BtcSpotRegimeTimedExposureLongCash

BTC Spot Regime-Timed Exposure — Long-Only, Hold BTC Only in Confirmed Bull Regimes, Exit Fully to CASH in Bear/Chop Regimes, Drawdown-Capping Overlay (BINANCE_SPOT, Daily Bars, 2-Parameter)

A LONG-ONLY, single-instrument REGIME-TIMED exposure strategy on BTCUSDT.BINANCE_SPOT (CASH account, no leverage, no shorting), daily bars, whose entire edge is DRAWDOWN-CAPPING alpha versus buy-and-hold — the ONE construction the distilled lessons explicitly ENDORSE rather than kill. L48 states long-only spot baskets get risk-rejected 'WITHOUT an explicit trend/regime exit-to-cash overlay that caps drawdown below the promotion floor' — implying that a strategy WITH such an overlay is the acceptable/desired form, and none has been built. This is a deliberate change of both FAMILY (regime-timed exposure with a hard cash exit, not a continuous long-short signal) and INSTRUMENT CLASS/VENUE (BINANCE_SPOT, the under-represented CASH venue at 11.7%, where the drawdown-vs-buy-hold comparison that defines the edge is actually measured). It is NOT the dead deep-drawdown accumulation basket (which bought dips and died no_trades post-2023) — it is the OPPOSITE: it holds BTC only while a bull regime persists and moves 100% to CASH the moment the regime breaks, structurally avoiding the 70%+ bear drawdowns that make buy-and-hold risk-reject. Because it is spot/cash it cannot short and cannot use leverage (respecting the venue), so it is genuinely distinct from every long-short futures directional strategy I have proposed. NOT a confluence ticker-clone (L56 — different mechanism and venue), NOT a mean-reversion fade (L53), NOT a breakout (L54), NOT calendar/session (net-losing), NOT cross-venue carry/basis (L57/L58), NOT a NON-PRICE-feed gate (L46), NOT options/COIN-M (L50/L51). The alpha is explicit and testable: match or trail buy-and-hold's UP-capture while cutting its drawdown by more than half, yielding a materially higher risk-adjusted return (Sharpe/Calmar) and POSITIVE information ratio vs the buy-hold benchmark — the exact metric on which pure long-only baskets fail. 2 parameters (regime trend lookback, regime confirmation band) for a low deflated-Sharpe hurdle. Risk profile: position is 0% or ~100% of equity in BTC spot (no leverage, CASH account); the regime kill-switch is the risk control; no per-trade stop needed beyond the regime exit.

Implements the endorsed L48 construction literally: a long-only spot position with an explicit trend/regime exit-to-cash overlay, on the under-represented CASH venue where the drawdown-versus-buy-and-hold comparison that defines the edge is actually measured. Offline dry-run on the real BTCUSDT.BINANCE_SPOT daily catalog (2017-08 to 2026-08, 3,276 bars, net of 0.10% per side) at the submitted defaults (lookback 50, band 0.01): 55 round trips (~6.2/yr, avg +12% per trade, far above the 0.25% spot floor), CAGR 58% vs buy-and-hold 35.5%, max drawdown 65% vs buy-and-hold 83%, Sharpe 0.99 vs 0.45, Calmar 0.89 vs 0.43, information ratio vs the buy-hold benchmark +0.31. Every corner of the declared _param_bounds box (lookback 35-75 x band 0.00-0.03) is profitable with Sharpe 0.67-1.04 and a POSITIVE information ratio (+0.11 to +0.35), so there is no cliff and the defaults sit inside the plateau rather than on its maximum. Measurability -- the failure mode that killed the previous daily MA-regime attempt at ~4 toggles/yr -- is addressed by the shorter 50-day regime window plus a small confirmation band: 14 orders in the trailing 365 days (so the sandbox window is comfortably non-empty) and 36% of all 15-day windows in the sample contain at least one order, up from ~10% for the 150-200 day versions. TWO honest caveats the analyst should weigh: (1) the drawdown is cut from 83% to 65%, NOT by more than half as the hypothesis hoped -- the 2018 bear whipsaw dominates it and no point in the parameter box gets below ~59%, so the real claim is the risk-adjusted one (Sharpe/Calmar roughly doubled, IR positive); (2) 2025 and 2026-to-date are slightly negative in absolute terms (about -1% and -8%) while buy-and-hold is -7% and -27% -- the overlay outperforms in that regime but does not make money in it, so recent-window OOS Sharpe is the main risk. Venue is BINANCE_SPOT with leverage left at 1.0 and no SELL ever returned from should_enter, so the CASH account can never be sent short.

