Executive Summary

The explosion of short-dated options liquidity (0-7 DTE) has fundamentally altered equity market microstructure. Traditional indicators break down because option surface kinematics move non-linearly due to accelerating Greek decay curves. This technical brief details why Gammatic utilizes an optimized Gradient Boosting (XGBoost) architecture to evaluate directional regime probabilities, isolating structural edges across distinct tactical and strategic time horizons.

The Non-Linearity of Option Surface Kinematics

Retail technical indicators like the Relative Strength Index (RSI) or Stochastic oscillators assume that historical price movements are predictive of future price trajectories. In modern options-dominated markets, this assumption is structurally invalid. The underlying equity market is increasingly driven by the options market, specifically through the hedging requirements of institutional market makers.

As a short-dated option nears expiration, its Gamma and Vanna accelerate non-linearly. When an underlying index moves toward a major concentration of open interest, market makers are forced to aggressively buy or sell billions of dollars of the underlying equity to maintain a delta-neutral book. This creates rapid, non-linear acceleration that simple price-based indicators cannot quantify. To accurately project market direction, a quantitative model must evaluate the total options surface architecture simultaneously.

Building a 27-Feature Boosting Architecture

To navigate this multi-dimensional environment, Gammatic built an advanced data pipeline running an optimized Extreme Gradient Boosting (XGBoost) model. XGBoost is uniquely suited for financial data because it constructs an ensemble of sequential decision trees, where each new tree corrects the specific classification errors made by the previous ones. This prevents the model from falling into the traps of simple linear regressions or overly rigid neural networks.

Instead of looking at raw price data, our pipeline processes a high-density matrix of 27 structural features simultaneously. These feature weights include:

• Real-time implied volatility skew differentials (the premium variance between out-of-the-money puts and calls).

• Second-order and third-order option Greeks (Vanna, Volga, and Charm decay curves).

• Accelerated institutional order-flow imbalances and localized liquidity voids.

By analyzing how these 27 features interact historically relative to institutional flows, the machine learning engine outputs an objective, probability-weighted directional bias, completely free from human bias or subjective chart interpretation.

Dual-Horizon Isolation: Tactical vs. Strategic Matrices

A model optimized for a 48-hour order flow imbalance will completely fail if applied to a monthly portfolio rebalancing cycle. Gammatic solves this problem by training separate machine learning pipelines to isolate two distinct trading horizons:

1. The Tactical Matrix (2–7 DTE Horizon): This engine optimizes its tree structures to isolate micro-term liquidity flips and dealer hedging acceleration. It is designed specifically for short-term tactical operations and weekly premium captures where gamma risk is highly compressed.

2. The Strategic Matrix (21 DTE Horizon): This pipeline scales out its perspective, adjusting its node depth to model macro volatility skew-adjusted boundaries, monthly theta decay curves, and broad institutional rotation trends.

This dual-horizon architecture ensures that terminal operators never deploy a short-term strategy against a long-term macro current, keeping their execution rules perfectly aligned with true institutional position sizing.