Putting AI to the Test
September 15, 2023
AI has dominated the headlines this year.
And ChatGPT has taken the world by storm… amazing us with its ability to pass bar exams… conduct research… and diagnose patients.
But at Manward, we’re most interested in whether it can actually help everyday investors boost their returns.
That’s why Alpesh Patel just went live with his new AI-enhanced trading strategy in GVI Investor. He’s combining elite, hedge fund-level research with the data-crunching power of ChatGPT to target winning stocks. (See what he’s doing right here.)
You don’t need to be a hedge fund manager to use ChatGPT in your investment research (though it helps to have someone like Alpesh in your corner!).
That’s exactly what I set out to prove for you this week. I wanted to see if AI really could help everyday traders create a winning stock-picking strategy for the long haul.
So I asked ChatGPT…
It’s a simple prompt. But don’t let it fool you.
First off… getting a 25% annual return isn’t child’s play. It takes some serious trading chops to achieve that consistently.
If an AI-generated strategy could pull this off… it would rival the greatest investors of all time, like Peter Lynch and Warren Buffett.
No easy task.
And second… I wanted to make sure the strategy ChatGPT created properly balanced risk and reward.
That’s where the Sharpe ratio comes in. It’s a respected measure of how well a strategy can produce returns without exposing you to unnecessarily high risks.
A Sharpe ratio of 1.0 or higher means the strategy is very good at balancing risk and reward. And a ratio below 1.0 suggests the strategy’s a bit more volatile than we’d like.
In short, my simple prompt was a tall order to fill.
But was it too tall for ChatGPT?
Let’s move on to the results…
The response I got from ChatGPT included a fairly long set of guidelines. But it all basically boiled down to identifying stocks with the following traits:
- A market cap in the microcap to midcap range
- Five years of consistent earnings growth
- A lower than average P/E
- A history of paying dividends
- A 50-day moving average above the 200-day moving average
- An RSI of 70 or lower
- Six to 12 months of market outperformance.
It sure sounds like a good set of criteria. But would we be able to test the strategy to make sure?
You bet.
To do that, I went to our trusty Bloomberg Terminal and built a stock-screening model that mirrored ChatGPT’s suggestions.
Since ChatGPT noted the historical outperformance of smaller companies over their larger peers, I chose to use a market cap range of $6 million to $175 million. This allowed our model to mainly target small cap stocks while also including the largest half of microcaps and the smallest half of midcaps.
It took just a few minutes to set up. I ran a 20-year backtest, with quarterly rebalancing, to see how it performed against the S&P 500.
Then I left my desk to make a fresh pot of coffee. I returned some 10 minutes later with a fresh cup of joe in hand to see the results.
Over the past 20 years, the ChatGPT-suggested strategy would have delivered a very strong performance, returning more than 2,000% and beating the S&P 500 by more than 250%.
But as impressive as that is… it would have delivered only a 17% annualized return over that period. Not quite the 25% per year we were hoping for.
It did meet one of our criteria, though…
The Sharpe ratio for ChatGPT’s strategy was 1.09. That’s a big deal.
This means the strategy not only would have produced market-beating returns… but would have done so without exposing the overall portfolio to a ton of risk.
Not bad, ChatGPT. Not bad at all.
And that was with just a small bit of fiddling. Once you see what Alpesh has been up to with ChatGPT, I think you’ll be blown away.
There’s a reason Business Insider, the Financial Times and Yahoo Finance have all asked him to explain how to use ChatGPT to help target winning stocks.
And now… he’s gone live with his AI-enhanced trading strategy. Thanks to him… everyday investors can have the power of AI – backed up by elite, hedge fund-level research – at their fingertips.