MultiMajorVolExpansionConvexBasketLS

Multi-Major Volatility-Expansion Convex Basket — Long-Short, Apply ONE Shared-Parameter Vol-Expansion Directional Rule Independently Across 6 Majors, Flat Per-Name in Calm Vol (Daily Bars, 3-Parameter)

A LONG-SHORT, MULTI-INSTRUMENT, pure-price CONVEX directional basket applying a single vol-expansion rule INDEPENDENTLY across 6 liquid majors (BTC, ETH, SOL, BNB, XRP, DOGE, all USD-M perps). It combines two evidence-established pieces: (1) the VOL-EXPANSION trigger (engage on a sharp expansion of realized volatility from a low base WITH a directional move — a regime shift beginning), a genuinely different mechanism from OHLCV momentum/confluence (dead, L56, 0/213) and from the return-magnitude tail-regime; and (2) the DIVERSIFICATION-FOR-CADENCE construction that the session PROVED is REQUIRED for any convex sleeve — the DOGE tail-regime died on COUNT (only 14.7 trades/yr single-name, below the ~100-trade floor), so single-name convex is structurally non-viable; only spreading one shared-parameter convex rule across many majors reaches the count. Each name's vol-expansions occur at different times, so 6 sleeves at ~40-70 engagements/yr each aggregate to ~250-400 trades (~150-250 lifetime), decisively clearing the count floor that killed the single-name convex bets, while each sleeve stays flat in calm vol (decorrelated, convex). This is NOT a cross-sectional RANK rotation (L52 — no ranking/selection; each sleeve is a standalone flat-most-of-the-time convex trigger with SHARED parameters, not per-name fit), NOT an always-on momentum ensemble (L56 — flat in calm vol), NOT a squeeze breakout (L54 — vol EXPANSION with a large concurrent move, not a compression pop), NOT a mean-reversion fade (L53), NOT a NON-PRICE-feed gate (L46 — realized vol is 100% price), NOT carry/basis/liquidation/options (L57/L58/L50/L51). Distinct from my two queued baskets/sleeves: different from the single-name BTC vol-expansion (this is multi-major, solving its count risk) and from the multi-major MAGNITUDE tail-regime basket (different trigger — vol 1st-derivative vs return-magnitude). SHARED parameters across all 6 names keep DOF low (3 params total) against the deflated-Sharpe gate that failed AAVE. Risk profile: vol-scaled sizing per engaged sleeve (equal risk), aggregate gross capped at ~60% of equity*leverage (simultaneous engagements rare since vol-expansions are somewhat asynchronous); 2x leverage cap (reads self.config.leverage).

Implements the hypothesis as specified: one shared-parameter vol-expansion rule applied independently to six majors, flat per name in calm vol, no ranking and no cross-sectional selection. Offline dry-run on the real daily catalog (2019-12 to 2026-08) at the submitted defaults (fast_win 7, exp_ratio 1.4, chandelier 6.0): 384 engagements over 6.3 years = 61/yr, avg +3.19% NET of fees per trade (21x the 0.15% fee floor), win rate 42% with the expected convex right tail, avg hold 5.9 days, per-trade Sharpe proxy 0.97, equity path x3.10 with a 19.6% max drawdown at the submitted sizing. The 384-trade count is the point of the construction and it clears the ~100-trade validation floor that killed the single-name convex sleeves (DOGE tail-regime managed ~15/yr) by roughly 4x. ROBUSTNESS: every cell of the declared _param_bounds box is positive -- fast_win 5/7/10/12 at exp_ratio 1.4 gives +2.12/+3.19/+3.39/+4.14% per trade (84/61/45/37 trades per year), exp_ratio 1.25/1.3/1.4/1.5/1.6 at fast_win 7 gives +2.45 to +3.79%, and chandelier 4/5/6/7/8 leaves per-trade capture flat (+3.13 to +3.28%) while trading drawdown against return; the weakest corner (fast_win 12, exp_ratio 1.6) is still positive at +2.03% but thins to 18 trades/yr, and the defaults sit mid-box rather than at the best cell. LEVERAGE IS GENUINELY CONSUMED, and I checked rather than assumed: simulating daily aggregate exposure at the defaults gives mean gross 0.28x equity, p95 0.68x and PEAK 1.07x -- above 1x equity notional, so the 2x margin allowance is what makes the busiest days executable, and at leverage 1.0 the 0.6 aggregate cap would clip ~10% of trading days. HONEST CAVEATS: (1) the sample is regime-dependent -- 2021 (+58%), 2024 (+41%) and 2022 (+21%) carry it while 2025 is clearly negative (-17%) and 2020/2026 are modest, so a walk-forward window landing in a 2025-like low-expansion regime will look bad; (2) the trigger is a close cousin of the return-magnitude tail-regime basket already queued (both are convex, both engage on violence and stand aside otherwise) -- the mechanisms differ (vol first-derivative vs return magnitude) but their engagements will correlate, so treat them as one family for portfolio-level allocation rather than two independent sleeves.

