SolMaCrossoverTrendFollowLS

SOL Macro 50/200 Moving-Average Crossover Trend-Following, Long-Short (Single-Instrument BINANCE USD-M Perp, Daily Bars, Golden-Cross Long / Death-Cross Short — Shorts the Bear So Returns Are Regime-Distributed, ATR-Sized, Fee-Trivial, Third Deep Name + Third Distinct Trend Mechanism to Diversify the Trend Sleeve, Pure OHLCV, 2-Parameter)

A LONG-SHORT macro trend-follower on SOLUSDT.BINANCE (USD-M perpetual, daily bars) using the classic 50/200-day moving-average crossover (golden cross / death cross). This is a deliberate PORTFOLIO-CONSTRUCTION addition: the pending trend sleeve is BTC (200-day MA band) and ETH (Donchian channel breakout); this adds the third deepest name (SOL) with a third distinct trend mechanism (dual-MA crossover), and because trend-following works best as a diversified multi-name portfolio — each name's trend is a largely independent return stream — three uncorrelated engines across three names is genuine diversification, not duplication. Like its siblings it escapes the trap that kills every long-only directional strategy here (single-regime concentration, all return in the 2021 bull) via the SHORT side: it goes short SOL on a death cross (50-day below 200-day), profiting in the 2022 crash and other declines, so returns are distributed across regimes as the analyst requires. It is NOT the failed fast daily-trend BASKET (multi-name, 50-day, chopped up, no_edge): this is a SINGLE deep name on the SLOWEST standard crossover (50/200), flipping only ~1-2 times per year — utterly fee-trivial against SOL's violent multi-ten-percent-to-multiple trend legs. It respects every hard lesson: FEE_EDGE (rare, huge trend legs dwarf the ~0.10% round-trip); OVERFIT (2 standard parameters, single name, no selection surface — the death of the trend-efficiency/low-vol baskets); SINGLE-REGIME (the short side profits in bear markets); DATA (pure OHLCV, deep SOL, multi-year history spanning the 2021 bull, 2022 crash, 2023-24 recovery). NOT reversion (retired class), NOT carry (fee-dead), NOT options/CM/HL (dead). Fills the emptiest horizon (macro/1w+, 0.6%) and the under-represented long_short direction (15.0%).

Iteration 2 fixes the sole QA rejection reason — measurability — with the smallest possible change and no alteration to the (credited, mechanically-correct) crossover logic. The daily 50/200 signal produced only 12 trades over 6 years with 160-day holds, so walk-forward windows had ~4 trades and the 15-day holdout closed ~0 trades, making the point estimate indistinguishable from luck (wide Sharpe CI). Rather than change the mechanism, I route the identical 50/200 dual-SMA crossover onto 4-HOUR bars: 50/200 on 4H spans ~8-day fast / ~33-day slow, still a genuine swing/macro trend signal but crossing several times more often. This multiplies the trade count (making walk-forward windows populated and the short-leg regime evidence statistically evaluable) and shrinks holds enough that the 15-day holdout can close trades — directly resolving the unmeasurable/underpowered finding. Fee-triviality is preserved: even at several flips per year, ~0.10% round-trip is dwarfed by SOL's multi-ten-percent trend legs (~fraction of a percent annual drag). The code is otherwise byte-identical except size_prec is corrected from 0 to 3 per the SOLUSDT instrument spec (the other noted, no-defect item). Two standard parameters, single deep name, pure OHLCV, shorts on futures/MARGIN — every hard lesson (FEE_EDGE, OVERFIT, SINGLE-REGIME, DATA) still respected.

