About NeuralStock
An educational dashboard for US stocks and ETFs
What we do
NeuralStock is a free, educational stock market dashboard. We combine four algorithmic prediction models with an AI research companion so that anyone — beginner or experienced — can understand what the numbers behind a stock actually mean.
We cover more than 250 popular US stocks and ETFs. Every stock page shows a 90-day price chart, four independently computed model forecasts, their historical accuracy, and a chat assistant that answers your questions in plain English.
We are not a brokerage and we do not give financial advice. NeuralStock is a learning tool — it helps you understand market concepts and model-based thinking, not tell you what to buy or sell.
How the prediction models work
Trend follower
Compares a 10-day moving average against a 50-day moving average. When the short-term average rises above the long-term average the model predicts 'up'; when it falls below, 'down'. Moving averages smooth out daily noise so you can see the underlying direction.
Momentum
Uses the Relative Strength Index (RSI) — a 0–100 gauge of how strongly a stock has moved recently — combined with the 5-day rate of change. Extreme readings often signal a reversal is coming. The model weighs these two signals together into a direction and confidence score.
Statistical trend
Fits a straight trend line through the last 60 days of prices using linear regression, then measures how far today's price has strayed from that line (z-score). A large positive or negative deviation suggests the price may revert toward its trend.
Multi-factor + AI news sentiment
Our most complex model. It blends four weighted signals: trend (25%), momentum (25%), unusual volume (15%), and AI-analysed news sentiment (35%). An AI reads recent news articles about the stock, scores each one bullish or bearish, and that score feeds into the final prediction alongside the technical signals.
All predictions are computed nightly after US market close and stored in our database — pages load instantly and never run models in real time. Historical accuracy figures are always out-of-sample: a prediction is only scored using price data from after it was made.
The AI research companion
Every stock page has a chat panel powered by Anthropic's Claude AI. It is pre-loaded with the stock's current price, predictions, and key facts so it can answer specific questions — not generic ones.
Ask it things like: "What does a high RSI mean for this stock?", "How should I interpret the trend model's confidence score?", or "What sectors compete with this company?" It will give you a grounded, educational answer and always remind you that it is not a financial advisor.
Our honesty commitment
Prediction models can look impressive on paper and fail in the real world — often because they were overfit to historical data. We guard against this by:
- Keeping models intentionally simple and rule-based.
- Scoring accuracy only on out-of-sample predictions.
- Always showing the sample size alongside any accuracy figure ("57% over 200 predictions").
- Always showing a baseline rate so you can judge whether a model is actually useful or just slightly better than a coin flip.
We believe the most valuable thing this site can teach is how to think about model predictions sceptically — not just what the predictions say.