Stock Analysis Methodology
How TechTars generates the daily 'why this stock moved' analysis: the data inputs, the technical signals we score, how the AI explanation is produced and reviewed, and the limits of the model.
Formula
SignalTone = sign(SentimentScore + TechnicalScore + NewsImpactScore) · ConfidenceWeight
Variables
| Symbol | Name | Description |
|---|---|---|
| SentimentScore | News sentiment | Aggregate sentiment of headlines from the last 24 hours, weighted by source authority and recency. |
| TechnicalScore | Technical reading | Composite of RSI position, MACD position vs its signal line, distance from 20/50-day MAs, and volume vs the 30-day average. |
| NewsImpactScore | News impact | Materiality of headlines (earnings, guidance, M&A, regulatory) weighted higher than routine coverage. |
| ConfidenceWeight | Confidence weight | Multiplier from 0 to 1 based on how aligned the three scores are with each other and with the day's actual price move. |
Worked example
If RSI is 72 (above 70, technical −1), the MACD line has just crossed below its signal line (technical −1), the price closed below the 20-day MA on 1.8x volume (technical −2), but the day's news flow is a strong earnings beat (sentiment +2, news impact +2), the composite is +0. The page shows 'Mixed signals' with Medium confidence. The label describes the conflict in today's data, and 'key levels to watch' lists the nearest support and resistance. It is not a forecast.
Assumptions and limitations
Every model leaves something out. Here is what this calculator assumes, and what it does not model, so you can interpret the output honestly:
- All inputs (price, volume, technical indicators, news headlines) come from publicly available sources via our data providers. We do not use insider information or private datasets.
- AI-generated prose passes a contract validator before display (must include the ticker, percent change, support/resistance, and a one-sentence reason). Stories that fail the validator are not shown, and users see a 'temporarily unavailable' state.
- The model does not predict future prices. It explains today's move and surfaces the data points to watch. It is not a buy/sell recommendation.
- Single-day analysis only. Multi-day trends, fundamental valuation (P/E, P/B, cash flow), and corporate-action timelines (dividends, splits, AGMs) are not modelled in the daily story.
- Sentiment scoring works best for high-coverage tickers. Small-cap stocks with thin news flow can produce lower-confidence readings, and these are flagged in the confidence breakdown.
Authoritative sources
Where the formula, rates, or framework come from: