Backtesting Methodology
How we collect event data, calculate real outcomes, and validate predictions. Outcome calculations are based on verified stock price movements from public market data.
Real Data, Not Mock Data
Impact Radar uses real historical event data and verified stock price movements. We do not use simulated or fabricated data. Events are sourced primarily from SEC EDGAR filings and FDA announcements, with additional event types being integrated as the platform expands.
All price data comes from public market data APIs, and all return calculations are computed using the exact formulas shown on this page. This ensures complete transparency and reproducibility.
Coverage Status
Primary Coverage: SEC 8-K filings, FDA announcements, earnings, and more are continuously monitored with automated ingestion and scoring. Approximately 1,490+ events across 440+ tickers have been validated with verified price outcomes.
Expanding Coverage: Additional event types (earnings, M&A, guidance updates, dividends, analyst ratings, corporate announcements) are being integrated incrementally. Coverage breadth is prototype-level and expanding as the platform matures.
Important: Impact Radar is a research and analysis tool. All predictions, scores, and analytics are designed for educational and analytical purposes, not investment advice or guarantees. Always conduct your own due diligence and consult with qualified financial professionals before making investment decisions.
Data Collection
11 automated scanners monitor SEC filings (8-K, 10-Q, EDGAR), FDA announcements, earnings calls, M&A activity, guidance updates, dividends/buybacks, product launches, press releases, and Form 4 insider trading. Events include source attribution and timestamps.
Impact Scoring
Each event receives a deterministic impact score (0-100) using quantitative rules. This base score considers event type, historical patterns, sector context, and market conditions.
Price Tracking
We track actual stock price movements at 1-day (64.6% accuracy), 5-day (51.7% accuracy), and 20-day (51.7% accuracy) horizons. Benchmark returns (SPY) are calculated to isolate event-specific impact from general market movements.
Outcome Validation
After sufficient time has passed, we calculate the actual returns and compare them to our predictions. This data trains the Market Echo Engine to improve future predictions.
Data Sources
SEC EDGAR Filings
Automated scanning of SEC 8-K current reports capturing material events like M&A, executive changes, material agreements, and regulatory updates. Additional SEC filing types are being integrated.
FDA Announcements
We monitor FDA drug approvals, rejections, clinical trial results, and regulatory actions. These events have high impact for biotech and pharmaceutical companies.
Additional Event Sources
Targeted coverage of corporate announcements including product launches, partnership agreements, and strategic initiatives. Coverage is expanding as the platform matures.
Quantitative Formulas
Every EventOutcome record is calculated using these deterministic formulas. There is no AI or estimation involved in these calculations—they are pure mathematics applied to real price data.
Raw Return
Simple percentage change in stock price over the specified horizon (1d, 5d, or 20d).
Benchmark Return
S&P 500 (SPY) return over the same period, representing general market movement.
Abnormal Return
The event-specific impact after removing market noise. This is our primary target metric for ML training.
Direction Accuracy
Binary check: did we correctly predict if the stock would move up or down?
Market Echo Engine Integration
The quantitative formulas above generate EventOutcome data, which serves as the ground truth for training the Market Echo Engine. Here's how the two systems work together:
- EventOutcome data serves as ground truth labels for training ML models
- Market Echo Engine learns which event types, sectors, and market conditions lead to stronger/weaker moves
- When Market Echo Engine makes a prediction, it stores ml_adjusted_score on the event record
- Backtesting engine uses ML predictions (ml_adjusted_score) when available, otherwise uses deterministic scoring (impact_score)
- Accuracy dashboard tracks performance across 3 model versions: v1.0-deterministic, v1.5-ml-hybrid, v2.0-market-echo
- Continuous learning pipeline: new outcomes → model retraining → improved predictions
Key Insight: EventOutcome data is created with quantitative formulas (deterministic, reproducible, transparent). The Market Echo Engine uses this data as training labels to learn better predictions. When ML predictions are available, the system uses ml_adjusted_score; otherwise it falls back to the deterministic impact_score. Formulas provide truth, AI learns to improve predictions.
Strategy Framework
Build, test, and refine custom trading strategies based on event signals. The backtesting framework supports comprehensive strategy definition with entry/exit rules, position sizing, and risk management.
Strategy Builder
Define custom entry conditions based on event types, impact score thresholds, direction filters, and ticker selection. Build strategies that match your trading style.
Risk Management
Set stop-loss and take-profit levels, choose position sizing methods (fixed, percent of equity, volatility-scaled), and define maximum positions per strategy.
Performance Metrics
Comprehensive metrics including Sharpe ratio, Sortino ratio, CAGR, max drawdown, win rate, expectancy, and detailed trade-by-trade history.
Position Sizing Methods
Choose how your strategy allocates capital to each trade. Different methods suit different risk tolerances and market conditions.
Fixed Size
Use a constant dollar amount per trade regardless of equity or volatility
Percent of Equity
Risk a percentage of current portfolio value on each trade
Volatility-Scaled
Adjust position size based on the asset's volatility (ATR-based)
Performance Metrics
Comprehensive metrics to evaluate your strategy's performance. All calculations follow industry-standard methodologies.
