Strategy
Trend Following: A Systematic Implementation Guide
Mechanics, fit ranges, and failure modes for trend-following strategies — the oldest profitable systematic style and the hardest to hold psychologically.
Sentivue Capital··7 min read
Trend following is the longest-running profitable systematic style — documented edge across centuries of futures and equity data. It is also the hardest to hold. Win rates of 35–45% with average winners 2–4× average losers means the strategy feels wrong most days while compounding over decades.
Core mechanic
A trend-following rule answers: is this market trending, and if so, in which direction? Two implementations dominate:
- Channel breakout — go long on a break above the rolling N-day high; flat or short on the symmetric break of the rolling low. Original Donchian / Turtle formulation.
- Moving-average crossover — long when fast MA > slow MA. Variants use exponential weighting or volatility-adjusted MAs.
Both produce statistically similar return streams when sized to identical target volatility.
Fit ranges
- Lookback windows: 50–250 days for diversified futures portfolios. Shorter than 50 produces signal noise; longer than 250 misses regime turns.
- Universe: 30+ uncorrelated futures markets (commodities, rates, currencies, equity indices). Diversification is the engine; single-instrument trend-following has a brutal Sharpe distribution.
- Sizing: vol-targeted per instrument. Equal-notional sizing systematically over-bets currencies and under-bets commodities.
Why it works
- Behavioral. Investors are slow to update beliefs; large moves take months to fully price in.
- Risk-transfer. Producers and consumers hedge by holding the opposite side; trend-followers are paid for absorbing that risk.
- Crisis alpha. Trend tends to perform in extended dislocations because dislocations are trends.
Failure modes
- Mean-reverting markets. Range-bound regimes whipsaw breakout systems repeatedly.
- Vol crash. When realized vol collapses, a vol-targeted trend system levers up into a complacent market and gets caught when vol returns.
- Correlation collapse. When markets that historically traded independently become correlated (2008-style), diversification stops working precisely when it's needed.
Implementation discipline
- Don't add filters to smooth equity curve. This is the single most common overfitting vector in trend-following research. See backtest overfitting.
- Trade through drawdowns. Turning the system off in drawdown realizes the loss without capturing the inevitable recovery rallies.
- Monitor at the portfolio level. Per-instrument drawdowns are noise; portfolio-level drawdowns are signal.