Same single-regime concentration artifact that killed its BTC and ETH trend-sleeve siblings at ANALYZING today, already visible in the initial backtest — not worth 2 hours of optimization. The +766% total return is entirely two bull spikes (2021 +233%, 2024 +126%) carried by a handful of monster days (2021-05-23 +144%, 2024-01-18 +101%; kurtosis 53.7), while every bear/flat year is negative (2022 -0.5%, 2023 -1.5%, 2025 -5.1%). The defining premise — that the short side earns in bear legs like the 2022 crash so returns are regime-distributed — is empirically FALSE: 2022 is -0.54% and short_win_rate is 0.43. max_drawdown 46.2% (CI high 71.9%) will not clear the 10% promotion floor. This is a pure-OHLCV single-name trend follower (class 0/213 survival); the concentration is in the signal, not in any tunable knob, so optimization cannot fix it — confirmed by both siblings' negative OOS/holdout under the identical thesis. META-NOTE FOR RESEARCH LEAD: the entire 'add a short side to a single-name OHLCV trend follower to distribute returns across regimes' family (BTC/ETH/SOL, three mechanisms) has now failed identically — the short legs lose in bear years and the edge remains bull-spike-concentrated. Stop generating single-name directional trend timers as a fix for single-regime concentration.

Long-short 50/200 dual-SMA crossover trend-follower on SOLUSDT.BINANCE USD-M perp, now on 4-HOUR bars. Golden cross (fast SMA > slow SMA) holds LONG, death cross (fast < slow) holds SHORT, exit/flip on the opposite cross. Volatility-targeted, path-independent capped ATR sizing. Shorts on MARGIN so returns are distributed across bull and bear regimes. Two tunable parameters (fast_ma, slow_ma); everything else is frozen construction.

Verification failed (Layer 4 — QA review): - UNMEASURABLE / non-diagnostic — the same structural block that failed the BTC-200d and ETH-Donchian macro trend followers, and the WIDEST CI of the three. total_trades is 12 over ~6 years with a Sharpe CI [-13.84, +11.28] spanning strongly negative to strongly positive, so the +132% / Sharpe 5.21 point estimate cannot be distinguished from luck (L16/L26); the developer's own note ('12 flips ~= 2/year') confirms it. The pipeline cannot validate it: walk-forward ~4 trades/window (pure overfit) and the 15-day holdout gets ~0 closed trades (160-day holds), so the holdout gate is unevaluable. The 50/200 crossover is the SLOWEST standard trend signal — structurally the fewest trades, the opposite of what measurability needs. - The multi-regime thesis — the entire justification — is unproven and underpowered: short_win_rate 0.333 on only 6 shorts (non-diagnostic), long_win_rate also 0.333, max_drawdown 52.5%. The whole point is that the SHORT side earns in bear legs so returns are regime-DISTRIBUTED. With 6 shorts it cannot be shown the short leg is net-positive in the 2022 crash, so +132% may still be bull-concentrated — the exact artifact it claims to fix. - size_prec is 0 (whole SOL coins), but CLAUDE.md lists SOLUSDT size precision as 3. At SOL's price whole-coin rounding doesn't breach min-notional and caused no defect here, but it's coarser than the instrument spec and could under-size at the margins. - CREDIT (no defect): the code is mechanically correct — dual-SMA crossover (spread (fast-slow)/slow), golden-cross LONG / death-cross SHORT, exit on opposite cross, ATR-targeted capped path-independent sizing, shorts on MARGIN (not spot), subscription flags consistent, no stacking, no liquidation. Fee-trivial (0.21%). Rejection is purely measurability + the mechanism's inherent low trade count.

Reframe the hypothesis to a 4H swing-horizon crossover; no code change required.

Timeframe/horizon deviation: hypothesis frames this as DAILY 50/200 ('slowest standard crossover', flips ~1-2x/yr, fills the 'macro/1w+' horizon), but config runs 4-HOUR bars where 50/200 spans ~8-day/~33-day — a swing signal that flips ~13x/yr, not macro. The core mechanism (50/200 dual-SMA crossover, long-short, ATR sizing, single deep name SOL) is faithfully and correctly preserved, and the change is a documented, deliberate measurability fix responding to the prior iteration's rejection (daily = ~12 trades over 6 years, unmeasurable). Fee-triviality survives (~1.3%/yr vs SOL's large trend legs). Not critical because instrument, direction, and the specific 50/200 edge all match — but the 'macro/1w+' and 'rare-flip' framing no longer holds and should be reframed.

