XrpRealizedSkewnessReversalLS

XRP Realized-Skewness Reversal, Long-Short (Single-Instrument XRPUSDT.BINANCE USD-M, Daily Bars — Short When Recent Returns Are Strongly Positively-Skewed / Lottery-Priced, Long When Strongly Negatively-Skewed / Crash-Compensated, Deadband-Gated, 3-Parameter)

A LONG-SHORT, single-instrument, pure-OHLCV directional strategy on XRPUSDT.BINANCE USD-M perpetual (DAILY bars) whose signal is the THIRD MOMENT of the recent return distribution: rolling REALIZED SKEWNESS of daily log-returns. Strongly positively-skewed recent returns (lottery-like up-jumps) predict underperformance; strongly negatively-skewed (violent flushes) predict crash-compensated outperformance. XRP is chosen for the most jump/lottery-dominated distribution of the deep-history majors. Slow daily horizon keeps turnover low so per-trade capture sits well above the ~0.10% USD-M round-trip.

Implements the hypothesis literally: the decision variable IS the third moment of the recent return distribution, computed incrementally from a bounded deque (O(lookback) per bar, no history rescan) and returned every bar so the signal is continuous rather than gated. Reversal polarity is exactly as specified (positive skew -> short, negative skew -> long) with the deadband as the entry/exit gate. Venue is the USD-M futures MARGIN account because the strategy goes short, which spot cannot do; leverage stays 1.0 because the edge does not need amplified notional, and sizing still reads self.config.leverage so it stays consistent if tuned. Defaults were chosen for signal frequency and cost viability rather than peak in-sample fit: on XRP daily history a 45-day skew window with a 0.75 entry band fires roughly 8-10 round trips per year (holds of ~2-6 weeks), so the per-trade capture is measured in whole percent against the ~0.10% USD-M round trip. Note for the record: the shorter 20-30 day skew windows are much noisier and their sign of edge is unstable, so if the optimizer pushes the lookback down expect the mechanism to degrade — the declared _param_bounds mirror the code's clamp of [15, 90].

analyzing_stuck_3_timeouts

Long-short single-instrument reversal on XRPUSDT.BINANCE USD-M perpetual, DAILY bars. Every bar it computes the rolling realized SKEWNESS (bias-corrected third standardised moment) of the last 45 daily log-returns and returns it as a continuous signal in natural units. Strongly positive skew (lottery-like up-jumps, over-priced right tail) triggers a SHORT; strongly negative skew (violent flushes, crash-risk compensation) triggers a LONG. A deadband gates entries at |skew| >= 0.75 and exits when skew mean-reverts inside |skew| <= 0.25 or flips to the opposite extreme. A 12% close-based protective stop and a 40-calendar-day time stop bound the tail. Sizing is capital-relative: 50% of account equity of notional per trade (leverage 1.0), quantised to the instrument's size precision.

No change required if the base template dispatches to the subclass min_bars_required(); the value returned is appropriate for the 45-day skew window.

min_bars_required() overrides a framework method (base_shadow static-analysis warning). The override is deliberate and correct (returns skew_lookback + 2 = 47 bars for warm-up), but confirm the base template honors the override for its warm-up gate rather than using its own default.

Faithful implementation but edge is regime-dependent: sandbox shows the calm-vol tercile is negative (-7.9%, Sharpe -0.16) while all the P&L comes from the normal/stressed terciles, and the Sharpe CI [-0.16, 1.34] straddles zero. Return distribution is heavy-tailed (return_kurtosis 55, return_skew 2.8), so the headline is right-tail-dependent. Worth checking on full history whether the skewness-reversal sign is stable out-of-sample rather than a few large flush/rebound windows; short leg (46 trades, 45.7% win) is weaker than long leg.

Numerically viable on a reliable, adequate sample: 72 trades over 6.7 years (consistent with the slow, low-turnover daily-skewness thesis), PF 1.54, avg_trade_return_pct 5.56% (far above the 0.15% USD-M fee floor; commission only 0.91% of gross, so this is a real gross edge, not fee-fragility).

Genuine directional edge with low market coupling: positive alpha 0.31, beta -0.05, benchmark_correlation -0.12 — the return is not just XRP beta. Sharpe 0.655, Sortino 0.75, PSR 0.934, recovery_factor 7.84.

Trades implement the stated mechanism: balanced long/short book (26 long / 46 short), fading positive-skew and buying negative-skew; entry_diag shows 72 signaled = 72 submitted with zero drops, so the signal fires as designed and this is a fair test.

Profitable across most calendar years (2021 +68%, 2022 +34%, 2023 +31%, 2025 +40%) and positive in both normal (Sharpe 1.20) and stressed (Sharpe 0.88) vol terciles.

High max_drawdown 46.75% (DD CI up to 82%) with a 450-day max drawdown duration — the risk-review drawdown floor is the real hurdle, and optimization must tighten this via stop/exit tuning.

Regime-dependent: calm-vol tercile is negative (-9.87%, Sharpe -0.22) — QA's concern; all P&L comes from normal/stressed regimes, and the equity is right-tail-dependent (return_kurtosis 56, return_skew 2.77).

Sharpe CI low is marginally negative (-0.06); the short leg (46 trades, 47.8% win) is weaker than the long leg (57.7% win) — walk-forward/holdout will be the decisive test of sign stability.

XrpRealizedSkewnessReversalLS-8e0ff3c3ab

XrpRealizedSkewnessReversalLS bet that the third moment of XRP's recent daily returns predicts reversals — fading positively-skewed lottery moves and buying negatively-skewed flushes. Its initial backtest passed the pre-optimization gate (Sharpe 0.655, PF 1.54, 5.56% per-trade return, verdict 'optimize'), despite a heavy 46.75% drawdown and a losing calm-vol regime. Optimization then exposed the edge as fragile: walk-forward OOS Sharpe averaged -0.27, the holdout came in at -0.47 and failed, PBO reached 0.70, and the deflated Sharpe was essentially zero, with the vault too thin (~14 trades) to prove anything. The pipeline never reached a clean analyst decision — the run stalled and was abandoned after the analyst timed out three times during the ANALYZING stage.

A slow daily-skewness edge can look strong in-sample yet be structurally untestable — the OOS vault held only ~14 expected trades, so there was no history slice both short enough to withhold and long enough to validate, and the strategy failed every out-of-sample check.

The run was abandoned in the ANALYZING stage after the analyst agent failed to return a verdict across three timeouts (abandon_reason: analyzing_stuck_3_timeouts); no analyst or risk-officer verdict was recorded, though the underlying optimization signals (failed holdout, negative OOS Sharpe, high PBO, candidate not surviving programme FDR) were already weak.

A long-short daily-bar strategy on XRPUSDT.BINANCE USD-M that used rolling realized skewness of daily log-returns as a reversal signal — shorting lottery-like positively-skewed windows and buying crash-compensated negatively-skewed windows, gated by a deadband.

The initial backtest over 6.7 years (72 trades) showed Sharpe 0.655, profit factor 1.54, total return 275.5%, and avg_trade_return_pct 5.56% well above the fee floor, but with a 46.75% max drawdown and a negative calm-regime tercile. After optimization the out-of-sample evidence collapsed: walk-forward average OOS Sharpe was -0.27, the one-shot holdout Sharpe was -0.47 (11 trades, holdout failed), PBO was 0.70, and the deflated Sharpe was ~0.02.