SolSpotFourHourNegativeSkewnessMeanReversionLong

SOL Spot 4H Negative Realized-Skewness Mean-Reversion Long with Trend Filter

A long-only, single-instrument BINANCE_SPOT strategy on SOLUSDT 4H bars using a higher-moment statistical signal: realized skewness of log returns over the last 20 bars. When the third moment of returns becomes strongly negative (large drawdown bars dominate the recent window) AND the broader trend remains intact (close > sma_100), enter long anticipating mean-reversion of the skewness toward neutral as volatility normalizes. This is a fundamentally different mechanism class from everything currently in the pipeline: (a) classical first-moment indicators (MACD, OBV, SAR, RSI, Heikin-Ashi) all use price-level or smoothed-price information — empirically refuted on ADA 4H; (b) volume-z and bar-range mechanisms (ADAVolumeSpikeMeanReversionLong, ETHSpotNR4BreakoutLong, ADAHammerReclaimAbsorptionLong) use volume or range geometry; (c) Donchian/breakout and trend-continuation mechanisms (most pipeline entries) use first-moment momentum. Realized skewness is the third standardized moment — orthogonal to all of these. Spot venue chosen because BINANCE_SPOT is critically under-quota at 6.3% vs 15% target. SOL spot specifically because no SOL spot strategies exist in the pipeline (current spot entries are BTC and ETH only), adding within-venue instrument diversity. Long-only-only because (i) the strategy is on a CASH-account spot instrument; (ii) the skewness-mean-reversion mechanism has the strongest empirical support on the bullish side per the crypto leverage-effect literature; (iii) short-only directional on crypto majors has been empirically refuted.

Implements a genuinely orthogonal third-moment mechanism: realized skewness standardizes returns by their own mean/std before cubing, so it is scale-free and carries no first-moment (price/momentum) or second-moment (volatility/volume) information that the existing MACD/RSI/Donchian/volume-z/range strategies already exploit. The negative-skew entry plus 100-SMA uptrend filter targets the bullish-side skewness unwind supported by the crypto leverage-effect literature. calculate_signal returns the skewness in natural units (may exceed 1) so the signal is continuous and varies every bar, satisfying Layer-2 frozen-signal detection; should_enter thresholds it at <= -skew_entry AND the cached uptrend flag, and should_exit uses skew>=skew_exit as the mean-reversion-confirmation exit plus PnL and time stops. All moments are computed causally from buffered closes with zero/NaN/std guards (no lookahead). The strategy is correctly routed to BINANCE_SPOT per the routing rule (long-only, no leverage, edge is not funding-based) and fills the under-quota spot venue plus adds SOL spot instrument diversity; should_enter only ever returns BUY (never sells to enter, respecting the CASH constraint) and the single position is sized once from full equity so spot equity-depletion is a non-issue. leverage stays 1.0 and is deliberately not referenced in sizing (spot caps at 1x), avoiding the leverage_set_but_unused gate. SOLUSDT spot is deeply liquid with multi-year 4H history, avoiding data_unavailable; the 6%/8% exits and multi-bar holds give per-trade moves above the ~0.20% spot round-trip fee.

Correctly-implemented negative-realized-skewness mean-reversion with NO edge on SOL spot 4H: profit_factor 0.84, expectancy -$218/trade, Sharpe -1.24, total return -29% over 5.7 years, max_drawdown 46.5%, negative in 6 of 7 years (171-trade sample, so this is a fair read, not noise). Buying after large-drawdown (negative-skew) bars even in an uptrend does not revert profitably — win_rate 46.2% with avg_loss >= avg_win, and fees consume 7.88% of gross on top of the negative edge. The failure is in the signal, not the parameters: there is no positive base or robust region for optimization to build on, and tuning to 171 trades would overfit the lone 2026 positive year and fail DSR/holdout. Not worth 2 hours of optimization. NOTE for research: the third-moment (realized-skewness) mean-reversion premise does not hold on SOL spot 4H — downside-skew clusters do not reliably unwind on the long side intra-trend. A different instrument/timeframe would be a separate hypothesis, not an iteration of this one; the orthogonal-mechanism appeal does not survive contact with the data.

Long-only higher-moment mean-reversion on SOLUSDT.BINANCE_SPOT 4H bars. It computes the realized skewness (third standardized moment) of log returns over the last 20 bars and enters LONG when skewness turns strongly negative (<= -0.5, downside bars dominating) AND the close is above the 100-bar SMA (uptrend intact), anticipating the skewness reverts toward neutral. It exits when skewness normalizes (>= 0.0), on a 6% take-profit or 8% stop-loss, or after an 18-bar time stop. Spot CASH account (long-only, leverage 1), ~50% equity notional, OHLCV-only.

Mechanism correctly implemented (realized third-moment skewness signal, uptrend-filtered); genuinely orthogonal mechanism class; clean diagnostics; adequate 171-trade sample.

No edge — consistently losing: profit_factor 0.84, expectancy -$218/trade, Sharpe -1.24, total return -29% over 5.7 years, max_drawdown 46.5%.

Negative in 6 of 7 years (only 2026 marginally positive) — structural, not a single-regime artifact; rolling Sharpe deeply negative across most of the history.

win_rate 46.2% with avg_loss (~$2,478) >= avg_win (~$2,413) — negative payoff geometry.

commission_pct_of_gross 7.88% — heavy fee drag on a negative-expectancy signal.

SolSpotFourHourNegativeSkewnessMeanReversionLong introduced a deliberately orthogonal higher-moment signal — buying negative realized-skewness extremes in an uptrend on under-quota SOL spot — to escape the first-moment indicators that had been refuted elsewhere. The mechanism was cleanly implemented across 171 trades, but it lost decisively (profit factor 0.84, Sharpe -1.24, -29% over 5.7 years, negative in 6 of 7 years) with losers outsizing winners and fees eating ~8% of gross. The analyst abandoned it at backtest review after one iteration, concluding the failure was in the signal rather than the parameters. The takeaway is that the skewness-mean-reversion premise doesn't survive contact with the data on SOL spot 4H, so a different instrument or timeframe would be a separate hypothesis.

A genuinely novel, orthogonal mechanism class (realized skewness) is no guarantee of edge — the third-moment mean-reversion premise simply doesn't hold on SOL spot 4H, and a negative-expectancy signal with adverse payoff geometry and heavy fee drag can't be salvaged by tuning.

The analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate: the orthogonal third-moment mechanism was correctly implemented over a fair sample, so the loss was a clean read of a premise that doesn't hold — downside-skew clusters don't reliably unwind on the long side intra-trend — leaving no positive base for optimization.

A long-only SOLUSDT spot 4H strategy using a higher-moment statistical signal — the realized skewness of the last 20 bars' log returns — that bought when skewness turned strongly negative (≤ -0.5, large drawdown bars dominating) while price stayed above its 100-bar SMA, anticipating skewness mean-reverting toward neutral, and exited on skewness normalization (≥ 0), a 6% take-profit, 8% stop, or an 18-bar time-stop.

Over ~5.7 years and 171 trades it consistently lost: profit factor 0.84, expectancy -$218/trade, Sharpe -1.24, -29% total return, and a 46.5% max drawdown. It was negative in 6 of 7 years (only 2026 marginally positive), with a 46.2% win rate and adverse payoff geometry (average loss ≥ average win), while commissions consumed 7.88% of gross.