A number, not a paragraph

When Keith Kaplan demonstrates Predictive Alpha in the TradeSmith presentation, he does something unusual. Instead of showing you a chart or a page of analysis, he shows you a number. One number per stock.

For Clorox in July 2020, the number was 5.19%. Predictive Alpha projected the stock would go up 5.19% over the next 21 trading days. It went up 6%. For Block in August 2020, the projection was 5.61%. The stock went up 26%.

This is the core of AI stock prediction as TradeSmith presents it. Not a narrative, not a thesis, not a story about why a company is undervalued. A number, generated by a machine, with a time frame attached.

As we detail in our full review of the PRA Top 5 presentation, this approach has real appeal for investors who are tired of trying to parse contradictory analyst opinions. But it also raises real questions about what AI can and cannot do in financial markets.

The Time Series difference

Kaplan makes a point of distinguishing Predictive Alpha from ChatGPT and other large language models. ChatGPT is trained on words and images. Predictive Alpha is built on what Kaplan calls a “Time Series” model, trained specifically on numbers.

This distinction matters. Language models are designed to predict the next word in a sentence. Time Series models are designed to predict the next number in a sequence. Stock prices, trading volumes, and market data are sequences of numbers, and the theory is that a model built for numbers will outperform a model built for language on numerical prediction tasks.

The London Stock Exchange Group shares 7.3 million price updates per second, as Kaplan notes in the presentation. That is 220 trillion price updates per year. No human can process that volume of data, and no team of human analysts can either. AI’s advantage is not intelligence in the human sense. It is the ability to process volumes of data that are physically impossible for a person to review.

The Devon Energy example

One of the most compelling examples in the presentation is Devon Energy. On February 1, 2022, Predictive Alpha flagged Devon Energy with a projection of 6% upside for the month. The stock went up 19%, an annualized gain of 252%.

What makes this example interesting is what happened that same month. Russia invaded Ukraine. Oil and gas prices exploded. Energy stocks soared. But Predictive Alpha, as Kaplan points out, “sure as hell didn’t know that Russia’s army was planning to invade Ukraine.” The AI does not read newspapers or geopolitical analysis. It reads data.

Something in the data, patterns that were invisible to human eyes, told the model that Devon Energy was positioned to move. The AI could not explain why, because it does not know why. It just saw the pattern.

This is the promise and the limitation of AI stock prediction in a single example. The AI saw something real. But it cannot tell you why it saw it, and it cannot guarantee the pattern will repeat.

The backtest problem

The presentation cites a 71% win rate and a 1,110% total return over a five-year backtest period. As we note in our review, backtested results are notoriously sensitive to methodology. Look-ahead bias, survivorship bias, and overfitting can all inflate simulated returns well beyond what a live trader would experience.

The backtest includes the AI boom and two major market crashes, which is a genuinely challenging period. But the gap between a backtest and live results is where most algorithmic products fall short, and the presentation does not disclose the specific methodology, position-sizing rules, or turnover assumptions that produced the 1,110% figure.

The honest read is that the backtest is suggestive, not definitive. It tells you the model found something in the historical data. It does not tell you whether that something will persist in live trading.

The SaaSpocalypse proof

Kaplan offers a more recent and more convincing piece of evidence than the backtest. In early 2026, a new AI release from Anthropic disrupted the software industry. Salesforce, Adobe, and Intuit all dropped sharply. IBM went into free fall.

Right as SaaS stocks crashed, Predictive Alpha issued a buy call on Western Digital, a data-storage firm in a completely different part of the market. The stock went up 28% that month, and the system issued further buy calls through November, December, January, February, March, and April. The stock went up every single month.

Meanwhile, Lumentum Holdings got a buy call on November 1. The stock went up 50% in a month, and further calls in December and January also produced gains.

The contrast is instructive. While SaaS stocks collapsed, the AI pointed investors toward hardware and AI infrastructure names that were benefiting from the same trend. The AI did not know that Anthropic’s release would disrupt SaaS. It saw the data pointing in a different direction and followed it.

The honest assessment

AI stock prediction is not magic. Kaplan is honest about this in the presentation: “I don’t want to pretend that AI is some sort of magic bullet that gets it right every time. That’s obviously impossible.”

But the underlying logic is sound. Markets produce more data than humans can process. AI can process that data and find patterns that are invisible to the human eye. The Time Series approach, trained on numbers rather than language, is a reasonable architecture for financial prediction.

The questions are about execution, not theory. Can the model adapt to changing market conditions? Does the 71% win rate hold in live trading? And does the simplicity of “buy five stocks on the 1st” capture the full value of the system, or does it undersell a product that also includes more active trading features?

For investors curious about AI stock prediction, Predictive Alpha is a genuine product with genuine technology behind it. The backtest is promising but unverified, and the $499 annual price with a 60-day refund policy makes it a low-risk way to test the thesis with real money.

NewsletterVetter is an independent publication. We receive compensation from some of the services we review through affiliate links. Nothing on this site is investment advice. Always do your own research.