MB L Fade

Mean-reversion long · All regimes · Demo

MB L Fade goes long the most beaten-down tokens in the universe - names ranked in the bottom 25% by cross-sectional momentum (XS < 25) - when they are in freefall (TS 5–35) and therefore primed to snap back. This is the mirror image of MB S, which shorts recent winners expecting mean reversion. L Fade targets recent losers expecting the same mechanic to work in reverse.

The short-term reversal effect is well-documented in crypto and stronger than in equities: extreme losers in one period tend to outperform in the next, especially when the broader market stabilises. The strategy's primary risk management is a consecutive loss gate- any symbol that loses twice in a row is blocked from re-entry, preventing capital being repeatedly deployed into “falling knives” that never bounce.

The Edge - Backtest Evidence

Before deployment, the strategy was backtested against 3 weeks of live MB score data (June 15 – July 5 2026, 784k score rows, 4h candle forward returns). The edge is real and materialises across all market regimes.

Win rate (with gate)

92.8%

610 simulated trades · $200 notional

Avg P&L per trade

+$23

After fees · 4h hold

vs current long approach

+$26/trade

Current XS 70–90 longs avg −$3.45

The XS 90+ short bucket (Bot22) already confirms the mechanic: extreme gainers mean-revert so hard that shortingthem loses money (−$2.79 avg expectancy). By symmetry, extreme losers (XS < 25) bounce. The consecutive loss gate raises win rate from 74% (no gate) to 92.8% by blocking the small set of chronic non-bouncers.

Entry Logic

Every 4 hours the rebalance loop pulls the full 150-symbol universe from the MB API, sorts it ascending by XS score (lowest first - the opposite of all other MB strategies), and filters through three gates. The five most beaten-down names that pass all gates are entered long.

Gate 1XS score - structural underperformer

XS_SCORE_MIN = 5 · XS_SCORE_MAX = 25

XS score ranks all ~150 active perpetuals by cross-sectional momentum. A score below 25 means the token is in the bottom 25% - persistent underperformance vs the universe, not a temporary dip. The floor of 5 excludes tokens with essentially zero price movement (TS = 0 flatliners and de-listed symbols).

Gate 2TS score - freefall, not dead

TS_ENTRY_MIN = 5 · TS_ENTRY_MAX = 35

The time-series score (TS) measures own-history momentum. The window TS 5–35 targets names still actively falling but not completely inert. Backtest showed a hard cliff at TS = 10: win rate drops from 75% to 59% if the floor is raised above 10. TS > 35 names have already started recovering - we are too late.

Gate 3Consecutive loss gate - no falling knives

CONSEC_LOSS_GATE_N = 2

A symbol is blocked from re-entry after 2 consecutive losing trades. The block resets when the symbol records a winning trade. In backtesting, only 6 symbols out of 23 were ever blocked - HUSDT (302 blocked attempts) was the main offender. All winning symbols (BEATUSDT, LABUSDT, BTWUSDT, etc.) were never blocked. This gate lifts win rate from 74% to 92.8%.

RegimeAll regimes - bounce works everywhere

REGIME_MAX_POS = 5 (all)

Unlike the long-momentum strategies (longs only in risk_on), L Fade is valid in all regimes. Backtest: risk_off 92.9% WR, cautious 89.4%, neutral 98.6%. Names get dumped hardest in defensive/risk_off regimes and bounce most aggressively when selling pressure eases - regardless of overall direction.

Exit Mechanisms

Three independent exits fire in order of aggression. The first one to trigger wins.

MechanismTriggerPurpose
Stop-loss−8% vs entryHard floor - exits the rare true falling knife that doesn't bounce
Max hold8 hours after entryBounces resolve within 8h; holding longer captures a different (and unvalidated) thesis
Book trailProfit ≥ 2.5%; then 1% pullbackLocks in gains on sharp bounce moves without waiting for the next 4h rebalance

Risk Parameters

Max positions

5

$500 notional each · 2× leverage

Stop loss

8%

Tight - bounce trade; if it doesn't bounce quickly, exit

Max hold

8h

Bounces resolve fast or not at all

Data-gathering phase

MB L Fade launched July 2026 and is in active data collection on Bybit Demo. The backtest covers 3 weeks of live score data - a strong signal but a short window. All five regime slots are open. Configuration will be reviewed after the first month of live operation.

How It Differs From MB L

DimensionMB LMB L Fade
ThesisMomentum - ride recent winners higherMean reversion - buy recent losers expecting a bounce
XS entry zoneXS 70–90 (top performers)XS < 25 (bottom performers)
Sort orderDescending combined score - best firstAscending XS - most beaten-down first
Regime restrictionLongs only in risk_onAll regimes - bounce works everywhere
Stop loss16% (momentum needs room)8% (bounce trade - quick or exit)
Max hold4h rebalance cycle8h hard cap

Timing & Execution

  • Rebalance fires every 4 hours. At each cycle the full 150-symbol universe is rescored.
  • Candidates are sorted ascending by XS score - the five most beaten-down names are entered first.
  • Up to 5 positions open simultaneously. $500 notional per leg, 2× leverage = $1,000 exposure per position.
  • The consecutive loss gate is checked at every rebalance. Blocked symbols are logged but not counted against the slot limit.
  • Book trail arms at +2.5% profit and trails 1% - captures fast bounce moves without waiting 4 hours.
  • No BTC direction gate, no alt-BTC spread gate, no dispersion gate - macro directional filters are irrelevant for a short-duration bounce strategy.