How Probabli.AI creates probabilistic stock forecasts, defines bullish, neutral, bearish, P10, P50, and P90, and handles model uncertainty.
Probabli.AI presents ranges and scenarios instead of claiming a single certain outcome. Analyses are generated from structured market, company, analyst, earnings, and headline data supplied to a supported large language model.
How probabilistic forecasts are generated
The model returns structured scenario probabilities, price ranges, reasoning, risks, catalysts, and confidence. The system validates the response format and normalizes bullish, neutral, and bearish probabilities to total 100%.
Forecast terminology
- Bullish: an upside scenario; neutral: a base or middle scenario; bearish: a downside scenario. Their percentages are model-estimated weights, not promises or audited odds.
- P10: a pessimistic but plausible price outcome; P50: the median or base outcome; P90: an optimistic but plausible outcome.
- P5 and P95 may describe more extreme tails. These values are AI-generated estimates, not statistically calibrated confidence intervals.
Horizons and update frequency
Analyses may cover 1-year, 5-year, and 10-year horizons. The model is instructed to reflect greater uncertainty over longer periods, but the system does not independently calibrate or guarantee that relationship. A saved forecast changes when a new analysis is run; every populated stock forecast identifies its generation date.
Data categories and sources
Forecast inputs can include current price and change, market capitalization, 52-week range, selected valuation ratios, earnings estimates and surprise measures, analyst consensus and targets, dividend and debt measures, and recent headline metadata and sentiment. These inputs are primarily obtained through Yahoo Finance and public financial-data interfaces. SEC EDGAR powers separate filing and ownership features but is not represented as a direct input to every forecast. Availability varies.
Models and disagreement
Supported models can come from OpenAI, Google, Anthropic, and xAI. A saved single-model analysis reflects the selected model. In the multi-model view, individual outputs remain available side by side so disagreement can be inspected. Displayed consensus values are arithmetic averages of available outputs and are not adjusted for inter-model dispersion.
Limitations and disclosure
Models can misread inputs, omit facts, reason inconsistently, or produce inaccurate estimates. Data may be delayed, incomplete, or revised, and forecasts cannot reliably anticipate unexpected company or market events. Results are for informational and educational use only, not financial advice or guaranteed returns.
Review ownership
Methodology and disclosures are maintained by Probabli.AI creator Matt Farley, an engineer and investor since 2012 who does not hold formal financial credentials. Read his background on the About page and see forecast validation standards.
The information on this page is for educational and informational purposes only and does not constitute professional financial advice.