase rates on the same balance basis ^[31]. Both are program totals. A view by asset class exists only because the loan file carries each borrower's industry code, and every class figure in this article is built from that code. Table 1 gathers the figures the article leans on most, with the numerator and denominator behind each cell rate.

Loan Data Behind This Article

Every figure in this table comes from the Loan Analytics loan dataset, analytics-loan-data freeze-3 (commit 360c856), and carries its key. SBA rates are approval-cohort floors and are withheld when fewer than 30 loans stand behind them.

Twelve classes, two programs, and the cells that can carry a rate

The asset-class group of this series covers fifteen kinds of project. Eleven of them have a class of their own in the dataset, and the restaurant topic has two, so twelve classes carry the series' asset-class work. Each is assigned from the borrower's industry code by the method set out in the series piece on class assignment: gas station and truck stop, car wash, RV park and campground, self storage, assisted living without nursing care, multifamily, hotel and motel, marina, data center, fitness center, full-service restaurant and limited-service restaurant. Of the four remaining topics, RV and boat storage is folded into self storage, which the dataset's asset map says cannot be separated from it, while retail, glamping and short-term rental have no class of their own in that map, so this article leaves them out.

Each of the twelve classes has two national program cells, one for 7(a) and one for 504, and each cell carries its own rates. A rate is printed only when at least 30 loans stand behind it; below that the dataset withholds it, for the reasons set out in the piece on thin samples. Three of the twenty-four cells fall short somewhere. The 504 multifamily cell holds only 11 ^[39] approvals across FY2017 to FY2026, so both of its rates are withheld: fewer than 30 loans. The marina and data center cells under 504 print cancellation rates, but their cohorts of disbursed loans are too small for a charge-off rate, and those two charge-off rates are withheld: fewer than 30 loans. Every other rate in this article stands on 30 loans or more.

Several classes carry a limitation that changes how their rates should be read, and the asset map records each one. The gas station class cannot split out truck stops and travel centers. Hotels and motels share a single code, so limited service and full service cannot be told apart. The data center code also holds data processing and hosting businesses, so the class mixes loans for colocation buildings with loans to hosting and data processing firms and cannot separate the two. The multifamily class is thin, because SBA loans to residential lessors are rare and nearly stop from FY2024. A reader who skips those notes will misread the class.

Table 2 prints the 7(a) rates for all twelve classes, ordered by charge-off floor, with the program row at the foot. Every rate in it is an approval-cohort floor.

Where 7(a) approvals cancel

On cancellation the 7(a) classes sit in a fairly tight band around the program floor of 11.24 percent ^[22]. The highest in the series is multifamily at 18.37 percent ^[42], a rate that rests on only 196 ^[42] approvals in the cohort. Gas stations come next at 15.42 percent ^[32], on a far broader base of 5,142 ^[32] approvals, followed by marinas at 14.44 percent ^[46] and RV parks at 13.50 percent ^[45]. At the low end, hotels and motels have cancelled 8.76 percent ^[33] of their approvals, the lowest figure among the twelve, with assisted living at 10.21 percent ^[36] and car washes at 10.84 percent ^[43] close behind.

The 7(a) ceilings sit close to their floors. Gas stations, for instance, could reach no more than 15.67 percent ^[32] even if every pending approval in the cohort fell away, which says that few of them are still waiting and that the floor is near its final value. Marinas show the widest 7(a) range of the twelve, from 14.44 percent ^[46] to 16.04 percent ^[46]. At the other extreme the multifamily ceiling equals its floor, and the hotel ceiling of 8.81 percent ^[33] sits barely above the hotel floor. For a reader comparing classes, a narrow range means the floor can be quoted with some confidence about where it will settle, and a wide one means it cannot.

What does a cancellation say about a class? Less than the word suggests. It records that an approval ended before any money moved. It does not record who ended it or why, because SBA's disclosure rules keep back the non-statistical record of a cancelled application just as they keep back that of a declined one ^[1]. A gas station approval dropped by a borrower who found cheaper money and one abandoned because the project's own numbers stopped working leave the same status behind. The defensible reading is narrow: in a class with a high cancellation floor, a larger share of approvals did not reach disbursement, and a study for a project in that class should treat the stretch between approval and first disbursement as a period it has to document rather than one it can take for granted.