- Total Return: Overall profit/loss as percentage of starting capital
- CAGR: Compound Annual Growth Rate normalized for time period
- Sharpe Ratio: Risk-adjusted return measuring excess return per unit of volatility
- Sortino Ratio: Like Sharpe but only penalizes downside volatility
- Max Drawdown: Largest peak-to-trough decline during the backtest period
- Win Rate: Percentage of trades that were profitable
- Expectancy: Average expected gain per trade accounting for win rate and average win/loss
- Trade Count: Total number of trades executed in the backtest period
Equity Curve: Track your strategy's performance over time with interactive equity curve charts. See exactly how your portfolio value changes with each trade.
Technical Highlights
SEC 8-K/10-Q/10-K, FDA, Earnings, M&A, Guidance, Dividends, Product Launches, Insider Trading, and more
1-day (64.6%), 5-day (51.7%), and 20-day (51.7%) price movement tracking when historical data is available
SEC EDGAR, SEC 8-K, SEC 10-Q, FDA, Earnings, M&A, Guidance, Product Launch, Dividends, Press, Form 4 Insider
Events with verified price movements across all scanner types, growing daily
Complete Example Workflow
Step 1: Event Detection
SEC scanner detects an 8-K filing for ACME Corp (ACME) announcing a major partnership agreement on November 15, 2025 at 2:30 PM ET. Deterministic scoring engine assigns impact_score=72 with direction="bullish".
Step 2: ML Enhancement (Optional)
Market Echo Engine reviews the event and stores ml_adjusted_score=78 based on historical partnership announcements in this sector. The system uses this ML prediction (78) when displaying the event to users with Pro or Enterprise plans.
Step 3: Price Tracking
System records: price_before=$50.00 (Nov 15), price_after_1d=$53.50 (Nov 16), price_after_5d=$54.75 (Nov 22). SPY moved from $450 to $452.70 (+0.6%) over the same 5-day period.
Step 4: Outcome Calculation
OutcomeLabeler runs on November 23, 2025 and calculates:
Step 5: Learning & Improvement
This EventOutcome record (+8.9% abnormal return) is added to the training dataset. Market Echo Engine learns that partnership announcements in this sector tend to outperform the base model's predictions. Next time a similar event occurs, the AI will make a more calibrated prediction.
Result: Validated Prediction
The prediction was directionally correct (bullish) and the magnitude was strong (+8.9% abnormal return vs. 78/100 impact score). This outcome validates the model and helps improve future predictions for similar events.
Trade Signal Generation
AI-Generated Trade Recommendations
Building on our backtesting methodology, the Trade Signal system transforms validated event predictions into actionable trade recommendations with precise entry/exit targets.
Entry Price
Current market price at signal generation
Stop Loss
5% below entry or 2x ATR for volatility-adjusted protection
Take Profit
Calculated from stop distance x impact multiplier x 1.5
Position Size
1-5% of portfolio based on confidence level
Risk/Reward Calculation
Signals with R/R ratios above 2:1 are highlighted in green, 1.5-2:1 in yellow, below 1.5:1 in red.
Model Explainability (SHAP)
SHAP-Based Feature Contributions
Every prediction from the Market Echo Engine now includes SHAP (SHapley Additive exPlanations) values that show exactly which features contributed to the prediction and by how much.
What is SHAP?
SHAP is a game-theoretic approach to explain the output of any machine learning model. It connects optimal credit allocation with local explanations using Shapley values from game theory.
How We Use It
For each event prediction, we calculate SHAP values for all input features (event type, sector, market conditions, historical patterns, etc.). The visualization shows which features pushed the prediction higher or lower.
Multi-Horizon Support
SHAP explanations are available for all three prediction horizons: 1-day, 7-day, and 30-day. Different features may dominate at different time scales.
Example Feature Contributions
Sector Analysis & Rotation Signals
Sector-Level Performance Tracking
Our backtesting framework now includes sector-level analysis to identify rotation signals and momentum trends across market sectors.
- Rotation Signals: Detect capital inflow/outflow between sectors based on event activity and price movements
- Momentum Scoring: Track sector momentum using event-driven price changes over multiple time horizons
- Event Clustering: Identify when multiple high-impact events concentrate in specific sectors
- Cross-Sector Correlation: Discover relationships between sector movements and event types
Integration: Sector analysis feeds into the Trade Signal generation, allowing for sector-aware position sizing and risk management across your portfolio.
Data Quality & Transparency
Impact Radar is built on verifiable, traceable data. Events from automated scanners include links to original source documents where available. Price calculations use publicly available market data. All formulas are documented and reproducible.
Our Data Quality Dashboard (available in the app) provides real-time freshness indicators, pipeline health monitoring, and data lineage tracking. You can see exactly when data was last updated, which scanner collected it, and how outcomes were calculated.
We believe in radical transparency. If you can't verify it, you shouldn't trust it. That's why events include the base deterministic score and, when available, ML-adjusted predictions with links to original source documents.
Research & Educational Tool
Impact Radar is designed for research and educational purposes. All predictions, scores, and analytics are research tools intended for analysis and learning, not guarantees or investment advice. Always conduct your own due diligence and consult with qualified financial professionals before making investment decisions.
Full strategy framework | Custom entry/exit rules | Position sizing | Comprehensive metrics (Sharpe, Sortino, CAGR) | Continuous learning since November 2025