Knowledge · Strategies · Cross-Sectional Momentum

Cross-Sectional Momentum

Jegadeesh & Titman — 'Returns to Buying Winners and Selling Losers' (Journal of Finance, 1993)
Momentum Evidence: Very strong position equitiescryptofuturescommodities
3/10
Relevance for Botty
Long the top winners of the last 3-12 months, short the top losers. The largest anomaly in finance research since 1993.
Within a universe of similar assets, rank by 3-12-month returns. Long leg: top 10-20%. Short leg: bottom 10-20%. Rebalance monthly. One of the most robustly documented anomalies across 30+ years and all asset classes — it remains an unsolved challenge for asset-pricing models.
Relevance Score 3/10
Botty trades single-asset BTC perp — cross-sectional is not applicable. A hypothetical multi-coin Botty (BTC + ETH + SOL + ...) could use momentum ranking: rotate monthly into the top coin by 30-day return. A significant architectural extension would be needed.

Entry

  • Choose a universe (e.g. all top-100 coins on Hyperliquid, or S&P 500 stocks)
  • Month-end: compute the 12-month return for each asset (often 12-1: 12M excluding the most recent month)
  • Sort; long leg = top decile/quintile; short leg = bottom decile/quintile
  • Equal-weight within each leg

Exit

  • Re-rebalance at the next month-end — close on exit from the top/bottom zone
NameTyp. valueDescription
lookback_months 12 (12-1 Standard) Formation period; the most recent month is typically excluded (short-term reversal)
holding_months 1 Rebalance frequency
portfolio_size 10-20% je Leg Deciles are common

Pros

  • The most robust academically documented anomaly
  • Works across all asset classes
  • Simple rule, low costs
  • Rational aggregate of behavioral biases (underreaction, herding)

Cons

  • Momentum crashes: after bear-market bottoms the 'loser' shorts often bounce back first → long-short loses heavily
  • High turnover — fees eat up alpha at the retail level
  • Needs a multi-asset universe
  • Crowded since the 2000s — margins reduced
notes
Robust over 30+ years, all asset classes, except during sudden market reversals (2009, March 2020). The Fama-French 5-factor model cannot explain it.
sharpe
0.5-1.0
max drawdown
20-40% (momentum crashes after bear markets are notorious)
annualized returns
5-15% (long-short), 10-20% (long-only)
Very good for diversified multi-asset bots; not for single-asset bots.

The original paper

Jegadeesh & Titman (1993) showed on US stocks 1965-1989: if you buy the top winners of the last 6 months and short the top losers, you achieve ~1% excess return per month — statistically highly significant, and not explainable by the asset-pricing models of the time.

Thirty years later: the effect holds in all asset classes studied — stocks, bonds, commodities, currencies, crypto. Fama & French called it 'the main embarrassment of the three-factor model' — a persistent phenomenon for which theory has no clean explanation.

The rule (12-1 standard)

  1. Month-end: compute each asset's return over the last 12 months, excluding the most recent month (because of short-term reversal effects).
  2. Rank: sort in descending order.
  3. Long leg: equal-weighted position in the top decile (or top 10%).
  4. Short leg (optional, for long-short): equal-weighted position in the bottom decile.
  5. Rebalance: monthly (weekly is also possible).

Why it works

Research suspects behavioral biases: - Initial underreaction: investors adapt too slowly to news → slow pricing-in of known information - Delayed overreaction: momentum traders and herding amplify the trend with a lag - Anchoring: investors cling to anchor-price-bound old valuations

The combination produces a 3-12-month persistence, followed by long-term reversals (3-5 years).

Momentum crashes

The biggest drawback: after bear-market bottoms (e.g. March 2009, March 2020) the depressed 'loser' shorts explode upward quickly → the long-short portfolio takes heavy losses. Daniel & Moskowitz (2016) documented these momentum crashes in detail.

Modern implementations use volatility scaling (Moskowitz/Pedersen) or trend adjustment (Barroso/Santa-Clara) to dampen crashes.

Crypto evidence

Research across 3,900+ coins 2014-2022: cross-sectional momentum exists with a magnitude similar to that in stocks, especially at the weekly level. The crypto-specific variant favors shorter lookbacks (2-4 weeks) than the equity standard (12M).

Relevance for Botty

Botty trades single-asset — cross-sectional is not directly applicable. If Botty were extended to multi-coin (e.g. the top 50 Hyperliquid perps):

  • Weekly ranking by 14-day return
  • Long the top 3-5, equal-weighted
  • (no short leg initially, to avoid momentum crashes)

That would be a substantial extension, but on a robust evidentiary foundation.