Where 7(a) loans fail

On charge-off the spread is far wider, and the order changes. Four classes sit above the program floor of 6.33 percent ^[26]: fitness centers at 9.48 percent ^[34], limited-service restaurants at 7.61 percent ^[35], data centers at 6.95 percent ^[40] and full-service restaurants at 6.67 percent ^[41]. The other eight sit below it, most of them well below: multifamily at 4.27 percent ^[42], car washes at 2.47 percent ^[43], assisted living at 1.44 percent ^[36], gas stations at 1.16 percent ^[32], hotels and motels at 0.99 percent ^[33] and self storage at 0.50 percent ^[44]. RV parks and marinas show no charge-off at all so far, with 274 ^[45] and 118 ^[46] disbursed loans behind those zeros.

A zero floor is still a floor. The RV park cell holds 110 ^[45] open loans whose status the file withholds, and the marina cell 37 ^[46], and any one of them can still be written off. A zero says that no loan in the cohort has been charged off yet. It does not say the class cannot lose money, and no study should describe it that way.

The within-five-years rate tells much the same story with the seasoning held level. On that measure fitness centers stand at 5.28 percent ^[34] against 3.72 percent ^[27] for the program, and hotels at 0.27 percent ^[33]. Multifamily climbs the order, to 4.27 percent ^[42], above the program on this measure though below it on the other, because 5 ^[42] of its 5 ^[42] charge-offs came inside five years. That figure rests on 117 ^[42] disbursed loans and on so few events that a single loan more or less would shift it visibly, a reminder that the 30-loan rule guards the denominator and says nothing about how many events stand in the numerator.

The data center figure needs its limitation printed beside it. Because the class code mixes colocation buildings with hosting and data processing businesses, the charge-off floor of 6.95 percent ^[40] cannot be read as the record of loans on data center buildings. The cell's median loan of $200,000 ^[40] is the smallest among the twelve 7(a) cells apart from multifamily, but the file cannot say what each borrower was financing, and a study for a colocation building should cite the figure only with that warning attached.

How far each class sits from its program floor

A ranking says which classes sit high. It does not say by how much. Table 4 measures six 7(a) classes against the program floors: the three classes of the series' SBA worked examples linked at the foot of this page (gas station, car wash and assisted living), the class with the lowest cancellation floor (hotel and motel), and the two classes with the highest charge-off floors (fitness center and limited-service restaurant). Its inputs, every one a loan figure with its key, are in Table 3. For each class the table prints the floor, the gap in percentage points, the floor as a multiple of the program floor and, for charge-offs, the number of loans the class would have written off had its disbursed loans charged off at the program floor.

Rounded as shown; lines may not foot by one unit in the last place.

Start with the cancellation rows. The gas station floor stands 4.18 points above the program's, a multiple of 1.37, and the hotel floor 2.48 points below it, a multiple of 0.78. Car washes, assisted living, fitness centers and limited-service restaurants all sit within about a point of the program floor in one direction or the other. On this measure, many of the series' classes behave much like the program as a whole.

The charge-off rows spread much further. Fitness centers carry a floor one and a half times the program's, 3.15 points above it, and their disbursed loans show 147 more charge-offs than the program floor would put on the same base. Limited-service restaurants sit 1.28 points above, which on a cohort of their size comes to 106 charge-offs more than the program floor. At the other end, hotels show 254 fewer write-offs than the program floor would produce on their disbursed loans, gas stations 152 fewer, car washes 52 fewer and assisted living 47 fewer.

Two cautions keep that arithmetic honest. First, the program floor includes these classes. It is an average over every 7(a) loan in the cohort, so each gap measures distance from that average and not from some independent standard of safety. Second, every count in the table is a floor. The hotel cell still holds 1,804 ^[33] open loans and the fitness cell 1,276 ^[34]; as those loans resolve, both floors can rise, and nothing in the file says the gaps will keep their present size.

