CFRM 422/522 Strategy Project: Funding Rate Contrarian

Overview

This project implements and evaluates a contrarian trading strategy on BTC-USDT perpetual futures (Binance). The strategy exploits mean-reversion following extreme funding rate events.

Core hypothesis: When perpetual futures funding rates reach extreme z-score levels, the dominant side of the trade (longs or shorts) is crowded. These crowded positions tend to unwind within 8-24 hours, causing price to mean-revert. We trade against the crowding direction.

Economic mechanism: High funding rates mean longs pay 0.01% or more every 8 hours. At annualized rates of 13%+, longs reduce exposure — especially leveraged players who cannot sustain the cost. This creates predictable short-term selling pressure (Brunnermeier & Pedersen 2009).

Key Results (Optimal Parameters: threshold=2.5, lookback=60d, hold=8h)

Metric	Value
Calmar Ratio	0.111
Sharpe Ratio	0.771
Max Drawdown	13.77%
Win Rate	54.40%
Total Trades	125 (2020-2024)
Avg Hold Period	8 hours

Walk-Forward OOS Performance

Year	Calmar	Trades
2021	-0.587	23
2022	-0.249	25
2023	+0.567	28
2024	+0.634	23

Regime note: Strategy underperformed in 2021-22 (low-funding trending bull/bear), recovered in 2023-24 as funding rate volatility returned.

File Structure

Strategy-project/
|-- strategy_notebook.ipynb   # Main deliverable — all 10 graded sections
|-- generate_notebook.py      # Script that builds the notebook
|-- references.bib            # BibTeX bibliography (9 papers)
|-- README.md                 # This file
|
|-- data/
|   |-- fetch_data.py         # Downloads real Binance data (public API)
|   |-- btc_funding.csv       # BTC-USDT funding rates (2020-2024)
|   |-- btc_ohlcv_1h.csv      # BTC-USDT 1h OHLCV (2020-2024)
|   |-- eth_funding.csv       # ETH-USDT funding rates (for extension)
|   |-- eth_ohlcv_1h.csv      # ETH-USDT 1h OHLCV (for extension)
|   |-- grid_search_cache.pkl # Cached grid search results
|
|-- src/
    |-- indicators.py         # Funding z-score, cumulative, IC computation
    |-- signals.py            # Signal generation with optional OI filter
    |-- backtest.py           # Vectorized backtester, performance metrics
    |-- optimization.py       # Grid search and simulated annealing
    |-- overfitting.py        # DSR, bootstrap Sharpe, top-N removal

How to Run

1. Install dependencies

pip install numpy pandas scipy matplotlib statsmodels jupyter nbformat requests

2. Fetch data (optional — notebook uses synthetic data if this fails)

python data/fetch_data.py --start 2020-01-01 --end 2024-12-31 --symbols BTC ETH

The script uses Binance's public futures API (no API key required).

3. Run the notebook

jupyter notebook strategy_notebook.ipynb

Or execute headlessly:

jupyter nbconvert --to notebook --execute strategy_notebook.ipynb --output strategy_notebook_executed.ipynb

4. Regenerate the notebook (if needed)

python generate_notebook.py

Data Sources

• Funding rates: Binance USDT-M Futures public API (/fapi/v1/fundingRate)
• OHLCV: Binance USDT-M Futures public API (/fapi/v1/klines)
• No API key required for historical data access
• Fallback: If the API is unreachable, the notebook generates realistic synthetic data using GBM with regime switching and historical-style funding rate distributions. This is clearly flagged in the notebook.

Strategy Details

Signal Generation

z = (funding_rate - rolling_mean(window)) / rolling_std(window)
signal = +1 if z < -threshold  (fade crowded shorts)
signal = -1 if z > +threshold  (fade crowded longs)
signal = 0  otherwise

Execution Rules

• Enter at next open after signal fires
• Hold for hold_hours (default: 8h, optimized in Section 7)
• Optional stop-loss at 0.5% adverse move
• Optional OI filter: skip if OI change < 1%
• Fees: 0.04% taker each side
• No leverage (1x exposure)

Objective Function

Calmar Ratio = Annualized Return / Maximum Drawdown

Preferred over Sharpe because BTC return distributions have fat tails and occasional extreme positive events that inflate Sharpe artificially. Calmar focuses on the worst-case loss, which is more relevant for leveraged crypto strategies.

Notebook Sections

Section	Title	Points
0	Setup and Imports	--
1	Hypothesis and Testing Plan	10
2	Constraints, Benchmark, Objective	10
3	Data Description and EDA	10
4	Indicators (Standalone Testing)	10
5	Signal Processing (Standalone Testing)	10
6	Trading Rules (Incremental Testing)	10
7	Parameter Optimization	10
8	Walk-Forward Analysis	10
9	Overfitting Evaluation	10
10	Extension (ETH, regime, Sortino)	10

References

See references.bib for full citations. Key papers:

• Bailey & Lopez de Prado (2014) — Deflated Sharpe Ratio methodology
• Brunnermeier & Pedersen (2009) — Funding liquidity and market liquidity spirals
• Shleifer & Vishny (1997) — Limits of arbitrage
• Pardo (2008) — Walk-forward analysis methodology
• Liu, Tsyvinski & Wu (2022) — Cryptocurrency factor models