Somewhere this week, a trader looked at a calendar reminder that has probably sat there for years: October is anniversary week for market disasters, so be careful. The reminder was on time. The market did not care. South Korea's KOSPI fell almost 8 percent in a single session as chipmakers took a beating, the Nikkei 225 dropped nearly 5 percent, and a tech-led rout ran through markets worldwide — with hedge funds, the professionals with the risk models, among the biggest casualties of the AI selloff, according to Reuters.
Then the experts arrived with the comparison everyone had been waiting for. The S&P 500's Shiller CAPE ratio is back above 40 — its highest level since the dot-com bubble, when it peaked near 44. More stocks are hitting 52-week lows than highs, even on days the index rises — something that last happened just before March 2000. Therefore, the essays conclude, history is repeating.
The numbers are real. The conclusion is a prediction. And the difference between those two things is the entire subject of this piece.
It is tempting to call the selloff inexplicable because no single event triggered it. That is a misreading of how markets actually move. The ingredients were stacked in plain sight all month: the IMF's Global Financial Stability Report warning that AI-linked equity and credit corrections could spread well beyond tech; seven companies now above 35 percent of the S&P 500's total value; 1.4 trillion dollars of future revenue riding on hyperscaler-AI deals the IMF itself notes may not be able to pay; rate fears; and a semiconductor chief executive telling a conference that investors buying the AI frenzy are "in a casino" with "no clue what they are investing in."
Nothing "happened" on Thursday. A story that had been quietly repricing all week simply finished repricing. That is not an anomaly — it is the standard shape of a drawdown, and it is why waiting for a clean catalyst before managing risk is a strategy that only ever works in hindsight.
The dot-com echoes are genuinely striking: valuation ratio near the 2000 peak, breadth deteriorating beneath a rising index, an AI narrative that "will change the world" the way the internet one did. The internet did change the world — after its bubble took the Nasdaq down 78 percent, and the index needed fifteen years to reclaim its peak.
But notice what the comparison actually is. A CAPE of 41 versus 44 tells you the market is expensive. It does not tell you the week, the month, or the year. The CAPE ratio has been above 30 for years while indexes made record after record — everyone who sold at "historically expensive" and waited for the repeat sat out the entire run. Every "this is 2000 again" essay, however well-researched, is a forecast dressed as history. Forecasts about markets do not survive contact with the week that follows, and neither will this one, whichever way it resolves.
There is one thing the comparison is good for, and it is not timing. It is the account math. The Nasdaq's 78 percent decline and fifteen-year recovery is not a story about the index — it is a story about what an uncontained drawdown does to every account that was fully exposed when it began.
Which leads to the observation that makes the whole comparison useful: the trader who predicted 2000 and the trader who did not needed exactly the same preparation. Sizing that assumed a multi-year bear was possible. Losses taken small, because the recovery arithmetic punishes depth without mercy. Nothing about their week-of-the-crash behaviour differed based on whether they had called the top. If the prediction changes nothing about the preparation, the prediction was never the edge. The preparation was.
The calendar reminder deserves its own honesty, because it is a real risk practice — and a flawed one. "Be careful this week" assumes risk lives on dates. It does not. It lives in position size. The trader who is careful on anniversary week carries full risk the other fifty-one, which is precisely how the professionals got caught this week: their models were careful every day, and the damage still arrived, because careful-with-size is a floor, not a forecast.
A pre-committed rule does not know what week it is. Stops are placed before the entry, size is set before the session, and the daily maximum exists on quiet Tuesdays exactly as it does on crash anniversaries. That is not a limitation of the approach — it is the approach. The week that hurts is never the one on the calendar.
Same arithmetic governs us, so plainly: our automation carries the rule set into the market — stops and limits placed before any position exists. A monitor that watches the engine's behaviour for drift from its verified baseline and cuts its authority when drift appears. Parameters localised per market and broker, because a stop distance correct for one instrument's volatility is a larger loss on another. In build, not yet shipped: an analysis layer that reads a trader's own closed-trade history and reports what it shows — including how their account behaved in weeks exactly like this one.
None of it predicts. None of it needed a catalyst. All of it held before this week, and will hold after it, whichever way the 2000 comparison resolves.
History does not repeat. It audits. The market changes. The rules don't.
