The simplest signal is the sign of the past return:

Raw return signal:

• signal = sign(r_t-N:t) where r is the cumulative return over N periods
• Binary: +1 (long) or -1 (short)
• 12-month lookback is the canonical choice from the academic literature

Risk-adjusted signal:

• signal = r_t-N:t / sigma_t where sigma is realized volatility
• Continuous: signal strength reflects conviction
• Better risk-adjusted performance than binary signals

EWMA (Exponentially Weighted Moving Average) variants:

• Instead of a fixed lookback window, use the difference between a fast and slow EWMA of prices
• signal = EWMA_fast(price) - EWMA_slow(price)
• Common pairs: (8, 32), (16, 64), (32, 128) day half-lives
• Smoother transitions, fewer whipsaws than binary signals
• This is what most institutional trend followers actually use

No single lookback window dominates across all regimes. Best practice is to blend signals across multiple horizons:

• 1-month (21 days): Captures fast-moving trends, higher turnover, more whipsaws
• 3-month (63 days): Medium-term, balances responsiveness and stability
• 12-month (252 days): Classic academic signal, slow-moving, lowest turnover

A common blending approach:

combined_signal = w1 * signal_1M + w3 * signal_3M + w12 * signal_12M

Equal weighting (1/3 each) is a strong default. AQR and other practitioners have shown that blending across lookbacks improves Sharpe ratios by 0.1-0.3 versus any single window, because different horizons capture different trend speeds.

This is the critical implementation detail that separates naive momentum from institutional-grade strategies. Raw momentum signals produce wildly different risk exposures across assets (e.g., natural gas vs. bonds). The fix:

position_size = signal * target_vol / realized_vol

• target_vol: Desired annualized volatility per position (e.g., 10% or 15%)
• realized_vol: Trailing estimate of asset volatility (e.g., 60-day exponential)
• Effect: Every position contributes roughly equal risk. A low-vol asset gets more leverage; a high-vol asset gets less.

This directly connects to Quant - Volatility Targeting at the portfolio level. Combine per-position vol scaling with portfolio-level vol targeting for a complete risk management framework.