Sleep Futures: Behavioral Time-Series Quant Research

Built with Python, CatBoost, Pandas, Matplotlib, and Streamlit

Project Overview

Sleep Futures is a behavioral-finance quant research project that transforms daily sleep behavior into a synthetic futures contract and evaluates whether sleep patterns contain predictive directional information.

The project defines a Sleep Index, forecasts its next-day direction using a nonlinear classifier, and simulates a long/short trading strategy with full backtesting, risk metrics, and visualization.

This work is designed as a research and signal-evaluation framework, demonstrating how nontraditional behavioral data can be structured, modeled, and evaluated using real quant methodologies.

Core Concept

Sleep Index Definition:

The Sleep Index measures deviation from a target sleep duration:

$$ SleepIndex_t = hours!sleept_t - 7.5 $$

Positive = overslept Negative = underslept

Synthetic Futures Contract:

A synthetic derivative is defined on the Sleep Index:

$$ Payoff_{t+1} = ContractMultiplier \times SleepIndex_{t+1} $$

Directional exposure is taken based on the model’s forecast.

Trading Logic:

• Predict ↑ → LONG
• Predict ↓ → SHORT
• Predict → → FLAT

Methodology

1. Feature Engineering

• Features combine statistical structure and behavioral context:
• Rolling means (3, 7, 14 days)
• Rolling volatility
• Sleep deficit
• Circadian drift (bedtime deviation)
• Sleep efficiency
• Day-of-week one-hot encoding
• Weekend vs weekday indicators

2. Modeling

• CatBoostClassifier (nonlinear, tabular-friendly)
• Multi-class direction prediction (↑ / ↓ / →)
• Chronological train/test split
• Early stopping to reduce overfitting
• CatBoost was selected for its robustness on small, noisy, nonstationary datasets.

3. Backtesting & Evaluation

• Custom long/short backtesting engine
• Daily PnL calculation
• Cumulative returns
• Sharpe ratio
• Maximum drawdown
• Performance visualization

Results

Directoin Strategy Performance:

• Total PnL: $76.83
• Sharpe Ratio: 9.287
• Max Drawdown: –$13.17

Classification Metrics:

• Accuracy: 66.67%
• Directional Accuracy (ex-neutral): 66.67%

These results indicate strong signal-to-noise within sample, while acknowledging limitations related to dataset size and non-independence

Visualizations & Interpretation

1. Daily Sleep Index

Shows how many hours above/below target slept each day.

Interpretation:

• Positive spikes = overslept
• Negative dips = underslept
• Volatility reflects behavioral inconsistency
• This time series is the core forecasting target

2. Sleep Volatility Histogram

Shows the 7-day rolling standard deviation of SleepIndex.

Interpretation:

• Most 7-day volatility is between 1.0 and 2.5 hours
• High volatility means irregular sleep patterns
• Helps explain why forecasting sleep is harder during chaotic periods

3. Strategy PnL Curve

Displays cumulative profit of the long/short strategy.

Interpretation:

• The model produced positive cumulative PnL
• Sharpe ratio > 9 indicates very high signal-to-noise
• Max drawdown shows the worst peak-to-bottom decline
• This demonstrates alpha generation from behavioral patterns

Project Architecture

.
├── data/
│   ├── raw/                      # Input sleep CSV
├── src/
│   ├── data_loader.py            # Load + clean raw sleep data
│   ├── features.py               # Feature engineering
│   ├── model.py                  # CatBoost classification model
│   ├── backtest.py               # Synthetic futures backtesting
│   ├── visualize.py              # Matplotlib visualizations
│   └── run_pipeline.py           # Full pipeline runner
├── app.py                        # Streamlit dashboard
├── charts/                       # Saved PNG plots
├── requirements.txt
└── README.md

Installation & Usage

1. Clone the repository
git clone https://github.com/YOUR-USERNAME/sleep-futures.git
cd sleep-futures-qm

2. Create virtual env
python3 -m venv .venv
source .venv/bin/activate

3. Install dependencies
pip install -r requirements.txt

4. Add sleep.csv

data/raw/sleep.csv

5. Run the full pipeline
python3 -m src.run_pipeline

6. Run the Streamlit dashboard

streamlit run app.py