Knowledge · Strategies · Time-Series Momentum (TSMOM)

Time-Series Momentum (TSMOM)

Moskowitz, Ooi & Pedersen (AQR / NYU Stern) — JFE 2012
Trend Following Evidence: Very strong position futuresequity_indicesfxbondscommoditiescrypto
3/10
Relevance for Botty
Go long an asset when it has performed positively over the past 1-12 months; short when negative. Robust across 58 futures markets.
Core concept
An asset's own past performance (absolute, not relative) predicts the near-term future. 1-12 months of persistence, then partial reversal. Explains the bulk of managed-futures hedge fund returns.
Relevance for Botty
Relevance Score 3/10
SCOUTED 2026-06-23 on 20 HL perps (Binance-UM daily, 2021-2026, vol-targeted, after costs — backtesting/tsmom_scout.py + tsmom_xsect_overlay.py). (1) STANDALONE weak: long/short is broadly positive (Sharpe>0 in 6/6 lookbacks) but low — Sharpe 0.75@lb30 declining to 0.40@lb90 to 0.19@lb180; net +3.6..+14%/yr. Convex trend-following signature confirmed (BTC-β≈0, crisis-α strongly positive), but below the deploy threshold. (2) LONG-ONLY trap: the higher Sharpe (up to 1.17) is pure crypto beta (β+0.30, crisis-α -500..-660%/yr) — redundant with DONCHIAN, no TSMOM value-add. (3) OVERLAY on XSECT refuted: every TSMOM dose LOWERS the Sharpe of the market-neutral book (1.49 solo -> 1.12 at 50/50); the maxDD does drop (-19%->-10.5%) but the Calmar stays the same (1.38 vs 1.34) -> pure de-risking, no diversification edge (the same DD is reachable by de-leveraging XSECT). Correlation positive (0.15-0.28), not negative. Walk-forward: combo >= XSECT in only 2/6 years. CONCLUSION: real (confirmed by the literature) but too weak for Botty — no dedicated wallet.
Rules

Entry

  • Choose a lookback window (classically: 12 months, also 1/3/6/12 combined)
  • Compute the cumulative excess return over the window
  • Long if positive, short if negative — binary or scaled by volatility (vol-targeting)
  • Rebalance monthly

Exit

  • Signal flip at the next rebalance date
  • Vol target: shrink the position when realized vol exceeds the target
  • An ATR-based trailing stop is possible as a supplement
Parameter
NameTyp. valueDescription
lookback_days 252 (12M) Window for the momentum calculation; longer = less noise, more lag
rebalance_interval monthly How often it is re-evaluated
vol_target_annual 10-20% Target volatility for position-size scaling
Pros & Cons

Pros

  • Strongest academically documented anomaly alongside value
  • Robust across all asset classes
  • Performs better than buy-and-hold in crises (crisis alpha)
  • Easy to automate, low trade frequency

Cons

  • Long drawdown phases possible in trendless markets
  • Weaker performance since the 2010s (crowded factor)
  • Needs several asset classes for diversification to reach full effect
Typical performance
cagr
~10-15% for diversified portfolios
notes
Uncorrelated with classic asset-pricing factors. Managed-futures hedge funds load fully on TSMOM.
sharpe
~1.2 diversified across asset classes
best regime
Extreme markets (up or down) — performs best in crises
worst regime
Extended phases after financial crises; whipsaw sideways markets
Bot suitability
Excellent in theory — low frequency, clear signals. But empirically too weak on our crypto universe (see botty_applicability).
Background

Core idea

Time-series momentum is the simplest and best-documented form of trend-following: if an asset has run positive over the past 12 months, it tends to keep running positive in the next month. Not relative to other assets (that would be cross-sectional momentum), but against itself.

Why it works

Moskowitz, Ooi & Pedersen (2012) documented the effect across 58 liquid futures markets — equities, FX, bonds, commodities. The authors attribute it to initial underreaction (investors adjust to news too slowly) and delayed overreaction (momentum traders enter with a lag).

Particularly important: the managed-futures hedge fund index loads fully on the TSMOM factor. Almost the entire average performance of this multi-billion-dollar industry can be explained by it.

Practical rules

  1. Choose a lookback window — 12 months is standard; many funds combine 1/3/6/12M equally weighted.
  2. Compute the return over the window (excess return after the risk-free rate).
  3. Long on a positive return, short on a negative one — or binary, leveraged to a vol target.
  4. Rebalance monthly (weekly is also possible; transaction costs rise).

Weaknesses

  • Crowded: has weakened since the 2010s — too many CTAs play it.
  • Whipsaw: in sideways-volatile markets it flips constantly, and costs eat the returns.
  • Long drawdowns: 2-3 years of underperformance are not uncommon.

Application to Botty

Botty trades BTC perp on Hyperliquid — single-asset, so a classic TSMOM portfolio is not possible. But: TSMOM as a regime filter works. Example: only accept long signals when BTC's 90-day return is positive. This excludes roughly 30-40% of the historically worst signals without cutting into the good ones.

Literature

  • Moskowitz, Ooi, Pedersen — Time Series Momentum, JFE 2012 (original paper, freely available on AQR)
  • Clenow — Following the Trend (practical implementation as a managed-futures system)
  • Hurst, Ooi, Pedersen — A Century of Evidence on Trend-Following Investing (215-year backtest)