Purpose
Uses percentile ranking of historical volatility (HV) to capture volatility mean reversion: build positions in low-volatility regimes while waiting for volatility expansion, and exit or short in high-volatility regimes to capture contraction.
Signal Logic
- Compute HV: annualized standard deviation of returns over the past
hv_windowdays - Percentile ranking: percentile position of HV within the past
lookbackdays (0-100) - Signal generation:
- Percentile <
low_pct→ go long (volatility is low, waiting for expansion) - Percentile >
high_pct→ exit / go short (volatility is high, waiting for contraction) - Middle region → keep the current position
- Percentile <
Key Implementation Details
- HV =
returns.rolling(hv_window).std() * sqrt(252)(annualized) - Percentile =
hv.rolling(lookback).rank(pct=True) * 100 - For cryptocurrencies, use 365 instead of 252 as the annualization factor
Parameters
| Parameter | Default | Description |
|---|---|---|
| hv_window | 20 | Historical volatility calculation window |
| lookback | 120 | Lookback period for percentile ranking |
| low_pct | 20.0 | Low-volatility threshold (percentile) |
| high_pct | 80.0 | High-volatility threshold (percentile) |
| annualize | 252 | Annualization factor (252 for China A-shares, 365 for crypto) |
Common Pitfalls
- Before the lookback window is filled, there is not enough data to compute percentiles, so the signal should be 0 (
fillna) - Volatility is not direction. Going long in low-volatility regimes does not guarantee price appreciation; it only means volatility expansion is statistically more likely
- Cryptocurrencies trade 7x24, so
annualizeshould be set to 365
Dependencies
pip install pandas numpy
Signal Convention
1= long (low-volatility regime),-1= short (high-volatility regime),0= stand aside