Data and methods

Where the numbers come from, how we calculate them, what is wrong with the underlying data, and the tables to download.

Download the tables

You may quote, chart or republish these figures. Please credit The Mercantile Record and link to the page you took them from. The underlying loan records belong to the public.

What we use

SBA's 7(a) and 504 loan file. The U.S. Small Business Administration publishes a record for every loan it has guaranteed since fiscal year 1991, as a Freedom of Information Act release. It is posted on SBA's open data site and refreshed each quarter, usually about a month after the quarter ends. The version on this site is dated September 30, 2026.

The prime rate. Dates and levels of the bank prime loan rate come from the Federal Reserve's H.15 release, which we read through FRED: the dates of change, checked against the daily series. Prime has been 7% since September 17, 2026.

Government shutdowns. The dates of each lapse in federal funding, and the law that ended it, come from the Congressional Research Service report Federal Funding Gaps: A Brief Overview. The day funding runs out next comes from the law now in force, P.L. 119-103.

504 debenture rates. SBA's file has no interest rate for 504 loans. Effective rates by month of sale come from the interest rate history published by Growth Corp, a certified development company, which has no connection with this site. Other companies' fees may differ a little.

SBA's own rules. Fee schedules come from SBA's annual fee notices for the 7(a) and 504 programs, and lending rules from its Standard Operating Procedure 50 10. We read the documents themselves and cite the version in force. What a lender does when a loan is not repaid comes from a second manual, Standard Operating Procedure 50 57.

Method

Which loans. For current pricing we use 7(a) loans approved in the twelve months up to the date of the file. For outcomes we use loans approved in fiscal years 2010 through 2019, old enough for most of them to have been repaid or to have failed. SBA's fiscal year runs from October 1 to September 30.

What we leave out. Loans marked cancelled, and loans approved but never paid out. Many cancelled loans were never used, so their terms tell us little. The one exception is the study of funding times, which is about exactly those loans and counts them.

Rates. The "initial interest rate" in the file is the rate at approval. Most 7(a) loans are variable, so this is a starting rate.

Spreads. A loan's spread is its initial rate minus the prime rate on the first business day of the month it was approved. SBA's rules set a loan's base rate on that day. When finding that day we skip weekends but not bank holidays. In the period we cover, no change in prime fell on a date where that shortcut would matter.

Typical values. "Typical" always means the median: half of loans are above it and half below. "Middle half" is the range from the 25th to the 75th percentile.

Rate caps. SBA limits the spread on variable-rate loans by loan size. A loan is counted "at the cap" when its spread equals the limit to within half a hundredth of a point.

Fixed and variable. The file marks each loan as fixed or variable, as the lender reported it. Under SBA's rules a rate is fixed only if it holds for the whole term, so a loan with a fixed opening period should appear as variable. When we compare the price of the two kinds we use term loans only, since lines of credit are almost all variable. Fixed rates have a maximum of their own, set by a notice in the Federal Register, and we measure fixed-rate loans against that one.

State comparisons. States differ in the size of loans their businesses take, and size drives the rate. For each state we take the median spread in each of four loan-size groups and average them using the national share of loans in each group. Where a state has fewer than 15 loans in a group we use the national median for that group. States with fewer than 100 loans are not ranked.

Charge-offs. A loan counts as charged off if SBA's file gives it that status. "Within five years" counts the days from approval date to charge-off date. The share of dollars charged off divides SBA's "gross charge-off amount" by the gross amount approved.

Time to funding. Days are counted from a loan's approval date to its first disbursement date. We use the most recent fiscal year in which every loan has at least nine months of history. Shares "funded within" a number of days are shares of all approvals that were not cancelled, so loans not yet funded count against them.

SBA Express comparisons. Express loans are those the file labels "SBA Express Program". We compare them with loans labelled "Preferred Lenders Program" or "7a General", size group by size group, and leave out export loans, CAPLines and pilot programs, which follow rules of their own.

Business purchases. A loan counts as buying a business when its business-age field in the file reads "Change of Ownership". The label is the lender's. We count term loans only, not lines of credit. The file has carried that label only since fiscal 2018, so outcomes for these loans start there.

Collateral. The file records, as a yes or a no, whether the lender reported a loan as backed by collateral. Shares are of loans that were not cancelled. Where a loan failed, the share written off divides SBA's gross charge-off amount by the gross amount approved. We give a median only where at least 30 loans were charged off.

Franchise brands. A loan is a franchise loan when the file gives it a franchise code. SBA's codes and spellings changed twice in the years we cover, so one brand can appear under several names. We merged them by hand for every brand that could reach 100 loans, treating names as one brand when they carry the same brand name and the loans are in the same line of business. The download lists the names behind each brand. A listing that covers two brands is left out of both.

A brand's yardstick. For each of a brand's loans we take the charge-off rate of all 7(a) loans in the same size band approved in the same fiscal year, and average those rates. We say a brand's own rate is higher or lower than that only when chance would produce a gap that large less than 1% of the time, counting both directions. Otherwise we say there is no clear difference.

Shutdowns. We count 7(a) approvals by the date in the file, including loans cancelled later, on every day from six weeks before each shutdown to six weeks after it. A typical working day is the median number of approvals on working days, Monday to Friday and not a federal holiday, in the eight weeks that ended two weeks before the shutdown began, so that any rush before the deadline is left out. A typical week is the average week of those eight. The four weeks after a shutdown start on the first day with approvals. We checked that the records show no approval on any day of each shutdown.

