AI/ML Methodology
How our models analyze options data and generate strategy considerations
Non-Discretionary Advisory Disclosure
Ohey Inc. is registered with the SEC as an investment adviser (SEC File No. 801-135267, CRD# 340093), providing personalized, non-discretionary investment advisory services. All strategy considerations generated by our AI/ML models are delivered through our platform for your independent evaluation. You make all trading decisions. Ohey does not execute trades on your behalf, and no model output constitutes a trade order or instruction to trade.
Overview
The Ohey platform uses a 16-model machine learning ensemble operating across a 7-layer algorithm stack to analyze options market data and present strategy considerations. This page describes the architecture, data inputs, decision logic, and limitations of the system.
The system processes options data from Polygon.io market data feeds (aggregating OPRA exchange data), computes derived analytics (gamma exposure, implied volatility surfaces, dealer positioning, options flow), and feeds these into the ML ensemble. The ensemble outputs directional predictions, regime classifications, and confidence scores across multiple time horizons.
7-Layer Algorithm Stack
Each analysis request passes through 7 independent analytical layers. Each layer contributes a signal that is weighted and combined to produce the final confidence score and strategy consideration.
| # | Layer | What It Analyzes | Key Metrics |
|---|---|---|---|
| 1 | Gamma Topology | Net gamma exposure (GEX) direction and magnitude across all strikes and expirations | Total Call GEX, Total Put GEX, Net GEX, GEX by strike |
| 2 | IV Surface Analysis | Implied volatility levels, z-scores relative to historical norms, and term structure shape | IV z-score, IV percentile, term structure slope, vol-of-vol |
| 3 | Greeks Exposure | Aggregate positioning across delta, gamma, theta, and vega | Net delta, aggregate theta decay, vega exposure by expiration |
| 4 | Options Flow | Real-time order flow direction, size classification, and buyer/seller identification | Flow direction, net premium, institutional vs. retail ratio |
| 5 | Regime Detection | Current market regime classification using IV z-score analysis and price velocity | Regime label (Trending, Mean-Reverting, Volatile, Calm), regime confidence |
| 6 | ML Predictions | Ensemble model outputs for directional probability and expected move magnitude | Directional probability, expected move %, prediction horizon |
| 7 | Skew Analysis | Put-call skew patterns, term structure slope, and sentiment inference | Skew value (σ), skew percentile, put-call ratio |
16-Model ML Ensemble
The ML prediction layer (Layer 6) uses an ensemble of 16 independent models. Each model is trained on different feature subsets and time windows to reduce overfitting and improve robustness. The ensemble combines predictions through weighted averaging, where weights are determined by each model's recent out-of-sample performance.
- Gradient boosting models (XGBoost, LightGBM)
- Neural network models (LSTM, attention-based)
- Statistical models (regime HMMs, GARCH variants)
- Weighted ensemble aggregation with dynamic weight updates
- 0DTE: Intraday predictions for same-day expiration
- 1-Week: Short-term directional outlook
- 2-Week: Medium-term trend analysis
- 4-Week: Longer-term regime and trend assessment
- Options chain data via Polygon.io (OPRA-sourced)
- Computed gamma exposure by strike
- Implied volatility surface snapshots
- Real-time options flow and premium data
- Historical price and volume data
- Models retrained on rolling windows
- Ensemble weights updated based on recent accuracy
- Feature importance monitored for drift detection
- Out-of-sample validation enforced on all updates
Confidence Scoring
Every strategy consideration includes a confidence score (0–100%) computed from a 6-factor agreement model. The confidence score reflects the degree of agreement across the analytical layers — not a probability of profit.
How Confidence Is Computed
The system evaluates agreement across 6 factors: gamma topology direction, IV regime classification, options flow direction, skew sentiment, ML ensemble prediction, and regime detection output. When all 6 factors agree on direction and magnitude, confidence approaches 100%. When factors conflict, confidence drops accordingly.
| Confidence Range | Interpretation | Typical Strategy Types |
|---|---|---|
| 75–100% | Strong agreement across most analytical layers | Bull Call Spread, Bear Put Spread (defined-risk directional) |
| 60–74% | Moderate agreement with some conflicting signals | Long Call, Long Put (single-leg directional) |
| 50–59% | Weak directional signal, neutral regime likely | Iron Condor, Iron Butterfly (range-bound strategies) |
| Below 50% | Insufficient agreement for any strategy | Wait — no strategy consideration presented |
Important: Confidence scores represent model agreement across analytical factors at a point in time. They are not predictions of trade profitability and should not be interpreted as such. Market conditions can change rapidly, and historical model agreement does not guarantee future accuracy.