Outlier-driven convex basket in a confirmed-dead family, with a decisive recent-regime breakdown (convex tail/vol-expansion class). The vol-expansion signal was profitable 2021-2024 but the recent window is deeply negative: 2025 -16.3%, and rolling Sharpe collapses from ~+4 in mid-2025 to -12.8 by Jan 2026, staying negative through 2026 — exactly the segment the walk-forward OOS windows and 15-day holdout evaluate, so optimization would fit the dead early years and fail out-of-sample. The return is outlier-dependent (return_kurtosis 117, skew 5.47; a few days — 2021-02-20 +28.7%, 2024-12-04 +24.1%, 2021-04-17 +15.7% — carry it), and full-sample Sharpe is only 0.787 (CI-low 0.170) with 18.7% drawdown. This is the last untried permutation of a graveyard mechanism: the direct sibling multi-major MAGNITUDE tail-regime basket just failed full optimization (DSR 0.298, PBO 0.655, holdout Sharpe -0.03), and every convex relative before it was abandoned (BTC tail-regime fee_edge, DOGE overfit, single-name BTC vol-expansion overfit). This weaker variant will fail the same holdout. No iteration path: shared params are already minimal and the failure is a regime-dependent outlier edge, not a tunable. Abandon at BACKTEST_REVIEW rather than spend the optimization budget.

Long-short multi-instrument convex basket on six Binance USD-M majors (BTC, ETH, SOL, BNB, XRP, DOGE), DAILY bars, with SHARED parameters across all six. Each name runs an INDEPENDENT sleeve of one vol-expansion rule: sigma_fast = stdev of daily log returns over fast_win, sigma_slow = stdev over 60 days, ratio = sigma_fast/sigma_slow (the first derivative of realized vol), and move = log(C[t]/C[t-fast_win]) gives the direction the expansion is resolving in. calculate_signal returns ratio * sign(move) every bar. A sleeve ENGAGES when |signal| >= exp_ratio, taking the side of the move, and is otherwise FLAT -- in calm vol the book holds nothing. It EXITS when the directional move flips sign or price gives back chandelier_atr x ATR(14) from the best close since engagement. Sizing is equal-risk per sleeve (equity * risk_pct / chandelier stop distance), capped per leg at equity * max_notional_frac * leverage and in aggregate at equity * max_gross * leverage, all off CURRENT equity so losses cannot compound into size. Three shared tunables: fast_win, exp_ratio, chandelier_atr.

Backtest_review/analyst: weight the recent (2025-like low-expansion) regime heavily since the OOS windows and holdout sit in it and it is negative; confirm the long-leg engagements contribute across the pooled sample and that the aggregate clears the deflated-Sharpe gate on more than the 2021/2024 carry years.

Recent regime negative -- the dominant analyst/backtest_review consideration, not a code defect. The measurability fix works (384 full-sample engagements ~61/yr, 28 in the sandbox, decisively clearing the ~100-trade floor that killed the single-name convex sleeves), and full-sample economics are strong per the developer (+3.19%/trade, per-trade Sharpe proxy 0.97, equity x3.10). But the edge is regime-dependent and the recent window -- where the OOS windows and 15-day holdout sit -- is where it loses: the developer discloses 2025 is clearly negative (-17%) and the Layer-3 sandbox (trailing year) confirms it at total_return -1.43%, Sharpe -0.10, profit_factor 0.88, avg_trade_return_pct -0.59%, with a long-leg weakness (7 longs all lost, shorts carried it) and Sharpe CI [-1.66, 1.64] straddling zero. This is the ETC/SOL/premium-reversion pattern -- full-sample positive but recent-negative -- so the abandon/iterate decision belongs to the analyst.

Portfolio manager: treat this and the magnitude tail-regime basket as one convex family for allocation/correlation purposes; if both promote, size them jointly, not as two independent sleeves.