High on one measure, low on the other

Set the two orders side by side and they disagree. Gas stations rank near the top of the twelve 7(a) classes on cancellation and in the lower half on charge-off. Fitness centers sit in the middle on cancellation and first on charge-off. Hotels and motels come last on cancellation and among the lowest four on charge-off. Multifamily, thin as it is, leads on cancellation and sits in the middle on charge-off. A study that took a class's cancellation floor as a sign of its credit record would place gas stations among the riskiest classes in the series, and their charge-off record does not bear that out.

The disagreement is not a puzzle that this file can solve. A cancellation is a transaction that never closed, and a charge-off is a loan that closed and then failed; a class can be prone to the first and resistant to the second, and the file has no field that would show why. The sensible use of the pair is to keep the two rates in separate columns and to bring each one to the part of a study it bears on, as the last two sections of this article do. Nor should either rate stand in for the refusals the file never sees ^[1]: a class with few cancellations and few charge-offs may still be one that lenders often turn away at the door.

The 504 record: more cancellation, almost no charge-off

The cells for the 504 program tell a different story on both measures. Table 5 orders them by cancellation floor, and every rate in it is again an approval-cohort floor.

In most classes the 504 cancellation floor stands above the 7(a) floor for the same class. Gas stations have cancelled 19.82 percent ^[37] of their approvals under 504 against 15.42 percent ^[32] under 7(a), self storage 19.15 percent ^[49] against 11.14 percent ^[44], and hotels 13.13 percent ^[53] against 8.76 percent ^[33]. RV parks and data centers are the two exceptions in the series. The marina cell has the highest 504 floor of all, 27.94 percent ^[48], on 68 ^[48] approvals, a modest base, though above the line at which a rate is withheld.

The 504 ceilings also sit much further above their floors than the 7(a) ceilings do. Self storage runs from 19.15 percent ^[49] to 23.76 percent ^[49], and gas stations from 19.82 percent ^[37] to 24.64 percent ^[37]. The distance between floor and ceiling is the part of the cohort's approvals that had drawn no funds by the end of September 2026, and until those approvals resolve, the final rate can land anywhere in that range. The series' cancellation piece explains why a 504 approval can wait so long. One rule bears directly on what a waiting approval faces at the end. SBA may, in its sole discretion, decline to close the Debenture, direct the loan's transfer to another CDC, or cancel its guarantee of the Debenture before its sale, if any of a listed set of events occurs ^[57], and the list includes SBA finding an unremedied material adverse change since approval, for instance a borrower whose finances have weakened, or finding that closing would put SBA at unacceptable financial risk ^[58]. A 504 approval can therefore be questioned again at the end of the process, and a project whose numbers have slipped in the meantime can stop there.

On charge-off the cells for 504 barely register. The program floor is 0.35 percent ^[28], and the highest class floors in the series, car washes at 1.00 percent ^[38] and full-service restaurants at 0.94 percent ^[52], rest on 4 ^[38] and 13 ^[52] written-off loans. Self storage and RV parks show none. With numerators that small, a ranking of 504 classes by charge-off floor says very little: a single further charge-off in the gas station cell, for example, would lift it past both of those classes.

Two features of the 504 file limit what its charge-off floor measures. Under 13 CFR 120.920 the third party loans in a 504 project must add up to no less than the 504 loan itself ^[59], and the 504 loan usually takes a second lien behind them on the project property ^[60]. The FOIA file gives the third party loan a field for its amount but records the SBA and CDC loan as the approval ^[61], and the charge-off amount it reports is the SBA and CDC balance written off ^[62]. A 504 charge-off floor is therefore a floor for the CDC's loan alone. Whatever happened to the bank loan ahead of it is not in the file, and a study should not read a low 504 floor as the record of the project's senior debt. The series' 504 program piece sets out the structure in full.

