devsachan23/crude_oil_quant

🛢️ Crude Oil Systematic Walk-Forward Strategy

Brent Crude Total Return Index (2016–2024) Sharpe: 0.847 vs Baseline: 0.462

📌 Project Overview

This project implements a systematic multi-signal trading framework on the Brent Crude Total Return Index using:

• —Diversified signal construction (Trend, Mean Reversion, Volatility-aware, Orthogonal)
• —Walk-forward adaptive strategy selection
• —Volatility targeting (15% annualized)
• —Transaction cost modeling (0.015% per trade)
• —Regime attribution analysis

The primary objective is to maximize out-of-sample Sharpe ratio in a realistic setting, strictly avoiding look-ahead bias.

All strategy selection decisions are made using only information available at each rebalance date.

⚙️ Strategy Architecture

🔹 Signal Groups (11 Base Signals)

Trend-Following

• —1M Momentum
• —3M Momentum
• —200D Moving Average Crossover

Mean Reversion

• —20D Z-score Reversion
• —5D Reversal
• —Percentile Reversion

Volatility-Aware

• —Vol-Adjusted Momentum
• —Volatility Breakout
• —Volatility Trend

Orthogonal

• —Carry Proxy (Momentum Spread)
• —Momentum × Volatility Filter

Each base signal is extended with a volatility-scaled variant.

Final candidate pool:

• —11 base
• —11 vol-scaled
• —1 equal-weight ensemble = 23 competing strategies

🔄 Walk-Forward Framework

Every 6 months:

1. 1.Evaluate all 23 candidate strategies over the previous 18 months
2. 2.Score using composite metric:0.6 × Sharpe + 0.4 × Sortino
3. 3.Select best performer
4. 4.Deploy out-of-sample until next rebalance

This process ensures:

• —No look-ahead bias
• —Adaptive strategy selection
• —Reduced overfitting risk

Walk-forward logic mirrors the provided WalkforwardBacktester implementation.

🎯 Risk Management

• —Transaction Cost: 0.015% per trade
• —Volatility Target: 15% annualized
• —Volatility Estimate: 63-day EWM (1-day lag)
• —Leverage Scalar Clipped to: [0.25×, 4×]

Volatility targeting is applied before strategy selection so all candidates compete on equal risk footing.

📊 Performance Summary

Improvement is driven primarily by disciplined volatility control and adaptive selection rather than aggressive directional forecasting.

The strategy demonstrates resilience during high-volatility regimes, particularly during the 2020 oil crash.

🌡️ Regime Attribution

Performance is decomposed across:

• —High Volatility (>65th percentile)
• —Low Volatility (<35th percentile)
• —Neutral Regime

Results indicate consistent performance across regimes, with enhanced stability during extreme volatility periods.

💻 Interactive Dashboard

An interactive Streamlit dashboard allows:

• —Strategy selection
• —Walk-forward parameter tuning
• —Volatility targeting adjustments
• —Regime filtering
• —Rolling Sharpe visualization
• —Selection frequency diagnostics
• —Full performance comparison

Run locally:

streamlit run app.py
📂 Repository Structure
app.py                         # Streamlit dashboard
backtest.py                    # Walkforward backtester class
crude_oil_strategy.ipynb       # Research notebook
crude_oil_strategy.html        # HTML export of notebook
brent_index.xlsx               # Data file
performance_summary_improved.csv
requirements.txt
README.md
📦 Installation
pip install -r requirements.txt
🧠 Design Philosophy
This project emphasizes:

Out-of-sample integrity

Adaptive systematic allocation

Risk-adjusted performance over raw returns

Clean, reproducible implementation

The objective is disciplined quantitative design rather than curve-fitting.

🚀 Future Extensions
Multi-commodity extension

True futures carry implementation

Cross-asset regime conditioning

Longer historical backtesting

👤 Author
Dev Sachan
Quantitative Strategy Project
2026