How Our AI Works
Complete transparency into our AI models, methods, and data sources powering institutional-grade earnings intelligence
AI Models & Technology Stack
Primary Models:
- GPT-4 for earnings call analysis and sentiment extraction
- Claude 3.5 Sonnet for complex financial reasoning and trade opinions
- FinBERT for financial sentiment classification
Our NLP models analyze earnings transcripts, financial reports, and market news to extract actionable insights and sentiment indicators.
Machine Learning Models:
- XGBoost for earnings surprise prediction
- LSTM Neural Networks for price movement forecasting
- Random Forest for volatility estimation
Trained on 15+ years of historical earnings data, our models predict earnings outcomes and post-earnings price movements with quantified confidence scores.
Data Sources & Quality
Market Data
Real-time and historical data from Polygon.io, Alpha Vantage, and NYSE/NASDAQ feeds with sub-second latency
Earnings Reports
SEC filings, company press releases, earnings transcripts from EDGAR and Benzinga with automated parsing
Analyst Consensus
Aggregated analyst estimates from FactSet, Bloomberg, and Thomson Reuters with real-time updates
AI Analysis Pipeline
Data Ingestion & Cleaning
Raw data from multiple sources is normalized, validated, and enriched with metadata. Missing values are imputed using statistical methods and historical patterns.
Feature Engineering
We extract 200+ features including: historical earnings patterns, analyst estimate trends, options implied volatility, sector performance, macroeconomic indicators, and social media sentiment.
Model Ensemble & Prediction
Multiple specialized models analyze different aspects (fundamentals, technicals, sentiment) and their outputs are combined using weighted voting based on historical accuracy.
Confidence Scoring & Validation
Each prediction includes a confidence score (0-100%) based on model agreement, data quality, and historical accuracy in similar scenarios. Low-confidence predictions are flagged.
Model Performance & Accuracy
Earnings Beat/Miss Accuracy
Prediction accuracy for whether a company will beat or miss earnings estimates
Price Direction Accuracy
Accuracy in predicting post-earnings price movement direction (up/down)
High-Conviction Trades
Success rate on trades with 80%+ confidence scores
Performance Disclaimer: These metrics represent historical backtesting results and are updated quarterly. Past performance does not guarantee future results. All AI predictions should be used as one input in your trading decisions, not as the sole basis for trading.
Limitations & Ethical AI
•Models perform worse during extreme market volatility and black swan events
•Limited historical data for newly public companies (IPOs < 2 years)
•Cannot predict unexpected news events or management changes
•Reduced accuracy for companies with irregular earnings patterns
✓No data manipulation or cherry-picking of results
✓Transparent disclosure of model accuracy and confidence levels
✓Regular third-party audits of model performance claims
✓User data privacy and no selling of personal information