Trading Volatility Cooling - A Trend-Following Strategy in Python

In the world of crypto trading, high volatility is a double-edged sword. While it offers profit potential, it often precedes violent reversals. This article explores a systematic strategy designed to enter Ethereum (ETH) positions when the price is trending up but the “chaos” (volatility) is starting to settle—a concept known as Volatility Cooling.

---

The Core Thesis

Most traders chase breakouts during peak volatility. Our strategy does the opposite. It seeks a “quiet” entry into an existing trend. We look for three specific conditions:

1. Price Momentum: The current price is higher than it was \(n\) days ago.

2. Trend Confirmation: The price is above a long-term Moving Average.

3. Volatility Cooling: Volatility is high but decreasing relative to its recent past.

---

1. Defining the “Cooling” Logic

We use a Volatility Oscillator to measure the rate of change in standard deviation. If price is climbing while this oscillator drops, it suggests a healthy, sustainable trend rather than a blow-off top.

Python

# Volatility Oscillator calculation
vol = ret.rolling(vol_w).std()
vol_osc = (vol / vol.shift(vol_roc) - 1.0) * 100.0

# The "Cooling" Signal
vol_cooling = (vol_osc - vol_osc.shift(price_lb)) < 0
base_bull = price_up & vol_cooling

2. Risk Management with ATR Trailing Stops

To protect capital, we don’t use a fixed percentage stop. Instead, we use the Average True Range (ATR). This adjusts the stop-loss distance based on the asset’s current “noise” level. As the price moves in our favor, the stop trails behind, locking in gains.

Python

# Trailing stop logic inside the loop
if pos == 1:
    new_stop = price - float(atr.iloc[i]) * atr_m
    # Only move the stop UP (never down)
    stop = new_stop if np.isnan(stop) else max(stop, new_stop)

    if float(l.iloc[i]) <= stop:
        pos = 0 # Exit position
        ex[i] = True

3. Mass Backtesting & Optimization

Using vectorbt, we can test hundreds of parameter combinations (different window lengths for moving averages and volatility) simultaneously. We filter for a minimum trade count to avoid “fluke” results and rank the rest by their Sharpe Ratio.

Python

# Running the backtest across all parameter combinations
pf = vbt.Portfolio.from_signals(
    close=c,
    entries=entries,
    exits=exits,
    fees=0.001,
    freq="1D"
)

# Filtering for robustness
trade_count = pf.trades.count()
mask = trade_count >= 10
best_index = sharpe.where(mask).idxmax()

4. Results

BEST PARAMETERS (by Sharpe, filtered) (LONG-ONLY)
ticker: ETH-USD | period: 1y | fees: 0.001 | min_trades: 10
vol_window:     14
vol_roc_period: 5
price_lb:       10
trend_window:   30
atr_window:     7
atr_mult:       1.0
Trades:         21
Sharpe:         1.831
TotalRet:       74.52%

---

Why This Works

By shifting the data (df.shift(1)), we ensure there is no lookahead bias—the code only “sees” information available at the time of the trade. The combination of trend-following and volatility-filtering creates a smoother equity curve than simply “buying and holding” through ETH’s massive drawdowns.