The Day the AI Trade Broke
Case study · one simple rule, eighteen possible worlds
The idea in one paragraph
Nobody knows what the market will do next. But from today’s situation, only a handful of futures are realistic. A world model generates those futures, plays your trading rule in each one, and shows you where the rule makes money, where it loses, and why. Once you can see where a rule breaks, you can fix it before it costs you.
A market is a state.
The real chip-stock index from April to today: a peak in June, then the AI selloff. Everything that follows starts here.
Compress the past.
The world model condenses the last sixty days into a compact state: momentum, volatility, drawdown and macro context, thirty-two numbers in all.
Add an action.
Ask a question with consequences: what if I buy $100,000 of chip stocks today, and sell if I am down 15 %?
Imagine what follows.
From that state and that action, the model rolls the market forward: eighteen plausible worlds, three hundred futures each.
Audit every world.
Keep scrolling to step through all eighteen. Each world shows how likely it is, its chain of causes, and which part of the state it depends on. Nothing is a black box.
Open any future.
In the data-centre pause world, a policy shock arrives overnight and jumps straight past the stop-loss. The reason is visible, step by step.
Reverse-engineer the model.
Interpretability works on the model too. Probe what each part of the state encodes, find where the dynamics are blind, such as overnight jumps, and change the architecture. Better explanations make better models.
Fix what fails.
Collect every outcome. Where the rule breaks, change it: insure against the whole market and sell sooner. Same worlds, smaller losses. Below, you can do it yourself.
1. What happened today
Two days after the US midterm elections, the stocks that had led the market for three years, the companies building artificial intelligence, fell hard.
- The S&P 500 (the 500 largest US companies) fell 4.4 %, its worst day since April 2025.
- Chip makers, the companies that make the processors AI runs on, fell the most: AMD −17.3 %, Nvidia −9.3 %. The Philadelphia Semiconductor Index, which tracks the largest US-listed chip companies, closed at 9,011, down 12.2 % on the day and 38 % below its June peak.
- Why: building AI data centres takes huge loans, and borrowing is expensive. Investors worried that fewer data centres means fewer chips sold. The election results also raised the chance that governments pause new data centres, because they push up electricity bills.
- In the background: the interest rate on 10-year US government debt has come down to 4.84 %, from above 5.30 % last month. Oil (Brent) is at $98 a barrel during the Iran war. The Bank of England raised interest rates today.
Sources: Associated Press market report of 5 November 2026, via NBC Philadelphia; index closes from Yahoo Finance (^SOX, ^GSPC), cross-checked against FRED for the S&P 500.
2. Your rule
Your rule is called buy the dip: when chip stocks drop sharply in one day, buy a fund that tracks the chip index and hold it for four weeks, betting the fall was an overreaction. Today qualifies. Set the rule up the way you would:
I put into chip stocks, sell automatically if I am down , and insure against the whole market falling.
“Sell automatically” is a stop-loss. “Insure” means also betting against the overall market (the Nasdaq index), so a market-wide fall is partly offset.
3. Eighteen worlds that could follow
The black line is what really happened to chip stocks since April. From today’s close, the world model keeps only the futures that today’s facts make realistic, and weighs each by how likely it is. Hover over a line to read that world; click it for the full review.
4. Fix the rule
Now use the failures. The model tries a hundred versions of your rule (how much to put in, when to sell, how much to insure) and keeps the one that earns the most while keeping your worst 1-in-10 outcome under a $15,000 loss.
The point is not to guess which world will come. It is to hold a rule you understand in all of them.
How to read this page: the facts in section 1 are reported news. The worlds, their chances and every dollar figure come from a simple illustrative model built by hand for this page. They are not forecasts and not investment advice. In Almagest’s system the worlds are generated by a world model from the market’s state.