Being smart in trading isn't about having a perfect model.
It's about building models, getting them wrong, and updating them faster than everyone else.
Every up-down Polymarket trader looking to make a buck has figured out the Black-Scholes model. If you plug the bitcoin price and the recent realised volatility in, you figure out a theoretical price and you get an idea of if the price is high or low. When the price looks high, you lift the bid, when it looks low you sell the offer.
Then you start to figure out that flow is really important. The guy trying to coin flip a month's wages is going to move the price a hell of a lot more than your vol prediction. You can make more money dealing with that flow than you can praying it goes up when you have 2% of edge.
The smart trader adds in that order flow to their theoretical price, uses the book volumes as a signal to trade. They figure out what is driving the good trades and drop their hubris to do more of them.
The best trader has the best learning loop.
A lot of it comes down to finding links.
When the first strikes in Iran happened, everyone knew what it meant. Oil up, inflation fears, risk off.
Natural gas gets constrained — the rookie trader looks for reversion while the guys who traded through Ukraine start looking for what's next.
How do we fuel the world's crops? How do we meet the AI boom's helium demand? This is where you can actually make money. Which stocks are exposed to fertiliser upside? Are helium prospectors neglected?
But most of the real learning comes when things stop behaving the way you expect.
2nd August 2024. Nikkei options.
We had good models. We understood how the surfaces should look, how things lined up with the US and the rest of Asia.
Stocks were down. Yen was strengthening. Nikkei looked stressed.
It had become common knowledge that a strong yen meant a weak Nikkei. Japanese earnings decreased and their valuations fluttered. It was not so common knowledge how fragile the link between yen borrowers, US stocks and Japanese stock owners had become.
For years, ridiculously low interest rates had driven careless borrowing to carry out the carry trade. Borrow yen at 0%, buy dollars with the yen, use those to chase juicy returns from American stocks.
The world's boldest economic experiment had created a ticking time bomb. Was it time for the dues to be paid?
Vol was high. Skew was stretched. Everything looked elevated relative to what we trusted. US markets, Asia, history — it all said the same thing.
It looked too expensive. The market calmed throughout the European session.
We got to work as dutiful market makers and sold some of it as it went up.
US job numbers came out. Bad.
That was the trigger.
What looked like a normal move turned into forced selling. The kind where nobody cares about fair value anymore.
We woke up to the Nikkei down 12% and our trades from the day before down 10 million.
We got in for another day of trading, updated our models, prepared for the next time the JGB cracks started to show.
This was a smart model broken by a market that had learned to watch and wait for the fragility.
This is the part that translates cleanly to AI.
- They can try more strategies.
- Test more ideas.
- Catch more flaws.
- Update constantly.
But there are parts they don't get for free.
- Knowing when the model is missing something.
- Understanding how positioning warps prices.
- Recognising when you're in a different regime entirely.
They're very good inside a system.
They're not naturally good at spotting when the system has changed.
We need to constrain these models, test them against the world, and rapidly review their assessments.