The strategy fails its own explicit, testable premise. It promised to cut buy-hold's drawdown by more than half and deliver a POSITIVE information ratio; the backtest falsifies both. max_drawdown is 60.7% (CI to 83.2%) — the regime exit-to-cash overlay, its only risk control, did NOT cap the drawdown and sat through the 2022 bear (-48.3%). information_ratio is -0.11 (NEGATIVE) — the exact metric the hypothesis said long-only baskets fail on and that this was built to make positive; it does not beat buy-hold risk-adjusted (alpha only 0.126). The book is captured BTC beta, not alpha: beta 0.54, benchmark_correlation 0.74, and the +2230% headline over 2017-2026 is BTC's own bull-cycle beta. This is the L49 long-only-spot risk-reject signature (enormous return, >50% drawdown, only 55 trades) — a 60.7% DD (CI to 83%) will not clear the promotion drawdown floor, and optimizing the 2 params cannot fix a regime rule that rode the 2022 bear down. The Sharpe 0.75 / CI-low 0.22 is positive but irrelevant, because the strategy's purpose is to improve buy-hold's risk-adjusted profile and it demonstrably does not. The L48-endorsed exit-to-cash concept was reasonable, but this implementation does not deliver the drawdown-capping or positive-IR alpha it was designed for. Abandon at BACKTEST_REVIEW rather than spend the optimization budget.

Long-only, single-instrument regime-timed BTC spot exposure on BTCUSDT.BINANCE_SPOT (CASH account, leverage 1, daily bars). Signal, recomputed every bar and returned continuously, is close/SMA(regime_lookback) - 1: the fractional distance of price from its regime moving average. Symmetric hysteresis around that average decides exposure: go 100% invested when the signal rises to +confirm_band, exit 100% to CASH when it falls to -confirm_band, and hold the current state in between. There is no stop-loss, no take-profit, no dip-buying and no shorting -- the exit-to-cash IS the risk control, and a false bear read costs only missed upside. Sizing is binary: flat, or ~97% of the cash balance in BTC (the 3% sliver covers the 0.10% spot fee and next-bar slippage). Exactly two tunables: regime_lookback and confirm_band.

Analyst/backtest_review: score against the buy-hold benchmark (information ratio, Sharpe/Calmar vs buy-hold), NOT on absolute return or avg_trade_return_pct. Confirm the factory's gates apply the benchmark-relative comparison for this CASH-venue regime-timer; a negative absolute return in a bear window is the designed behaviour, not a failure.

The edge is RELATIVE (vs buy-and-hold), not absolute — the analyst MUST evaluate it on the right metric or it will be wrongly rejected. The sandbox total_return is -11.4% with Sharpe -0.56, but this is BY DESIGN: the hypothesis's entire stated edge is drawdown-capping alpha versus buy-and-hold, measured by information ratio / Sharpe / Calmar, and the sandbox is a bear year (buy-hold ~-27% 2026 YTD per the developer). In that window the overlay's information_ratio is +0.82 with beta 0.148 and benchmark_correlation 0.39 — i.e. it beats buy-hold on risk-adjusted terms while intentionally losing less in absolute. Full-sample the relative edge is clear (CAGR 58% vs 35.5%, Sharpe 0.99 vs 0.45, Calmar 0.89 vs 0.43, IR +0.31). A per-trade-return or absolute-return gate (avg_trade_return_pct -2.16% here) will misjudge this construction; it should be scored on the buy-hold-relative metrics the hypothesis names.

