Every stock picker is selling the same thing: a thesis. A narrative about why a stock should go up or down. It’s informed guesswork dressed up as analysis.
TradeSmith’s An-E does something different: it forecasts.
The difference matters. A guess has no measurable accuracy. A forecast comes with a confidence score, a historical track record, and a price target to the penny. You can test it. You can grade it. And when you’re wrong, you know exactly how wrong and by how much.
This is the shift that An-E represents — a fundamentally different way of thinking about what a stock prediction even is, rather than a better stock picker.
The Problem With Predictions
The financial advice industry runs on opinions. Analysts publish price targets, but most are vague ranges or outdated blasts from six months ago. The academic literature has known for years that single-model stock prediction is fragile. The Journal of Big Data published a comprehensive evaluation in 2020 showing that ensemble methods — models that combine multiple algorithms — significantly outperform single classifiers across every major exchange. The question has never been whether AI can forecast stocks. It’s whether a retail investor can access AI that’s actually built for the job.
An-E is built for the job. A team of 69 engineers, 22 ML specialists, and 23 quants spent $18 million and 50,000 man-hours training it on 1.3 quadrillion data points. The engine covers 2,300 stocks, funds, and ETFs. It generates a price forecast 21 trading days out, to the penny, with a per-forecast confidence score and a historical accuracy rating for each stock.
Those are the specs.
How An-E Actually Works
The engine is an ensemble of two AI models. One tracks longer-term trend patterns — the direction a stock wants to go over weeks. The other handles day-to-day volatility and short-term noise. Together, they produce a forecast that’s more stable than either model alone. This is the same approach the academic literature calls “stacking” — combining multiple models so the weaknesses of one are covered by the strengths of another. Studies from leading journals have shown stacking ensembles can achieve 90-100% directional accuracy on certain market data. An-E’s numbers are more modest and more honest: 57.53% directional accuracy, 60.30% target accuracy across all rated stocks. Past performance does not guarantee future results; the accuracy figures describe the engine’s historical hit rate, not a forecast of what Predictive Alpha will deliver going forward.
The output is a Prime Projection Date — the single day within the 21-day window where the model has the highest confidence in its forecast. You get a price target, an expected move percentage, a confidence score, and a historical accuracy gauge for that specific stock, with per-stock historical accuracy ratings ranging from 80%+ down to 40% so the system tells you which is which.
The weather radar analogy applies here. An-E doesn’t control the storm. It tells you where the storm is likely to go. A 60% accurate forecast means you plan for the likely path, carry an umbrella, and know it could still shift. The difference between a 60% forecast and a coin flip is that the forecast tells you which side of the coin is heavier.
What Predictive Alpha Gives You
Predictive Alpha is the product wrapper around An-E. The base tier gives you one weekly stock search through the engine — punch in any ticker from the 2,300+ universe, and An-E returns a complete forecast. You also get two analyst-vetted picks per month, weekly video market analysis, and access to the Options360 tool that maps An-E’s forecasts to options strategies.
The Prime tier opens the full system. Unlimited searches, real-time screening filters, and the ability to sort stocks by expected move, confidence score, and projection date. The list price is $5,000 per year, with the current offer at $1,799 — still a serious investment, but a fraction of what a single institutional-grade quant terminal costs.
I checked the published demo results. OXY was forecast to go from $46.21 to $49.23 over 21 trading days. It hit $49.19. That’s a 0.09% delta on a 6.44% move. Discover Financial was forecast bearish at -9.97%. It hit -9.93%. Duolingo was forecast at +10.18% with 66% confidence. It hit +10.89%. These are not cherry-picked home runs. They’re the published verifications, and they’re precise enough to make you stop and think. Past performance does not guarantee future results; the published demo results are historical forecast-versus-actual pairs, not a forecast of what the engine will produce going forward.
The bearish calls are the real test. Any model can find stocks going up. A model that can confidently forecast a -9.97% drawdown on a financial stock and land within four basis points — that’s a different caliber of signal.
Where This Sits in the Landscape
Hedge funds have been running ensemble forecasting models for a decade. The infrastructure is expensive, the talent is scarce, and the data pipelines are proprietary. The An-E engine is the first retail-facing product to package ensemble forecasting inside a subscription wrapper, built on a $18 million engineering investment with 69 engineers, 22 ML specialists, and 23 quants. The infrastructure is the kind that hedge funds have run internally for a decade; An-E’s contribution is making the approach available at a retail price point rather than an institutional terminal cost.
The academic research supports the approach. A 2024 study on Chronos models found that transformer-based stock prediction generated gross Sharpe ratios of 3.17 — but the edge disappeared after 3 basis points of transaction costs. That’s the difference between a backtest and reality. An-E promises a better map, with the execution still belonging to the user.
TradeSmith’s parent company, MarketWise, is publicly traded on Nasdaq. That means there’s actual accountability. The claims are subject to SEC scrutiny. The product is attached to a real business with 381,000 paid subscribers and $81 million in quarterly billings. That’s not nothing.
The demo is free and requires no credit card, which puts the forecast accuracy claim inside reach of direct verification. The published demo results — OXY, Discover Financial, Duolingo — are the verifications TradeSmith has put forward. The full set of rated stocks carries per-stock historical accuracy ratings, which is the mechanism a skeptical investor uses to separate the 80%-accuracy names from the 40%-accuracy names.
Predictive Alpha doesn’t claim to be a crystal ball. It claims to be a 60% accurate forecast engine that tells you which stocks it’s confident about and which ones it isn’t. In a market where most predictions are dressed-up opinions, the engine’s stated-uncertainty framing is the differentiator from the category it sits in.
For the broader context on why this matters, see Predictive Alpha: AI That Predicts Stock Prices. For the full index, see Promo Watch.