Momentum Exits — Research Companion
The evidence behind Momentum Exit & Position Sizing System. Four adversarially-verified deep-research sweeps (each: ~100 agents, ~20 sources, claims fact-checked 3-vote). Confidence and vote counts noted per finding.
Sweep 1 — Academic exit strategies for momentum
Headline: the high-value “exit” is not when to sell one stock on a clock but how to scale total exposure down when crash risk is elevated. Most “exit” findings are portfolio-level de-risking, not single-position sell rules.
- Volatility scaling (Barroso & Santa-Clara 2015, JFE, “Momentum Has Its Moments”): size momentum by inverse of 6-month realized vol → Sharpe 0.53 → 0.97, max drawdown −96.7% → −45.2%, worst month −78.96% → −28.40%. (3-0)
- Dynamic mean-variance scaling (Daniel & Moskowitz 2016, JFE, “Momentum Crashes,” NBER w20439): scale on forecasts of conditional mean + variance, cut in “panic” states → ~doubles alpha and Sharpe. (3-0)
- Crashes are forecastable — cluster after market declines, in high vol, at rebounds. Trigger is state, not position age. (3-0)
- 10% stop-loss (Han, Zhou & Zhu, JFQA 2016): per-stock stop → avg monthly return 0.99% → 1.69%, Sharpe 0.165 → 0.369, worst month −49.79% → −11.36%. (3-0)
- Holding-period decay (Jegadeesh & Titman 2001, JF): cumulative returns negative months 13–60, but reversal not significant until years 4–5 → slow decay is a weak concern vs. crash risk. (3-0)
- 52-week high (George & Hwang 2004, JF): forecast returns do not reverse long-run → signal-based hold, not forced exit. (3-0)
- Refuted: skewness-management dominating across all cost levels (1-2); 52WH nearness being a better predictor than past returns (1-2).
Sweep 2 — Extension risk (the most important correction)
Headline: extension is two different things by timeframe. Multi-day non-fundamental spikes revert; multi-month structural strength continues. Fading genuine strength is counterproductive.
- Short-term reversal (Jegadeesh 1990, Lehmann 1990): past-month winners underperform, ~2%/mo gross — but small/illiquid, net-of-cost contested. (2-1)
- Only the non-fundamental residual reverts (Da, Liu & Schaumburg 2014, Management Science): isolating it ~quadruples reversal profit (alpha 0.33% → 1.34%/mo). Strength driven by real news does not revert. (3-0)
- Extension above a long-run MA is a BUY (Avramov, Kaplanski & Subrahmanyam 2021, Review of Financial Economics): 21d-vs-200d MA distance positively predicts returns, ~9% annualized alpha, incremental to momentum, survives costs. Continuation, not reversion. (3-0)
- No rigorous evidence that trimming-into-extension helps a long-only book; cross-sectional evidence argues against fading strength.
- Best validated exit remains the stop-loss — and it does not cap winners: in the 4 worst momentum months, the stop-loss version returned +1.69%, +2.64%, −6.00%, −3.57% (surviving un-stopped names carried it). (3-0)
- Caveat: the short-MA reversion study (IJECM 2017) is low-tier, in-sample, non-monotonic — suggestive only.
Sweep 3 — Championship exit rules
Headline: the system in the plan is the championship method. Verified rules:
- Hard 7–8% stop, no exception (O’Neil, Ryan, Minervini; tighter on low-vol). (3-0/2-1)
- Sell into strength — unwind into the liquidity of a rising price. (3-0)
- Climax = exit: O’Neil 25–50% surge over 2–3 weeks; Kell “Exhaustion Extension” far above the 10-day MA. (3-0)
- Seven-Week Rule (Morales/Kacher): trail the 10-day MA only after a stock obeys it 7+ weeks, else the 50-day. Violation = close below + next-day undercut of that day’s low. (3-0)
- Schwartz/Kell anchor on short MAs — 10-day EMA as red-light/green-light; 10/20 EMA stack. (3-0)
- Minervini staged exit: limit loss → breakeven on profit → protect profit with trailing stop; never let a 20% gain become a loss. (3-0)
- Pyramid only into winners, never average down; finance adds with paper profit, raise stop so aggregate risk ≈ constant. (3-0)
- Ryan: 5% starters, 7% stops (~0.35% portfolio risk), hold strongest 6–12 months. (3-0)
- Refuted: O’Neil’s “sell at 20–25% profit” rule could not be confirmed (1-2) — don’t build around it.
Sweep 4 — Concentration
Headline: champions run 4–12 names (not 20–30), single positions 12–35%, rejecting diversification as “di-worsification.” But concentration is dynamic and the headline returns are levered.
| Trader | Positions | Max size | Leverage |
|---|---|---|---|
| Zanger | 1–2 | whole book | 2:1 (gave back ~75% in 2000) |
| Ritchie II | ~4 | ~25% | ~2:1 |
| Kell | 6–7 core (good) / 1–2 (poor) | up to 35% | 120–150% |
| Minervini | 4–6 small / 10–12 large (max ~20) | ~25% | varies |
- Progressive exposure (Minervini): 6.25% pilot → 12.5% → 25%; press when winning, cut when losing. (3-0)
- Position size falls out of the risk math: 1.25–2.5% risk ÷ 7–8% stop ≈ 25% full ≈ 4-name book.
- Contest vs. real money: +334.8% (Minervini 2021), +941% (Kell 2020), ~29,000% (Zanger) = short audited windows + ~2:1 margin + hot tape. Not repeatable unlevered process. (3-0)
- Concentration is a regime lever, not a constant: wide (10–12) in good tape, tight (4–6 or cash) in bad. Refuted: any fixed “8–12 max” (0-3 / 1-2).
- Gaps: no verified data this sweep on O’Neil’s account-size guidance, David Ryan, or Schwartz concentration.
The convergence
All four sweeps point the same way: let strength run, exit only on a confirmed break, tighten (don’t sell) into extension, size up via progressive exposure into a concentrated-but-dynamic 8–12 name book, and de-risk on regime — not on a clock. Academic theory (state-based de-risking, no-fade-on-strength, stop-loss doesn’t cap winners) and championship practice (7-8% stop, 50-day line, sell into climax, progressive exposure) describe the same system from two directions.
Cross-cutting caveats: most figures gross of costs; net-of-cost is the field’s weakest spot. Championship headline returns are levered sprints. Short-term reversal edges live in illiquid names. This is a long-only frame — the short-side “momentum crash” machinery is deliberately excluded.