The loop where the collateral and the lending standard move together

Start with a single loan. A bank lends against an asset, and how much it will lend depends on what the asset is worth. Nothing controversial yet. The trouble is that what the asset is worth depends on what the next buyer can borrow to pay for it.

That is the entire mechanism, and it is a loop rather than a chain. Easier credit raises what buyers can pay, which raises prices, which raises appraised collateral values, which makes the same loan-to-value ratio look more conservative than it did last year, which supports easier credit. The lending standard and the collateral value are not two independent facts a lender weighs against each other. They are one variable, observed twice, and each observation is used to justify the other.

Run it backwards and the symmetry is what makes it dangerous. Prices fall, collateral is worth less, the same loan now looks aggressive, so the lender tightens, which reduces what the next buyer can pay, which lowers prices further. Nobody in the chain has behaved irrationally. Each participant has responded sensibly to a price the previous participant's credit created.

This is why credit cycles overshoot at both ends, and why they are not the same thing as the business cycle even though they overlap it. A recession is a fall in output, dated after the fact by a committee. A credit contraction is a fall in the willingness to lend, and it can begin before output turns and persist long after output recovers.

The 1990 to 1991 episode is the cleanest demonstration. The recession lasted eight months, from July 1990 to March 1991. Commercial and industrial lending peaked in December 1990 at 641.6 billion dollars and did not stop falling until December 1993, at 583.7 billion, roughly thirty three months after the recession officially ended. Output had been recovering for nearly three years while credit was still shrinking.

So the transferable claim is about timing rather than direction. Credit tightens quickly, because tightening is a decision a committee can make in an afternoon, and it loosens slowly, because loosening requires a lender to believe the collateral again. If you are exposed to credit, the recovery arrives later for you than the headlines say it has arrived for everyone else.

Four things, and the reason to keep them separate is that they arrive in a fixed order. Confuse a leading indicator with a lagging one and you will act on the credit cycle exactly one turn too late.

First, the survey. Every quarter the Federal Reserve asks senior loan officers whether they have tightened or eased standards, and reports the net percentage saying tightened. This is unusual and valuable because it measures willingness directly rather than inferring it from a price. In the April 2026 survey the net share tightening standards on commercial and industrial loans to large and middle-market firms was 8.1 percent, which is a mild net tightening. For scale, that reading was 83.6 percent in the fourth quarter of 2008 and 50.8 percent in the third quarter of 2023. Standards on GSE-eligible mortgages were at minus 3.7, meaning a net share of banks were easing.

Second, the flow that actually happened. The H.8 release gives bank credit and its components weekly. Total bank credit was 19,741 billion dollars in mid July 2026, growing about 6.4 percent year over year, and commercial and industrial loans were 2,894 billion in June, up about 8.0 percent. A survey can say banks are cautious while the loan book still grows; when the survey and the flow disagree, the flow is the one that is spending money.

Third, the price of risk. The spread between what a middling corporate borrower pays and what the government pays is the market's live vote on how likely repayment is. A note on sourcing here, because it matters for reproducibility: the widely quoted ICE high-yield spread is now published on public data services only for a rolling three year window, so historical comparisons cannot be made from it without a license. The Federal Reserve's own Baa corporate yield less the ten year Treasury has the full history and is used throughout this brief instead. It sat at 1.58 percentage points on 23 July 2026, against 6.16 at its December 2008 peak, 4.31 in March 2020, and 2.28 during the 2023 bank episode.

Fourth, the damage, which arrives last and is therefore useless as a warning. Delinquency on all loans at commercial banks was 1.48 percent in the first quarter of 2026, against a peak of 7.40 percent in the first quarter of 2010. Note the date on that peak: nine months after the recession had officially ended. Credit card delinquency peaked in the second quarter of 2009 at 6.77 percent and stood at 2.92 percent in early 2026.

Behind all four sits the stock rather than the flow: how much leverage exists to be squeezed. Nonfinancial corporate debt was 14.45 trillion dollars in the first quarter of 2026, about 45 percent of GDP, against roughly 47 percent at the end of 2008. Treat that figure carefully, because different institutions define corporate credit differently and the definitions are not interchangeable. The Bank for International Settlements measure of total credit to United States nonfinancial corporations puts it at 72.2 percent of GDP for the fourth quarter of 2025, and the gap between 45 and 72 is definitional, not a disagreement about the world.

The move to carry: read these four in order and ask which one has turned. If the survey has tightened and spreads have widened while loan growth and delinquencies still look fine, you are early in a contraction, not safe from one.

