Two readings on the same economy

I signed a personal guarantee once. One page. It took me about forty seconds to read.

The sentence that matters is short. If the business can't pay, you pay.

You, personally. Your savings, your paycheck, and usually your house.

Everyone signs it anyway. There's a seller across the table and a lawyer waiting and forty more pages behind it.

Between 2021 and 2025, a great many Americans signed that page to buy a small business. Car washes. Plumbing companies. Dental practices. Landscaping outfits with six trucks and a yard.

Most of those deals ran through an SBA 7(a) loan. The government guarantees 75% to 85% of the balance for the bank. The buyer guarantees all of it.

For four of those years the down payment was optional. A 2021 rule change let a buyer take over a whole company without putting in cash. That window closed on June 1, 2025, when the SBA put the 10% equity injection back and pushed seller notes onto full standby.

The guarantee was written to protect the lender. The borrower signs the other side of it.

So the zero-down loans are now four and five years old. That is the age at which small-business loans tend to break.

The trailing twelve-month default rate across the whole 7(a) book reached 4.8% in March 2026. That is the highest reading since 2013, and roughly three times the 1.6% floor of 2021.

It got there without a recession. Unemployment never spiked. Bank earnings look fine.

The program is supposed to cost taxpayers nothing. Congress wrote that into the statute. Lender fees are meant to cover the failures.

In fiscal 2024 they didn't. The SBA reported negative cash flow of about $397 million in the 7(a) program, the first shortfall in over a decade, and said $460 million in upfront lender fees had gone uncollected from 2022 through 2024.

So the SBA repriced. It brought the fees back, rewrote the rulebook, and raised the amount of the buyer's own money that has to sit underneath the guarantee.

The lending stopped almost immediately. Through the first nine months of this fiscal year, approvals ran 33.4% below the same stretch a year earlier by count, and 20.9% lower in dollars.

Interesting. An insurer that pays out more than it takes in raises the premium and covers less. That is all this is.

But the loans already written don't get repriced. They sit there, four years old, at a variable rate, with a name on the guarantee.

I have no idea how far this runs. The vintage that worries me hasn't finished aging yet.

When a lender decides a loan program has turned against it, the lender does not announce anything. It just stops showing up.

In July 2026, 1,141 lenders were still participating in the SBA's 7(a) program. A year earlier there were 1,407. That is a drop of 18.9% in twelve months, and the lowest lender count in thirty years.

More than half of the ones left made five loans or fewer.

The retreat is not spread evenly. Approvals of $500,000 or less fell about 38% by count, against about 15% for larger loans. The band between $350,000 and $500,000 fell 64%.

That band moved out of streamlined processing. Since 2014, a passing credit score let a lender skip the full cash-flow analysis on smaller loans. The SBA ended that shortcut on March 1, 2026.

Every loan now needs a documented commercial credit analysis and a debt service coverage ratio of at least 1.1 to 1, at any size. Twelve years of underwriting shortcuts came off the books in one notice.

And it spreads from there. Losses in a guarantee program are never evenly distributed. They concentrate wherever the underwriting was loosest.

Community Advantage was the small experimental wing of the 7(a) program. It ran through non-depository lenders rather than banks, and it was built to reach borrowers the mainstream program missed.

It defaulted at 7% over a twelve-month stretch. That is more than double the rate of the overall 7(a) portfolio.

The distribution underneath was worse. Several individual lenders in the program posted early problem loan rates above 30%.

On May 19, 2025, the SBA stopped issuing new licences for it and raised capital reserve requirements on the lenders already in.

One in three loans going bad inside the first stretch of a loan's life is not an economic cycle. That is an underwriting result, and it took a regulator to find it.

Meanwhile, follow the risk to where it settles once the paperwork is done.

A bank writes a government-backed small-business loan and splits it in two. The guaranteed piece, usually 75% to 85%, gets sold to investors. The SBA estimates lenders sell that piece on nearly half of all the loans they write.

Once it sells, the government's guarantee to the buyer becomes unconditional. Full faith and credit. The buyer gets paid whether the plumbing company survives or not.

The bank keeps the other 15% to 25% and collects a servicing fee for the life of the loan.

Anyone owning 20% or more of the business signs a personal guarantee for the whole balance.

Three parties are in every one of these loans. Only one of them is uninsured.

And the loan is variable. It floats off the prime rate, which sat at 6.75% in late July, with the spread capped at three points above that on loans over $350,000.

So the buyer who signed in 2022 is paying near 9.75% today on a ten-year note for a business they bought with none of their own money. That is the structure. It works until the cash flow stops.

Three more things worth keeping track of this week.

Japan is running the same experiment on a longer timeline. Teikoku Databank reported on August 10 that 1,037 companies failed in July, up 8.5% from a year earlier. That is two straight months above a thousand, which has not happened since the end of 2009. Sixty-three of July's failures followed what the Japanese call zero-zero loans, the interest-free, unsecured emergency credit handed out during the pandemic. Failures tied to inflation hit a record 121.

France has its own version, and the labour side of it is worse than the headline. The Banque de France counted 68,961 restructurings and liquidations in the twelve months to January, up 4.1% on the year. Altares put insolvencies at 71,100 over the twelve months to March, with 19,000 new proceedings in the first quarter alone. The figure that stopped me: more than 75,000 French jobs were at risk in that single quarter, above the 73,000 peak of early 2009.

Meanwhile, Americans keep starting businesses anyway. The Census Bureau counted 531,423 business applications in June, seasonally adjusted, up 1.1% on May and well above the pre-pandemic run rate. Its own model expects 29,741 of those applications to turn into an employer within four quarters. New formation is booming while the last wave of small-business borrowing turns over. We'll see.

In 2020 the government lent money straight to small businesses, with no bank in the middle. About 3.9 million loans went out before the program closed in May 2022.

Thirty-year terms. They were loans, not grants, and none of them were forgivable.

By June 30, 2025, the SBA had written off $75.2 billion of them.

The inspector general looked closely at one slice. As of December 2024 the agency had charged off 369,588 of these loans, worth over $47 billion, none of them flagged for suspected or confirmed fraud. It recovered less than 1% of the original amounts.

Charge-off sounds like the end of the story. It only means the agency stopped trying.

The file then goes to the Treasury. On April 24, 2026, the SBA sent 562,000 pandemic-era loans worth $22.2 billion over for collection, and called it the largest referral package on record.

Treasury does not have to sue anyone. It can order an employer to withhold 15% of a borrower's take-home pay. It can take a tax refund before the cheque is written. It can reduce a Social Security payment.

So how long does that authority last? There is no statute of limitations on it.

A man named James Lockhart tested that once. He borrowed for school between 1984 and 1989 and never repaid. In 2002 the government began withholding part of his Social Security cheque. He sued, arguing a ten-year limit had long since run out.

The Supreme Court ruled against him in December 2005. The decision was unanimous.

More than a decade after the debt went bad, the government still had a claim on his retirement income. That is settled law now.

So the debt outlives the business, the deferment, and most of the arguments against it. A loan signed at a kitchen table in 2020 can still be taken out of a Social Security payment in 2050.

Nobody wrote a headline when the referral went out in April. There was no market reaction. Five hundred and sixty-two thousand files changed hands and the number moved from one column to another.

The paperwork is patient.

We'll see.