Health of Market from Complexity Medicine Point of View

Fuel, Wind, and a Match

What a century of crashes says about the market right now

Yoon Hang "John" Kim, MD, MPH  ·  September 2026

I spend most of my working life making sense of complex patterns in bodies. Lately I have been looking for them in markets, and I keep landing on the same uncomfortable one. So I did what I would do with a patient I could not quite read: I went back through the chart history and looked for every prior case that resembled the one in front of me.

This is a pattern study, not a forecast and not advice. I am a physician, not a financial advisor. But the pattern is clean enough that I wanted to write it down, partly to think it through and partly so I can look back later and see whether history rhymed.

Three things have to show up together

Every really big market crash of the last hundred years had the same three ingredients at the same time. Think of them as fuel, wind, and a match.

The fuel: prices are very expensive. Imagine a lemonade stand that earns one dollar a year, and someone pays forty dollars for it. That buyer needs everything to go perfectly for decades just to break even. The most common way to measure this for the whole market is the Shiller CAPE ratio, which compares today's prices to the average of the last ten years of earnings. Its long-run average is about 17.

The wind: people bought with borrowed money. If you buy a hundred-dollar stock with fifty of your own dollars and fifty borrowed, and the stock drops to sixty, your lender gets nervous and makes you sell. Your selling pushes the price down, which makes the next borrower sell too. That chain reaction is what turns a decline into a crash.

The match: interest rates go up. When borrowing gets more expensive, the people who borrowed to buy stocks feel it first. Rates rarely cause the crash by themselves. They light what is already stacked.

When only one of the three is present, you get a bear market that hurts for a while and then recovers. When all three are present, you get the declines people still talk about fifty years later.

Figure 1. Peak-to-trough decline in each episode, with the number of ingredients present. Blue bars had one or two; red bars had all three.

The cases

United States, 1929

The textbook version. The CAPE ratio hit about 32. Buying stocks with only ten percent down was common, and brokers' loans reached roughly eight to nine percent of the whole economy. The Fed had been tightening, with the New York discount rate at six percent by August 1929. The Dow lost 89 percent from its peak to the 1932 bottom. About half of that came from the crash itself; the other half came from bank runs and policy mistakes afterward.

Japan, 1989

The closest overseas twin. The Nikkei traded above sixty times earnings, and land in Tokyo was priced as if it were worth more than all of California. Stocks and property were used as collateral to buy more stocks and property. The Bank of Japan raised rates from 2.5 percent to 6 percent in about a year, on purpose, to deflate the bubble. The Nikkei lost 63 percent by 1992 and did not stop falling for good until 2003, at 82 percent below the peak. It did not get back to its 1989 high until 2024.

United States, 2000

The CAPE peaked near 44, the highest reading ever recorded. Margin debt spiked. The Fed raised its rate to 6.5 percent in May 2000. The S&P 500 fell 49 percent and the Nasdaq fell 78 percent. No banking collapse followed, because the borrowed money was sitting in brokerage accounts and startups rather than inside the banking system.

United States, 2007

Here the borrowing was the star and the valuation was only moderately high, with a CAPE around 27. Household debt was near one hundred percent of the economy, banks were running twenty to thirty times leverage, and a shadow banking system depended on short-term funding. The Fed had raised rates from one percent to 5.25 percent, which reset adjustable mortgages and started the defaults. The S&P 500 fell 57 percent, and because the leverage lived inside banks, it became a systemic crisis.

The contrast that proves the point

Paul Volcker raised rates to nearly twenty percent in the early 1980s, higher than any of the cases above. But stocks were cheap, with a CAPE under ten, and hardly anyone was buying them with borrowed money. The result was a 27 percent bear market and then one of the great bull runs in history. Fuel and wind matter more than the size of the match.

The 2022 bear market sits in between. Prices were expensive, with a CAPE in the high thirties, and the Fed hiked fast. But borrowing was moderate outside of a few pockets. Two ingredients, not three. The market fell 25 percent and recovered within two years.

Putting numbers on it

Here is every episode that fits the description, plus two partial fits from 1937 and 1973 where the tightening was strong but the borrowing or the valuation was less extreme.

