Quantitative Greyhound System
A deep dive into a high-frequency algorithmic betting system that generated £2.2M profit over a two-year live period using strictly engineering-led principles.
Net Profit (2 Years)
£2,200,000
Avg Monthly Return
£91,600
Primary Strategy
LAY-First
Risk Model
Fractional Kelly
Cumulative Performance Simulation
Visualizing the equity curve based on reported metrics (Simulated Data)
System Architecture
The alpha comes from a disciplined engineering pipeline, not a single "magic signal". Click a stage to explore.
1. Data & Filters
GBGB feeds, Betfair odds, strict 6-dog filtering, inline SQL operations.
2. Feature Engineering
Trap stats, Trainer forms, Odds momentum. No fabricated columns.
3. Comparative Model
Head-to-head scoring. Probability forecasting. Lay favorites, Back overlays.
4. Execution & Ops
File-based gating. Strict confirmation. Fractional Kelly staking.
Alpha Source: Comparative Scoring
The system doesn't predict "Who will win." It predicts "Who is priced wrong."
Decision Engine
Triggered when Market Prob > Model Prob (Overvalued).
Triggered when Model Prob > Market Prob (Undervalued).
Split exposure across multiple runners to reduce variance.
Operational Gating Log
Simulating real-time race filtering to avoid "Toxic Flow".
Strict Exclusion Rules
Risks & Counterpoints
Why this might fail (The "Pre-Mortem").
📉 Edge Erosion ▼
💸 Execution Slippage ▼
🔍 Overfitting ▼
Portfolio Composition
The system is LAY-dominant. This exploits the structural inefficiency where the public overbets favorites, pushing their prices down (probabilities up) beyond reality.
