How the AI model works. Documented. Disclosed. Verifiable.
The methodology behind the 75%+ directional accuracy claim includes the input categories, the backtesting framework, the measurement approach, and the conditions under which the model was tested. Not every proprietary detail is disclosed here. But everything you need to evaluate the accuracy claim honestly is.
How Accuracy is Measured
StakerIQ's deep learning model processes quantitative inputs across five primary categories. Within each category, multiple specific variables are analysed. The specific variables and their ML-derived weighting factors are proprietary: they represent the analytical framework that produces the 75%+ accuracy figure. The category-level framework is documented here.
Price Action
Momentum metrics across multiple timeframes, rate-of-change relative to historical norms, trend strength indicators, and pattern recognition across tested historical configurations. The model does not simply identify chart patterns, it quantifies price dynamics across multiple dimensions simultaneously.
Volume Dynamics
Trading volume relative to 20-day and 90-day historical averages, volume-price confirmation and divergence analysis, accumulation and distribution signal detection, and relative volume against sector and market-cap peers.
Fundamental Metrics
Revenue growth trajectories, margin trend analysis, debt-to-equity dynamics, and selected valuation metrics are each weighted by their historically demonstrated relationship to directional price movement in backtested data. Not all fundamental variables are equally predictive. The model weights them accordingly.
Sector Relative Performance
How the stock positions against its sector peers across every input category. The model applies sector-adjusted benchmarking, meaning a stock's quantitative profile is evaluated relative to its sector context, not in isolation. The same fundamental strength in a weak sector carries different analytical weight than in a strong sector.
Volatility and Risk Metrics
Historical volatility relative to the stock's own rolling norms, implied volatility dynamics where applicable, drawdown characteristics from historical data, and the current volatility regime. The model dynamically adjusts its variable weighting based on the volatility environment, so inputs that are predictive in low-volatility regimes carry different weight in high-volatility regimes.
The Backtesting Framework
How 75%+ directional accuracy was measured.
Directional accuracy definition
The model is evaluated solely on whether its directional indication (Bullish/Bearish) matched the actual price trajectory of the stock at the conclusion of a 30-day window.
Market conditions tested
Bull market periods (sustained upward price trends), bear market periods (sustained downward trends), high-volatility environments (VIX > 25), and low-volatility compression periods (VIX < 15). The 75%+ figure reflects aggregate performance across all of these conditions, not performance in any single favourable environment.
Measurement method
The model generates a directional score on a stock at a specific historical date. Thirty days later, the actual price direction is recorded. If the directional score indicated upward movement and the price moved upward within 30 days, the call is counted as correct. If the price moved in the opposite direction, the call is counted as incorrect. Neutral or mixed outcomes are recorded separately.
Reporting approach
The 75%+ accuracy figure represents the aggregate of all directional calls tested (correct, incorrect, and mixed) across the full backtesting dataset. Peak-period performance figures are not used. The full distribution is used.
Market condition coverage
The 75%+ figure is not derived from a single favorable market period. The backtesting dataset includes bull regimes, sustained bear markets, and high-volatility spikes to ensure the model's predictive capacity is not dependent on a rising broader market.