ProfitScout: High Gamma Trading System

ProfitScout is an autonomous machine learning system designed to identify stocks poised for explosive, short-term price moves (High Gamma). Unlike traditional "trend following" systems, it ignores slow movers and specifically targets volatility velocity.

Target: (Next_Day_Close - Close) > 0.5 * ATR(14)
Philosophy: "Buy Volatility, Sell Velocity."

---

🎯 The "Sniper" Methodology

The core innovation of ProfitScout is its Dynamic Thresholding system. Most models choke on market noise because they force a prediction for every stock, every day. ProfitScout operates differently:

1. Strict Filtering: The model is trained to recognize the "perfect setup."
2. The "Top 1%" Rule: During training, we calculate the probability threshold required to be in the Top 100 highest-confidence historical trades.
   • Current Threshold: 0.814 (81.4% Confidence)
3. Actionable Precision:
   • Baseline Precision (Random Guess): ~12%
   • ProfitScout Precision (at Threshold): ~36%
   • Implication: When the model pulls the trigger, the odds of a massive (>0.5 ATR) move trip by 3x.

The "Daily 50"

Every evening, the system scans the entire market (~1,500 tickers) and delivers a ranked list of the Top 50 setups.

• Tier 1 (Sniper): Probability > 0.814. These are the "Fat Pitches."
• Tier 2 (Watchlist): Top ranked but < 0.814. Good setups, but require manual confirmation.

---

🧠 The Brain: XGBoost & Features

The model does not care about "company fundamentals." It only cares about Price Structure and Velocity.

Key Drivers (Feature Importance)

1. Close Location Value (CLV): Where did the price close relative to the day's High/Low?
   • Signal: A close near the High suggests institutional buying into the close.
2. Distance from Low: How far has it already moved?
3. Momentum: ROC (Rate of Change) 1, 3, 5 days.
4. Volatility Compression: Bollinger Band Width & ATR Ratio.

Note: The model has "learned" that a strong close (CLV > 0.8) combined with rising short-term momentum (ROC) is the single best predictor of a gap-up or continuation.

---

⚙️ Automated Workflow

The system runs autonomously on Google Cloud Vertex AI.

1. Daily Inference (Mon-Fri @ 6:30 PM EST)

• Trigger: Market Close.
• Process:
  1. Loads fresh OHLCV data from BigQuery.
  2. Engineers features for the latest trading day.
  3. Applies the "Golden" model and thresholds.
  4. Saves the Top 50 picks to BigQuery: profit_scout.daily_predictions.

2. Weekly Retraining (Sun @ 9:00 AM EST)

• Trigger: Weekly Reset.
• Process:
  1. Retrains the model on the absolute latest data (capturing new market regimes).
  2. Recalculates the "Sniper Threshold" based on recent performance.
  3. Promotes the new model to the production environment automatically.

---

📊 Results & Performance

Metric	Value	Meaning
Target	> 0.5 * ATR	Predicting massive, tradable volatility.
PR-AUC	0.22	Area Under the Precision-Recall Curve (Global).
Precision @ 100	0.36	Win rate for the Top 100 predictions.
Risk/Reward	High	Winners typically pay 3:1 or 4:1 due to the magnitude of the move.

Status: Fully Operational. Inference and Training schedules are active.