Listen to the audio version of this article (generated by AI).
Managing Editor’s note: 2,927 TradeSmith folks joined CEO Keith Kaplan for yesterday’s AI trading event where he showcased the newest version of the Predictive Alpha AI trading model.
It forecasts prices for thousands of stocks up to 21 trading days out. It’s not looking for stocks that are about to shoot the lights out. It’s looking for high-probability setups with more modest gains.
And it learns as it goes. It tracks how each forecast turns out so it can improve its accuracy over time. As the model has evolved, the results have improved.
It recommends trades only when they have historical accuracy rates of 85% or more. That means the model’s forecasts on this stock have been right in the past roughly 8 times out of 10.
If you missed the event, catch up here. Keith included a free stock recommendation for anyone who attended. And it’s one of the most bullish trades across the more than 2,000 stocks it tracks.
This AI Is Giving Investors an Edge
BY KEITH KAPLAN, CEO, TRADESMITH
In early July 2024, Hurricane Beryl was tearing across the Caribbean with winds topping 165 miles per hour. By the time it made landfall, it had killed 36 people, left millions without power, and caused billions of dollars in damage.
For decades, forecasters have used giant supercomputers to track storms like this. They take readings from satellites, ships, and planes. Then they grind through complex physics equations about how air and water move. It can take more than an hour to produce a single 10-day forecast.
When Beryl formed, forecasts from a top European weather agency pointed to Mexico as the likely target for landfall.
But an experimental forecasting system called GraphCast, which can run on something as small as a laptop, disagreed. Days in advance, it predicted the storm would make landfall in Texas.
When Beryl struck Matagorda Bay, Texas, on July 8, it was GraphCast – not the world’s most advanced supercomputer – that had been right all along.
GraphCast was created by Google’s DeepMind AI lab. It’s trained on 40 years of weather data. And it can produce a 10-day forecast in 60 seconds.
Instead of solving complex equations, it spots hidden patterns in past weather data. This allows it to see further and faster than traditional physics-based models.
This was a turning point for meteorology. A laptop-scale AI beat the most powerful forecasting engines on Earth.
And if AI can decode the weather, what might it do with the other great chaotic system of modern life: the stock market?
My team of 74 researchers and developers at TradeSmith has been putting our $8 million annual budget to work to find out.
And after years of development and testing, we’ve created a new “Super AI.” Instead of learning from weather patterns, it learns from stock market data. And the results have blown us away.
It can project future prices of 2,334 stocks up to 21 trading days out – to the day and even the penny – with 85% backtested accuracy.
Today, I’ll show you how it works – and how you can use it to dramatically up your odds of success as an investor. First, it’s important to understand the common thread between predicting storm paths and stock market prices.
A Butterfly Flaps Its Wings…
In 1961, MIT meteorologist Edward Lorenz was tinkering with a rudimentary weather model on a Royal McBee computer.
It was the size of a fridge, spat out forecasts on long rolls of paper, and could take an hour just to process a handful of equations.
To save time, Lorenz tried a shortcut. He rounded one of his inputs from six decimal points to three. He let the computer run while he stepped out for coffee. When he came back, the forecast had transformed. A calm weather pattern had become a raging storm.
What Lorenz discovered is that tiny changes can snowball into huge effects. He called it the “butterfly effect” because a butterfly flapping its wings in Brazil might set off a tornado in Texas.
Weather systems can shift on a dime because they’re dynamic, not linear. Even the tiniest change, like a gust of wind or a seemingly minor increase in humidity, can cascade into a wildly different outcome.
Markets are the same.
Every trading day, markets absorb thousands of tiny shocks. A Fed comment, a surprise earnings miss, even a social-media post can shift sentiment.
These are financial “butterfly effects.” Tiny shifts that ripple through the system and hit the prices of thousands of stocks.
That’s why the world’s best hedge funds have spent decades building algorithms to capture them. Take Jim Simons’ Medallion Fund. It’s averaged 66% annual returns since 1988 by spotting financial butterfly effects hidden in vast stock market datasets.
At their core, forecasting storms and forecasting stock prices face the same problem. Both are dynamic systems where tiny changes create outsized consequences.
That’s why at TradeSmith, we built our own Super AI to read the market’s turbulence the way GraphCast reads the weather.
I don’t mind pulling back the curtain on it today. It’s nearly impossible to replicate. It takes decades of market data, thousands of hour of programming, and a dedicated research team to stitch it all together.
Which is why funds like Medallion keep their edge locked away.
From Storm Paths to Stock Prices
It’s a kind of AI model called a TimeGPT.
It isn’t designed to write text or generate images. Instead, it forecasts what’s known as time-series data.
Think of stock prices like storm tracks: data points lined up in time, where each moment connects to the ones before, and hidden patterns influence what comes next.
GraphCast works the same way. It doesn’t solve physics equations from scratch. It learns from decades of weather data to see how storms grow and move.
And just as weather forecasters don’t want to miss the next hurricane, investors can’t afford to miss the next “Hurricane Nvidia,” “Hurricane Apple,” or “Hurricane Tesla.”
Our AI is built to see those storms before they hit.
You can see what I mean from the results from our backtests…
On March 9, 2026, our model flagged BHP Group (BHP) with a projected move of 2.1% over the next month. We got in at $71.77 per share, with a target to hit our projected gain by March 26.
Just one day later, the stock surged to $74.26. And my team closed the position for a 6.2% gain in just two days.
That may not sound extraordinary at first. But when you annualize that return, it works out to an astonishing 1,135% annualized gain.
And traders using a special type of options trade could potentially have boosted that kind of move even further.
That’s the kind of short-term move our Super AI is built to catch.
And as impressive as that single trade is, we found you could have done even better with a portfolio approach.
There’s a lot of complex math going on under the hood with our five-stock strategy, but you won’t notice it’s there. You simply buy the best five trades – all with an unusually high 85% historical accuracy – and sell when they hit their prime projection date or price, whichever comes first.
And in backtests, this five-stock strategy could have made you an average gain of 239% over the last seven years. And that’s just the average gain.
That’s more than three times the return of AI rocket ride Nvidia (NVDA). And it’s more than 30 times the return of the S&P 500.
That seven-year study included the pandemic, the 2022 crash, soaring interest rates, the tariff tantrum, and even two wars – yet the five-stock strategy still delivered that kind of average annual gain.
It’s simple to follow. With just a couple of minutes’ attention every week, it crushes the returns most investors are making.
So if you haven’t yet used AI to help you invest, you should check out the replay for the event right here…
Sincerely,
Keith Kaplan
CEO, TradeSmith