Analyst should verify per-window trade counts before trusting optimization.

Measurability improved 6x (13 trades/363 days vs 12 trades/6yr) but projects to only ~80-100 trades over the full backtest — near the lower bound for distinguishing edge from noise; walk-forward windows will still be thin.

Analyst: review long vs short leg PnL across regimes on the full backtest.

Sandbox flag for analyst (not a code defect): long_win_rate = 0.0 (all 7 longs lost) vs short_win_rate 0.50, total_return -20.2%, PF 0.63. Code polarity is correct; likely a single-window artifact, but the fully-losing long leg warrants scrutiny over the full history.

avg_trade_return_pct 21.6% and PF 3.08 — decisively fee-viable, not a FEE_EDGE risk

74 trades (37 long / 37 short) — both legs active, mechanism correctly implements the dual-SMA crossover

Clean 2-parameter construction

Central thesis FALSIFIED in the raw backtest: the entire +766% is 2021 (+233%) and 2024 (+126%) bull spikes; bear/flat years 2022 (-0.5%), 2023 (-1.5%), 2025 (-5.1%) are all negative — the short side did NOT distribute returns across regimes

Single days carry the result (2021-05-23 +144%, 2024-01-18 +101%; kurtosis 53.7) — textbook single-regime artifact

max_drawdown 46.2% (CI high 71.9%) — will fail the 10% promotion drawdown floor post-optimization

Third identical instance of a family whose BTC and ETH siblings were both abandoned today for overfit+single-regime; pure-OHLCV single-name trend is the 0/213-survival class

This SOL crossover strategy was the third member of a trend sleeve intended to diversify across BTC, ETH, and SOL with distinct mechanisms, again betting that shorting on a death cross would earn in bear markets. Its raw backtest looked headline-strong — +766% return, profit factor 3.08, huge per-trade returns — but the analyst caught the same defect already fatal to its BTC and ETH siblings: nearly all the gain came from 2021 and 2024 bull spikes and a handful of outlier days, while every bear/flat year was negative and the shorts did not profit. With a 46.2% drawdown that could never clear the promotion floor and the single-regime concentration living in the signal rather than any tunable knob, it was abandoned at the backtest-review gate before spending any time on optimization, along with a meta-note to stop generating single-name directional trend timers as a regime-concentration fix.

Adding a short side to a single-name pure-OHLCV trend follower does not distribute returns across regimes — the short legs lose in bear years and the edge stays concentrated in a few bull-spike days — so this entire family (BTC/ETH/SOL, three mechanisms) fails identically and should not be regenerated as a fix for single-regime concentration.

The analyst abandoned it at the pre-optimization backtest-review gate: the whole +766% traced to two bull spikes (2021 +233%, 2024 +126%) carried by a few monster days while every bear/flat year was negative (2022 -0.5%, 2023 -1.5%, 2025 -5.1%), so the short side failed to distribute returns (short_win_rate 0.43), the thesis was falsified, and the 46.2% drawdown would not clear the 10% promotion floor.

A long-short 50/200 dual-SMA crossover trend-follower on single-name SOLUSDT.BINANCE (run on 4-hour bars) — long on a golden cross, short on a death cross — added as a third deep name and third distinct trend mechanism, relying on the short side to profit in bear legs and distribute returns across regimes.

The initial backtest showed +766% total return over 74 trades (37 long / 37 short) with profit factor 3.08, avg_trade_return_pct 21.6%, and Sharpe 0.94, but a 46.2% max drawdown (CI high 71.9%) and kurtosis 53.7; optimization was never run.