What the outcome record cannot say about why

Every pattern above is a description. The file holds no reason for a cancellation, a default or a charge-off, and the dataset's report gives no explanation of why one class sits above another. That leaves a long list of differences between classes that the cells do not test. The 7(a) cells differ in loan size: the median loan is $2,200,000 ^[33] in the hotel cell, $246,000 ^[34] in the fitness center cell and $297,000 ^[35] in the limited-service restaurant cell. They differ in how often the borrower operates under a franchise; the fitness center cell, for one, records a franchise code on 48.47 percent ^[34] of its approvals. They may also differ in the mix of approval years inside each cohort, in loan terms and in collateral. The dataset does not test whether any of these accounts for any part of a gap, and this article does not either.

Three further limits apply to any explanation a reader might reach for. Class membership comes from the borrower's industry code, so a class is only as clean as its code, and the limitations listed earlier say where it is not. The population is selected twice, first by the credit elsewhere test ^[10] and then by the lender's own judgment that the loan is sound enough to reasonably assure repayment ^[7], so the cells describe what happened to approved SBA borrowers in each industry and nothing wider. And the record moves in one direction as it ages: unless SBA revises a status it has already recorded, a cancellation floor can only rise as pending approvals resolve, and a charge-off floor can only rise as open loans do, so a pattern read from the September 2026 release ^[21] can shift in the next quarterly one ^[20].

Two archived Loan Analytics posts come at the same ground from other directions. Why Feasibility Studies Get Rejected: The Deficiencies SBA and USDA Reviewers Flag, and the Outline That Survives Them works from the review standards and the audit record, the side of a rejection that the loan file cannot see. SBA Loan Default Rate by Industry: Charge-Off Rates for 416 NAICS Codes ranks charge-off rates by industry code on an older cohort of 7(a) approvals that has largely resolved, which is a different basis from the floors in this article. Its rates and these should not be set against each other unless each basis is named. SBA's own FOIA rules treat statistical data on loans and defaults as disclosable ^[4], and both the archived post and this article are built on the loan files SBA publishes under those rules; they read the files at different dates and measure them differently.

What a study should do differently for a class with a high floor

A class floor describes a group of loans. It is never the probability that the project in front of a lender will fail, and it is never, on its own, a reason to decline one. What it can do is tell a study where to spend its evidence. Each recommendation below is a statement of the study's own method, not a description of what lenders require.

For a class whose charge-off floor sits well above its program's, such as fitness centers or limited-service restaurants under 7(a), the weight of the study belongs on the debt service coverage test and on how far that test can bend.

Conventions. This article follows the conventions of the Loan Analytics worked examples. Coverage is tested against the program floor the site's program page states, where the program has one, and against a stated 1.25x lender assumption, and the higher of the two binds. Equity is measured at the loan the cash flow supports, with fees recomputed on that loan. An acquisition is tested on the seller's historical earnings, not on projections.

Under those conventions a base-case coverage figure is a single point. The site's methodology page describes a method that recomputes the coverage ratio under each stress it applies, aimed at the assumptions with the most influence on the result, so that a single figure becomes a range ^[63]. For a high-floor class the study should push that range further than it would for a class near the program floor: a longer climb to stabilisation, a deeper revenue shortfall, an operating cost that runs over budget, each recomputed against both minimums. The series' piece on sensitivity and stress testing covers how shocks are sized. The point here is that a class record is a reason to show the downside in full, not a reason to change the method.

The study should also print the class record itself, with every limit attached. That means the class floor and the program floor side by side, the cohort years, the seasoning, the denominator, the open loans and any limitation in the class definition. A fitness center study that cites 9.48 percent ^[34] should say that it is a floor on 4,674 ^[34] disbursed loans from FY2017 to FY2021 approvals, with 1,276 ^[34] of them still open, and that it describes loans to fitness businesses that came to SBA rather than the fitness industry.

For a special-purpose project the downside runs past cash flow. The site's methodology page says that when a property would have little other use if the business failed, collateral and liquidation questions are weighed next to the going-concern projections ^[64]. A study may well treat a gas station or a car wash as that kind of building, and the low 7(a) charge-off floors of both classes are no reason to skip the step: a low floor says how often loans in the class were written off, not what a lender would recover on this one. Value is a question for the appraiser, and the series piece on how a study differs from an appraisal keeps the two documents apart.