Three, chosen because they fail in three different places. Read them as variations on one mechanism rather than three unrelated crises.

1990 to 1991, the lender is the problem. Commercial real estate lending had been aggressive through the 1980s, savings institutions were failing in volume, and the constraint on the economy became the balance sheet of the lender rather than the appetite of the borrower. 653 institutions failed across 1990 and 1991. The recession itself was mild and short, eight months, but as the previous section showed, lending fell for three years past its end. When the lender is the impaired party, credit recovery waits on recapitalisation, and recapitalisation is slow.

2007 to 2009, the collateral is the problem. This is the loop of the first section running in reverse at full speed on the largest asset class in the country. The net share of banks tightening commercial and industrial standards hit 83.6 percent in the fourth quarter of 2008, the highest reading in the survey's history. The Baa spread over Treasuries reached 6.16 points on 4 December 2008. Bank credit itself shrank, bottoming at minus 5.45 percent year over year in late 2009, and commercial and industrial loans fell 25.3 percent from peak to trough, hitting minus 20.2 percent year over year in February 2010. Delinquencies peaked in the first quarter of 2010, which is to say they peaked after the recession, after the spread peak, and after the survey peak. Every one of those lags is the mechanism, not noise.

March 2023, the funding is the problem, and the panic is not the peak. Silicon Valley Bank was closed on 10 March 2023 with about 209 billion dollars of assets, Signature Bank on 12 March with 110.4 billion, and First Republic on 1 May with about 229 billion. First Republic is the instructive one: it reported 103.9 billion dollars of deposits in mid April against 176.4 billion at the prior call report, and that gap is the run, visible in the accounts.

Then the part that breaks most people's intuition. The Federal Reserve's Bank Term Funding Program, created that month to stop the run, did not peak during the panic. Its balance peaked at 167.8 billion dollars on 24 January 2024, ten months later, because banks kept drawing on a facility that had become cheap relative to alternatives. The emergency response and the emergency are on different clocks.

And the credit consequence showed up in the survey rather than in failures. The July 2023 survey reported net tightening on commercial and industrial loans of 50.8 percent, the highest reading since the fourth quarter of 2009 outside the 2020 spike. Only five institutions failed in 2023, but the surviving banks tightened as though many more had, which is how a contained banking episode still transmits into a credit contraction.

The move to carry across all three: identify which link in the loop is impaired, because that determines what has to be repaired before credit can return. An impaired lender needs capital. Impaired collateral needs prices to stop falling. Impaired funding needs a facility, and it needs one for longer than the headlines suggest.

Four exposures, in rough order of how early they feel it, and the point of ordering them is that the same contraction arrives at different businesses months apart.

The lenders first, obviously, and not only through defaults. A bank whose funding cost rises faster than its asset yields is squeezed before a single borrower misses a payment, which is what happened in March 2023: the failures were about the value of safe assets and the speed of deposits, not about credit losses.

Then anyone whose customer borrows to buy the product. A homebuilder does not sell houses, it sells monthly payments, so a mortgage rate move reprices its entire market without touching its costs. The same is true of anything financed at the point of sale. Cancellation rates are the tell, because a cancellation is a customer who wanted the product and could no longer fund it.

Then anyone who finances their own inventory or their own input. This is where the credit cycle stops being macroeconomics and becomes a line in someone's accounts, and it is worth reading the NVR dossier beside this one for the specific case: a builder that controls its lots through forfeitable deposits rather than owning land is structurally less exposed to a land price collapse, and in a soft market it still took about 76 million dollars of deposit impairments in a single year while its lot costs rose. Optioning an input moves the price risk to a counterparty who charges for holding it, and in a credit contraction that counterparty's own funding gets harder, so the premium rises exactly when the option is most needed.

Then the leveraged borrower, who feels it last and worst. Nothing happens to a company with a five year loan when standards tighten. Something happens when the loan matures, and the something is that the price of rolling it is set by a lender who has just tightened. This is why maturity walls matter more than leverage ratios: the ratio tells you the exposure, the maturity schedule tells you the date.

What to do with all of this is narrower than it sounds. You cannot forecast the turn. What you can do is know, for any business you care about, which of these four exposures it has, and which indicator therefore governs it. A lender is governed by funding costs and the survey. A payment-selling business is governed by rates and cancellations. An input-optioning business is governed by the counterparty's cost of capital. A leveraged borrower is governed by a calendar. Match the business to its indicator and you will be reading the right number when it moves, which is most of what reading a cycle honestly consists of.