Episode	Index	Peak → trough	Loss	Duration	Back to old high
1929 Great Depression	Dow	Sep 1929 → Jul 1932	−89%	34 months	1954 (25 yrs)
Japan 1989	Nikkei	Dec 1989 → Aug 1992	−63% (−82% by 2003)	32 months	2024 (34 yrs)
2000 dot-com	S&P 500	Mar 2000 → Oct 2002	−49%	31 months	2007 (7 yrs)
2000 dot-com	Nasdaq	Mar 2000 → Oct 2002	−78%	31 months	2015 (15 yrs)
2007–09 financial crisis	S&P 500	Oct 2007 → Mar 2009	−57%	17 months	2013 (5.5 yrs)
1973–74 (partial fit)	S&P 500	Jan 1973 → Oct 1974	−48%	21 months	1980 (7 yrs)
1937 (partial fit)	Dow	Mar 1937 → Mar 1938	−50%	12 months	1945 (8 yrs)

Averaging all seven rows:

• Loss: 62 percent on average, with a median of 57 percent.

• Time to the bottom: about 25 months on average, with a median of 31.

• Time back to the old high: 14.5 years on average, dragged upward by 1929 and Japan. The median is 8 years.

Two things stand out. Every full-fit case lost at least 49 percent on the broad index, and the bubble asset itself, whether it was the Dow, the Nikkei, the Nasdaq, or American banks, lost 63 to 89 percent. Nothing in this group was a twenty-percent pullback. And the decline was always a process rather than an event: the first twenty percent is the fast part, and the rest grinds on for a year or two.

Where we are today

So how many of the three ingredients are present in September 2026?

The fuel is fully stacked. The Shiller CAPE is around 42. Only the 2000 peak was higher, and only barely. The story behind the price is the same one every generation tells: in 1929 it was a new era of permanent prosperity, in 2000 it was the internet, and today it is artificial intelligence. Each of those stories turned out to be true about the technology and wrong about the price.

Figure 2. Shiller CAPE ratio at each market peak. Today's reading is second only to the dot-com top.

The wind is blowing. Investors have borrowed a record 1.5 trillion dollars against their brokerage accounts, up 49 percent in one year. Growth that fast has only appeared in the run-ups to the 2000, 2007, and 2021 peaks. Relative to the size of the economy, margin debt is at roughly 4 percent versus a fifty-year median of 1.5. There is a newer form of wind as well: the largest technology companies are borrowing hundreds of billions to build AI data centers, with roughly a third of that spending now financed by debt, and a growing share of it moving into private credit and off-balance-sheet structures where losses are harder to see. Their balance sheets are still strong, and nothing today resembles the bank leverage of 2008. But the borrowing is real and it is growing.

Someone is reaching for the match. The Federal Reserve is at 3.5 to 3.75 percent and has not started raising yet. But its July vote was 9 to 3, with the dissenters wanting a hike, and the Fed chair said the pause was the beginning of a story rather than the end of one. As I write this, markets give better-than-even odds of a rate increase on September 16. Inflation is running above target, partly because of energy prices tied to the conflict in the Middle East.

Figure 3. Two ingredients fully present, one arriving.

My score is two and a half out of three. That is more than 2022 had and about what 2000 had at the same point in the cycle, with the difference that today's rates start at 3.6 percent rather than zero, so there is less room before the borrowed money starts to hurt.

What the pattern does and does not tell you

If the third ingredient arrives and the Fed keeps going, history says the broad market falls somewhere between 35 and 50 percent and the most crowded AI names fall 60 to 80 percent, with the bottom arriving about two years after the top. If the borrowing turns out to be more systemic than it looks, especially through private credit, the range shifts toward 55 to 65 percent and the recovery stretches to five years or more. If the Fed backs off after a hike or two and inflation cools, this becomes another 2022: a 20 to 30 percent bear that people forget in a few years.

What the pattern cannot tell you is when. In every prior case the market peaked near the end of the rate-hiking cycle, not the beginning. A hike in September would open the window, not close it. And in every case something showed up that the prior case lacked: bank runs in 1929, property in Japan, mortgages in 2007. The novelty this time might be private credit, or it might be something none of us are looking at.

A sample of five or six episodes over a century is a pattern, not a law. It tells you the shape of what tends to happen and how big it tends to be. It does not tell you the date, and it does not tell you what to do. Anyone who claims it does is selling something.

Luck favors the prepared bold. The preparation, for me, is simply knowing which chapter of the history book we are in. I think we are in the one where the fuel is stacked and the wind is up, and everyone is watching the Fed's hand.

Sources and notes

Valuation data from Robert Shiller's published CAPE series (multpl.com, shillerdata.com). Margin debt from FINRA monthly statistics. Federal Reserve statements and minutes from the July 28–29, 2026 FOMC meeting. AI capital-spending and debt-issuance figures from Goldman Sachs, J.P. Morgan, and PIMCO research published in summer 2026. Historical drawdowns are peak-to-trough on nominal closing prices. This post is for education and reflection only and is not investment, legal, or medical advice.