Why TSMOM uses a 12-month lookback (and not 6 or 3)

TSMOM (Time-Series Momentum) is one of the simplest and most-cited quantitative strategies in modern finance. The signal is almost embarrassingly simple: if an asset has gone up over the past 12 months, go long; if it has gone down, go short. But there's one detail in the original paper that almost every retail implementation gets wrong, and it matters.

The signal: sign(price[t-21] / price[t-252] - 1)

Most people implement TSMOM as: 'Is today's price higher than 252 trading days ago?' The actual paper formula is: 'Is the price 21 days ago higher than 252 days ago?' That 1-month gap is called the 'skip' and it exists for a very specific reason.

Why skip the most recent month?

Markets show short-term reversal at horizons under one month — this is documented in dozens of academic papers going back to Jegadeesh (1990). After a strong move in the last few weeks, prices tend to pull back slightly before resuming any longer-term trend. If you don't skip the last month, your TSMOM signal gets corrupted by this short-term noise: you'll go long right at the local top of a 1-week rally, then immediately suffer the reversal.

The 21-day skip filters this out. You're basically asking: 'Setting aside the last month of noise, what's the underlying trend over the year?' This is why it works.

Why 12 months and not 6 or 3?

Moskowitz, Ooi, Pedersen tested lookbacks ranging from 1 month to 36 months across 58 different futures contracts. The 12-month window came out as the sweet spot for two reasons:

• Statistical: 12 months is long enough to filter out short-term noise but short enough that the trend hasn't already exhausted itself. Shorter windows (1-3 months) get whipsawed by reversal effects. Longer windows (24+ months) are too slow to react when regimes shift.
• Behavioral: 12 months aligns with how investors and institutions actually think about performance — annual reviews, calendar-year returns, fiscal year reports. This creates a natural reinforcing feedback loop where money flows toward 12-month winners.

What happens if you change the lookback in OpenAlpha?

The strategy has a configurable 'lookback' parameter (default 252 trading days = 12 months). Pro users can override it. Here's roughly what we see when sweeping the parameter on the same dataset:

• Lookback = 63 (3 months): Sharpe drops to ~0.2. Too noisy, too many false signals.
• Lookback = 126 (6 months): Sharpe ~0.4. Decent in directional regimes, fragile in choppy ones.
• Lookback = 252 (12 months, default): Sharpe ~0.6 in our 3-year sample. The sweet spot.
• Lookback = 378 (18 months): Sharpe ~0.5. A bit slow on regime changes.
• Lookback = 504 (24 months): Sharpe ~0.4. Misses too many transitions.

When TSMOM fails

TSMOM has well-known failure modes. The biggest: 'whipsaws' during regime transitions. When a multi-year trend ends, TSMOM is by definition fully positioned in the wrong direction, and it takes 1-2 months for the new signal to flip. During those weeks, the strategy bleeds. The 2009 'risk-on' rebound after the GFC was a textbook example: TSMOM was massively short equities on January 1, 2009 — exactly when the 11-year bull market began.

We mitigate this in OpenAlpha by adding a max-drawdown circuit breaker (default 25%), which forces the strategy to step back when it's been losing for too long. It's not perfect, but it prevents catastrophic losses during regime flips.