Long-Term Capital Management had run the numbers…
Two Nobel Prize winners sat on its board… and Business Week had gushed that never had this much academic talent been given this much money to bet with.
And the firm’s models said it was unlikely to lose more than $35 million on any single day.
Yet on Friday, August 21, 1998, it lost $553 million… in a single day.
And within 2 years, the fund that was being called the greatest collection of talent in Wall Street history… was forced to shut down.
The problem? Their models.
Because, like all models - they were built on years of past prices… with the working belief was that tomorrow’s risks can be worked out from yesterday’s.
This is also known as a backest, and LTCM’s models were a very sophisticated one.
But then came leverage…
At the start of 1998 the fund held about $129 billion of assets against $4.7 billion of its own money. That’s a leverage ratio of over 27:1.
Which is why a small miss in the model became a huge loss on the books. Within five weeks the New York Fed scrambled 14 banks to bail out LTCM.
So while backtesting is nothing new… and can be useful… today we face another problem…
The accesibility for retail investors…
In May Robinhood began letting customers connect their own AI agents to a trading account. A couple months prior, Public announced agents that can place recurring trades.
One example it gave was a covered-call strategy aimed at $5,000 a month… with shiny backtests to prove how effective it is…
In 2014, a study published in the American Mathematical Society showed that tweaking a strategy a fairly small number of times can produce a spectacular backtest… that falls apart on new data. So the more variations you try, the more likely you have found noise rather than any usable signal.
And now we’ve got AI agents involved… that can try variations far faster than you can... that leaves us in a sticky situation…
The kind of situation which results in you losing money!
That is why I like strategies I can explain in a couple sentences, rather than ones that have to go through hundreds of rounds of backtesting to produce even a passable results.
The wheel… LEAPS… seasonal trades - all fit the bill
Which is why I’ll be doing my own trading rather than handing it off to an agent anytime soon…
Oliver