Regime Detection
The regime detection layer classifies the current market environment based on implied volatility z-scores and price velocity. This classification determines which strategy types are considered appropriate:
| IV Z-Score | Regime | Confidence | Strategy Implications |
|---|---|---|---|
| > 2.0 | Extremely Stressed | 95% | Elevated volatility — volatility-selling strategies may be considered |
| > 1.5 | Stressed | 85% | Above-average volatility — directional trades or vol strategies |
| > 1.0 | Elevated | 75% | Mildly elevated — spreads and defined-risk strategies |
| < -1.5 | Compressed | 85% | Unusually low volatility — vol-expansion strategies considered |
| |price velocity| > 0.02 | Trending | 70% | Strong directional momentum detected |
| Default | Normal | 50% | No strong signal — caution advised |
Strategy Selection Logic
Based on the ensemble output and regime classification, the system selects one of the following strategy types for your evaluation. The selection follows a deterministic rule set — there is no subjective or discretionary judgment involved:
| Regime | Confidence | Strategy Presented |
|---|---|---|
| Bullish | > 75% | Bull Call Spread |
| Bullish | 60–75% | Long Call |
| Bearish | > 75% | Bear Put Spread |
| Bearish | 60–75% | Long Put |
| Neutral | > 50% | Iron Condor |
| Neutral (tight range) | > 50% | Iron Butterfly |
| Volatile (expansion) | > 50% | Long Straddle |
| IV term structure anomaly | > 50% | Calendar Spread |
| Any | < 50% | Wait (no strategy) |
Each strategy consideration includes a written rationale explaining which analytical factors drove the selection. The rationale is generated dynamically based on the specific data conditions at the time of analysis.
Trader Level Personalization
The suitability questionnaire (accessible at Settings → Profile → Suitability) also captures your experience level. This trader level setting controls the complexity and type of strategy considerations presented — distinct from risk tolerance, which controls thresholds and allocation limits.
| Trader Level | Strategy Complexity | Strategy Types Shown |
|---|---|---|
| Beginner | Simple, 2-leg maximum | Long Call, Long Put, Covered Call. Multi-leg strategies suppressed. |
| Intermediate | Defined-risk spreads | Bull Call Spread, Bear Put Spread, Long Straddle added |
| Advanced | Full strategy set | Iron Condor, Iron Butterfly, Calendar Spread added |
| Professional | Full set + complex multi-leg | All strategies plus ratio spreads, diagonal spreads, and volatility overlays |
| Institutional | Unrestricted output | Full strategy output with maximum position size limits removed. Intended for institutional desks with their own risk controls. |
If the suitability questionnaire has not been completed, the system defaults to Intermediate level — showing defined-risk spread strategies but suppressing the most complex multi-leg structures. Trader level affects only which strategy types are presented; it does not alter the underlying model computations or confidence scores.
Risk Tolerance Personalization
Strategy considerations are filtered based on your risk tolerance setting, which you configure during onboarding or in Settings → Profile. Higher risk tolerance surfaces more strategy types and accepts lower confidence thresholds:
| Risk Level | Min. Confidence Threshold | Available Horizons | Max Options Allocation | Max Position Size |
|---|---|---|---|---|
| Very Low | 75% | 1W, 4W | 15% | 3% |
| Low | 70% | 1W, 4W | 20% | 3% |
| Moderate (default) | 60% | 0DTE, 1W, 2W, 4W | 30% | 5% |
| High | 50% | All | 50% | 8% |
| Very High | 40% | All | 80% | 10% |
Risk tolerance affects which strategies and time horizons are presented, not the underlying model computations. A "Very Low" risk tolerance user will only see strategies with confidence above 75% on the 1W and 4W horizons, while a "Very High" user may see more speculative or shorter-duration considerations.
Data Sources
All model inputs are derived from market data sourced via Polygon.io:
- Polygon.io market data API: Options chain data, pricing, implied volatility, and greeks sourced from OPRA-consolidated exchange data
- Derived Analytics: Gamma exposure, dealer positioning models, IV surfaces, and flow analytics are computed internally by Ohey's backend engine from raw chain data
- Data freshness: Updates during U.S. equity market hours (9:30 AM–4:00 PM ET). Outside market hours, models operate on closing snapshot data
For full details on data delivery architecture, see Data Infrastructure →
Limitations & Important Disclosures
- Not a guarantee of performance: Model outputs are analytical considerations, not guaranteed profitable trades. Past model accuracy does not predict future results.
- Point-in-time analysis: All outputs reflect market conditions at the moment of computation. Market conditions can change rapidly and unpredictably.
- No execution: Ohey does not execute trades on your behalf. All trading decisions remain yours. Strategy considerations are presented for your independent evaluation.
- Confidence ≠ probability of profit: Confidence scores measure agreement across analytical factors, not expected return or probability of a profitable outcome.
- Model limitations: ML models can underperform in unprecedented market events, flash crashes, or conditions outside their training data distribution.
- Options risk: Options trading involves significant risk of loss and is not appropriate for all investors. You should understand the risks before trading options.
- Third-party data: Data is sourced from Polygon.io market data API. While Polygon.io aggregates OPRA exchange data, we cannot guarantee data accuracy or completeness at all times. Data reflects exchange-reported values and may experience delays during high-volume periods.
SEC Registration
Ohey Inc. is registered with the SEC as an investment adviser:
| SEC File Number | 801-135267 |
| CRD Number | 340093 |
| Advisory Type | Non-discretionary |
| Services | Personalized, non-discretionary investment advisory services delivered through our web platform |
Registration with the SEC does not imply a certain level of skill or training. For more information, visit IAPD (Investment Adviser Public Disclosure) →
Related Pages
- PulseDeck™ — Unified dashboard with strategy considerations
- Data Infrastructure — How market data flows through the platform
- FAQ — Frequently asked questions
- RegimeShift — Detailed regime detection and classification