Inside Our AI Stock Predictor: What the Factors Actually Mean
Our predictor scores stocks across twenty-plus factors and draws a forecast cone. Here is what those signals measure, and the limits we want every user to understand.
Search any ticker in our Stocks tab and alongside the live chart you get an AI verdict: a buy/hold/avoid lean, a confidence score, a one-month target range, and a set of named signals. This article explains what is under the hood at a conceptual level, because a tool you cannot reason about is a tool you should not trust.
The factor families
The engine blends signals from a few well-studied families. None of them are our invention; they are the same ingredients quantitative funds have used for decades, applied at educational scale:
- Trend and momentum. Is the price above or below its own moving averages? Stocks in established trends statistically continue them slightly more often than they reverse. The engine reads several lookback windows, from weeks to a year.
- Mean reversion and stretch. The opposite force. When a price races far above its recent average (measured with tools like RSI and Bollinger bands), short-term snapback odds rise. Trend and stretch signals often disagree; the blend is the point.
- Volatility regime. How wide are the daily swings, and are they widening? The same bullish setup means something different in a calm tape versus a panicked one, so volatility scales the confidence and the width of the forecast cone.
- Fundamentals. Valuation ratios, earnings growth, and analyst consensus (ratings and targets) anchor the technical picture to the underlying business.
- Sentiment and context. Recent headlines for the symbol and the market-wide regime (risk-on versus risk-off) tilt the score at the margin.
How the pieces become a verdict
Each factor produces a score. The engine weighs them, and the weights themselves are adjusted over time by comparing past predictions with what actually happened (that is the self-learning part). The sum becomes the gauge position you see; the volatility estimate becomes the cone around the price path; the biggest contributing factors are listed as named signals so you can see why, not just what.
The limits, stated plainly
- It cannot see the future. No model can. Factor edges are statistical tilts, a few percentage points of advantage that only show up across many predictions, never certainty about one.
- It knows nothing outside its data. A surprise acquisition, a CEO scandal tonight, a factory fire: invisible until they hit prices and headlines.
- Confidence is not accuracy. The confidence score measures how strongly the factors agree, not how often the engine is right. Agreeing factors can all be wrong together, and in crashes they often are.
- It is a teaching instrument. We built it so learners can watch named, explainable signals interact with real markets. It is not, and will never be presented as, a reason to place a trade.
How to use it well
Treat the verdict as a structured second opinion. Form your own view first from the chart and fundamentals, then open the signal board and see what the engine weighted. When you disagree, articulate why; that argument is where the learning happens. Pair it with a paper portfolio and you can test the engine, and yourself, without risking a dollar.
Key takeaways
- The predictor blends momentum, mean-reversion, volatility, fundamental, and sentiment factors into one explainable score.
- Weights adapt over time by scoring past predictions against outcomes.
- Confidence measures factor agreement, not correctness; edges are statistical, small, and unreliable case by case.
- Use it as a second opinion and a teaching tool alongside paper trading, never as a trade instruction.
Educational content, not financial advice.