504 loans. SBA's file of 504 loans is separate from the 7(a) files and has its own columns. The loan amount in it is the debenture alone, and the bank's loan sits in a column of its own. Current figures use loans approved in the twelve months up to the date of the file, cancelled loans left out. A loan counts as refinancing when its processing method is the 504 Refinancing Program, the 504 Commercial Real Estate Refinance Program, ALP Express Debt Refi or PCLP Debt Refinance. For the industries in the fee waiver we use the codes in SBA's notice, leaving out two trucking codes that count only for some kinds of freight.

504 structures. The file does not record the borrower's contribution. We infer it from the debenture's share of the debenture and the bank's loan together: up to 39.5% points to a debenture of 30% of the project, up to 43.5% to 35%, up to 47.0% to 40%, and anything higher is "something else". These limits fall between the shares that the manual's structures produce once the fees financed in the debenture are added. A borrower who put in more than the minimum could fall in any of them, so the result is an indication and not a count.

The wait for a debenture. Months run from the approval date to the first disbursement date, at 30.4375 days to the month. For new projects we use the four fiscal years ending with the last in which every loan has had the 48 months the manual allows for disbursement, cancelled loans left out. Refinancing is kept apart. Curves count loans not yet funded against the share; medians use funded loans. An industry needs 100 funded loans, and a state or a development company 25, for a median to be shown.

Rate moves on 504 loans. For each new project of 20 or 25 years approved since fiscal 2022 and funded by the date of the file, we take the effective rate of that term in the month of approval and in the month of the first disbursement. Refinancings and 10-year loans are left out. The series begins in October 2021.

504 charge-offs. We use loans approved in fiscal 2010–2019, leaving out those cancelled or never funded. The yardstick is the same as for franchise brands: the charge-off rate of 7(a) loans in the same size group and fiscal year, averaged over the 504 loans. The 7(a) files have no loans above $2 million in fiscal 2010, so those 504 loans take the rate of the next group down. The share of an industry's 7(a) loans of 20 years or more leaves out loans charged off, because of the problem with the term field described below.

Small groups. We do not report a lender with fewer than 100 loans, an industry with fewer than 1,500 loans in the outcome study, a franchise brand with fewer than 100, or a like-for-like lender rate based on fewer than 25 loans. In the study of fixed rates a lender needs 10 fixed-rate loans to be listed and 25 for its median rate to be shown. In the study of 504 loans a development company needs 25 loans in the twelve months to be listed, and an industry 300 loans in the outcome years.

Privacy. The file names every borrower. We publish only totals for groups of loans, never a named borrower.

Known problems with the data

We found these while working with the file. They shape what we will and will not claim.

  • Active loans are a black box. A loan that has been paid out and is neither repaid, cancelled nor charged off carries a single status that SBA withholds detail on. Healthy loans and loans in default look the same. Charge-off rates for recent years are therefore floors.
  • The term field cannot be trusted for failed loans. For loans that are paid in full, the term is a round number such as 84, 120 or 300 months. For charged-off loans it is scattered across odd values, which shows the field was changed after approval. We do not analyze charge-offs by loan term.
  • The lender shown is the current holder. If a loan was sold or the lender merged, the file shows the lender it is assigned to now. The 504 file does the same with development companies.
  • Disbursement dates bunch at month-ends. 23% of first disbursement dates on 7(a) loans fall on the last day of a month, far more than chance would give, so some lenders evidently report the month and not the day. We report medians and ranges of days and do not rank individual lenders by speed.
  • On 504 loans, the first disbursement is the debenture sale. Every first disbursement date in the 504 file falls on one date of its month. We read that date as the funding of the debenture, not as the day the project's money first moved. The interim loan pays out earlier, and the file has no date for that.
  • 504 jobs are lenders' estimates. SBA's data dictionary calls the jobs column "self-reported, good faith estimates by the lender" and says SBA does not check it.
  • Third-party lenders are free text. One bank can appear under several spellings in the 504 file. We merged the variants we could recognize, so some counts may still be split.
  • Industry codes are entered by lenders, and the national classification was revised in 2012, 2017 and 2022. We group by SBA's industry description so that renumbered codes stay together.
  • Some rates are wrong. A few loans show rates below 4% or above the legal maximum. They are too few to move a median, and we leave them in.
  • Collateral is a yes or a no. The file does not say what was pledged or what it was worth, and it has nothing on personal guarantees.
  • Franchise names are free text. One brand is spelled several ways and filed under different codes over the years, and 40 loans in the outcome study carry a code with no name. The share of loans with a code also fell in fiscal 2024, between the end of SBA's franchise directory in May 2023 and its return in June 2025, so franchise loans from those months are probably undercounted.
  • Some approval dates look moved by a day. In most months about 1% of approvals are dated on a weekend. From August 2023 to April 2024 it is between 10% and 26%, and in most of those months few are dated on a Monday. Our studies mostly count approvals by month or by fiscal year, which a day's difference barely changes, and no government shutdown falls near those months. A few single working days elsewhere show almost no approvals, and the file does not say why.
  • Approved is not applied. The file has no declined applications, no credit scores and no financial statements. It cannot say how hard a lender is to borrow from or why a borrower got the rate they did.

How often this is updated

When SBA posts a new quarterly file, we rerun every calculation and reread each article against the new numbers. Each article shows the date of the data it uses, and its change log records anything we altered.

Fee and rate pages change when SBA issues a new notice or the prime rate moves.

Corrections

If you think a number here is wrong, tell us. If you are right we will fix it, and the page's change log will say what changed and when.