Mechanism correlation with the multi-major MAGNITUDE tail-regime basket I passed earlier this session -- a PM/allocation dedup concern, honestly disclosed by the developer. Both are convex 'engage on violence, flat otherwise' baskets on the same six majors; the triggers differ (vol first-derivative here vs return-magnitude there) but their engagements will correlate heavily, so they are effectively one family rather than two independent decorrelated sleeves. Not a code or hypothesis defect, but the portfolio manager should not allocate to both as if independent.

For live deployment, have the primary sleeve reconstruct its side from cache.positions_open() on restart, as the extra sleeves already do.

Minor restart-recovery inconsistency between the primary and extra sleeves (both unreachable in backtest, live-only). The five EXTRA sleeves correctly recover a lost side from the actual open position (pos.side.name) in on_extra_bar before calling _regime_over. The PRIMARY (BTC) sleeve's _regime_over infers a lost side from the signal sign (st.side = 1 if st.sig >= 0 else -1), which on a live mid-position restart could pick the wrong side. Harmonising the primary to read cache.positions_open() would make it consistent with the extras.

Healthy cadence: 328 trades over 6.6y, above the count floor the single-name convex bets failed

avg_trade_return_pct 3.86% clears fees; profit_factor 1.64

Low beta (0.051), decorrelated character

Recent regime deeply negative: 2025 -16.3%; rolling Sharpe collapses from ~+4 (mid-2025) to -12.8 (Jan 2026) and stays negative — precisely the window optimization OOS/holdout occupy

Outlier-driven: return_kurtosis 117, skew 5.47; days like 2021-02-20 +28.7%, 2024-12-04 +24.1%, 2021-04-17 +15.7% carry the return

sharpe_ratio 0.787 with sharpe_ci_low 0.170 — barely positive; max_drawdown 18.7% (CI to 34%)

Direct sibling (multi-major MAGNITUDE tail-regime basket) just failed full optimization: DSR 0.298, PBO 0.655, holdout -0.03 — this weaker variant will fail the same holdout

Whole convex family abandoned (BTC fee_edge, DOGE overfit, BTC vol-expansion overfit, magnitude basket overfit)

MultiMajorVolExpansionConvexBasketLS combined the vol-expansion trigger with the diversification-for-cadence construction, spreading one shared rule across six majors to clear the trade-count floor that had killed the single-name convex bets. It did reach a healthy 328 trades with a decorrelated +182.5% and PF 1.64, but the return was outlier-driven (kurtosis 117) and the recent regime had broken down hard — 2025 -16.3%, rolling Sharpe to -12.8 into 2026 — precisely the segment the OOS and holdout score. The analyst abandoned it at backtest review as the last untried permutation of a graveyard convex family whose direct sibling had just failed optimization on every robustness gate; it never reached optimization, analysis, or risk review.

Solving the cadence problem via diversification does not rescue a convex family that is outlier-dependent and regime-decayed — a Sharpe of 0.79 carried by a few extreme days, collapsing to deeply negative in the recent OOS window, will not survive the holdout, and successive permutations of an already-exhausted mechanism inherit the same fate.

The analyst abandoned it at backtest review as an outlier-driven convex basket in a confirmed-dead family with a decisive recent-regime breakdown: the vol-expansion edge worked 2021-2024 but is deeply negative in exactly the window the OOS/holdout evaluate, so optimization would fit dead early years and fail out of sample. Its direct sibling (the multi-major magnitude tail-regime basket) had just failed full optimization (DSR 0.298, PBO 0.655, holdout -0.03), and this weaker variant would fail the same holdout — the last untried permutation of a graveyard mechanism with no iteration path.

A long-short, multi-instrument convex directional basket (3 shared parameters) that ran one volatility-expansion rule independently across six majors (BTC, ETH, SOL, BNB, XRP, DOGE) — each sleeve flat in calm vol, engaging only when its own realized volatility expanded sharply from its base with a directional move, riding it with a chandelier stop — combining the vol-expansion trigger with the diversification-for-cadence construction proven necessary for convex sleeves.

The backtest (6-major daily basket, 2409 data days) cleared the cadence floor (328 trades) and looked reasonable: +182.5% with Sharpe 0.787 (CI-low 0.170), profit factor 1.64, avg_trade_return_pct 3.86%, decorrelated (beta 0.051). But it was outlier-driven (kurtosis 117, skew 5.47, a few days carry the return), max drawdown 18.7% (CI to 34%), and the recent regime was deeply negative — 2025 -16.3% with rolling Sharpe collapsing from ~+4 mid-2025 to -12.8 by January 2026 and staying negative.