For a class with a high cancellation floor, such as gas stations under either program, or self storage and car washes under 504, the weight of the study shifts to the stretch between approval and first disbursement. A feasibility study is one of the items SBA may make a condition of a loan, alongside appraisals and a survey ^[65], and the conditions of an approval are what has to hold before it can close. A study that lists which of its own assumptions remain open at approval, such as a contractor's price not yet fixed, an entitlement not yet granted or an equity source not yet documented, gives the lender a short list of what could end the approval before it funds. For a 504 project the study should date its market and financial evidence, so that a reader at debenture closing can see whether anything has moved since approval, because an unremedied material adverse change is among the events on which SBA may decline to close ^[58].

For a class whose rate is withheld, the study should say so in those words and use the program figure, or a neighbouring class, only with the substitution named. A withheld 504 multifamily rate is not a rate of zero, and a borrowed 7(a) rate is not a 504 rate. For a class whose floor is zero, such as RV parks and marinas under 7(a), the study should print the zero with its open-loan count and treat it as an absence of recorded write-offs, never as evidence of safety.

In every class the project's own evidence decides. The class record frames the questions, and the project's market, costs and cash flow answer them. The series' worked examples show that order at work: the 7(a) gas station example, the 504 express car wash example and the 7(a) assisted living example each test an assumed project's own cash flow against the coverage minimums and size the loan to what that cash flow carries.

Seven checks before a class figure goes into a study

A class figure quoted well answers each of the following without sending its reader anywhere else.

• The program. A 7(a) floor and a 504 floor for the same class are separate records. Compare each class only with its own program on the same cohort: 6.33 percent ^[26] is the charge-off floor that a 7(a) class is measured against, and 0.35 percent ^[28] the one that applies under 504.
• The ending. A cancellation is an approval that never disbursed, and a charge-off is a loan that disbursed and was written off. Neither one is a refusal, and refusals are not in the file at all ^[1].
• The cohort and its age. State which approval years the figure covers and how old they were when the data was cut, and give the ceiling with any cancellation floor, since a wide gap between them means the rate has not settled.
• The base. At least 30 loans must stand behind the rate, and the numerator matters too: a floor built on a handful of events, as most 504 charge-off floors are, can be reordered by a single loan.
• The class definition. Say where the code is shared, as it is for gas stations with truck stops and for data centers with hosting businesses, and say when a project type has no class of its own.
• The open loans. Print the count of loans still open in the charge-off denominator, since each one can still be written off and lift the floor.
• No cause, no forecast. The file records outcomes and withholds the business plans and projections behind them ^[2], so a class figure supports neither a statement about why the class performs as it does nor a probability for one project.

Read with those checks, the outcome record never tells anyone what lenders reject. It tells a study where approvals in a class have tended to stop short of funding and where funded loans have tended to fail, and it leaves the question of why to the project's own evidence.

Related Pages on Loan Analytics

• The Loan Analytics methodology series (all articles)
• How a feasibility study is done: the Loan Analytics methodology
• SBA feasibility studies (7(a) and 504)
• Loan data by state
• Loan data by asset class
• Worked examples (case studies)
• Worked Example: SBA 7(a) Gas Station Feasibility Study in Tennessee
• Worked Example: SBA 504 Express Car Wash Feasibility Study in Maryland
• Worked Example: SBA 7(a) Assisted Living Feasibility Study in Utah
• Why Feasibility Studies Get Rejected: The Deficiencies SBA and USDA Reviewers Flag, and the Outline That Survives Them
• SBA Loan Default Rate by Industry: Charge-Off Rates for 416 NAICS Codes
• Cancellation Is Not Charge-Off: The Two Ways an SBA Approval Can End Without Repayment
• The Approval-Cohort Charge-Off Floor: Why a Published SBA Loss Rate Can Only Understate

Commission a Feasibility Study

Loan Analytics prepares lender-grade feasibility studies for SBA 7(a), SBA 504, USDA Business and Industry, conventional and EB-5 financing, built on the methods this series sets out. A study is not an appraisal: it tests whether a project's market, costs and cash flow can carry the debt a lender is asked to approve. To discuss a study for your project, contact Loan Analytics.

Daniel Smith, MAI, Loan Analytics