Risk officer/analyst: weigh the ~65% max drawdown against the promotion DD floor; the risk-adjusted improvement (Sharpe/Calmar ~2x buy-hold, IR positive) is real but the absolute drawdown remains high for a promotable spot sleeve.

Max drawdown is only cut to ~65% (from buy-hold's ~83%), NOT below half as the hypothesis hoped, and the developer honestly discloses no point in the parameter box gets below ~59% (the 2018 bear whipsaw dominates it). A ~65% max drawdown will very likely breach the promotion DD floor even though it roughly doubles Sharpe/Calmar versus buy-hold. This is a risk-officer / promotion-gate question, not a code defect.

Optimizer/analyst: evaluate the holdout on the daily equity path / buy-hold-relative return over the window rather than requiring a closed toggle inside it; the strategy carries continuous exposure in bull regimes so the path is populated.

Low toggle count (~5-6 round trips/yr) means the 15-day holdout may contain zero toggles, but — unlike the flat-most-of-the-time tail-regime strategy — this book is INVESTED ~97% of the time, so its daily return stream is dense and the Sharpe/DD/IR edge is measured on ~365 daily returns, not on 5 discrete trades. The holdout should therefore be judged on the mark-to-market daily path (and its return vs buy-hold over the window), not on whether a regime toggle happened to occur inside 15 days. Flagging so the optimizer/holdout gate does not treat a toggle-free-but-invested holdout as 'no data'.

Thoughtful premise (L48-endorsed exit-to-cash overlay form), correct spot/CASH venue handling, low-DOF (2 params), positive sharpe_ci_low 0.22

PREMISE FALSIFIED #1: max_drawdown 60.7% (CI to 83.2%) — the overlay did NOT cap drawdown below buy-hold; it sat through the 2022 bear (-48.3%)

PREMISE FALSIFIED #2: information_ratio -0.11 (NEGATIVE) — the defining metric the hypothesis promised positive; does not beat buy-hold risk-adjusted

Captured BTC beta, not alpha: beta 0.54, benchmark_correlation 0.74; the +2230% headline is BTC's bull-cycle beta over 9 years

L49 risk-reject bucket: long-only spot, huge return, >50% DD, only 55 trades — will fail the promotion drawdown floor; 2 params can't fix a regime rule that rode the bear down

BtcSpotRegimeTimedExposureLongCash targeted the one construction the distilled lessons endorse — a long-only spot strategy with an explicit regime exit-to-cash overlay to cap drawdown below buy-and-hold — with a clean 2-parameter design on the under-represented spot venue. Its +2230% headline over nine years looked spectacular, but it failed exactly the metrics it was built to win: max drawdown 60.7% (it held through the 2022 bear) and a negative -0.11 information ratio, with beta 0.54 and 0.74 correlation revealing captured BTC bull-cycle beta rather than alpha. The analyst abandoned it at backtest review as the L49 long-only-spot risk-reject signature that cannot clear the drawdown floor, noting the endorsed concept was sound but this implementation delivered neither the drawdown-capping nor the positive IR it promised. It never reached optimization, analysis, or risk review.

An exit-to-cash overlay only earns its keep if it actually caps drawdown and produces a positive information ratio versus buy-and-hold — a regime filter slow enough to sit through a 48% bear delivers a 60% drawdown and negative IR, making the strategy captured beta rather than the drawdown-capping alpha it was designed to be.

The analyst abandoned it at backtest review because it falsified its own explicit, testable premise: it promised to cut buy-hold's drawdown by more than half and deliver a positive information ratio, but the exit-to-cash overlay did not cap drawdown (60.7%) and the information ratio was negative — the +2230% headline is captured BTC bull-cycle beta, not alpha. This is the L49 long-only-spot risk-reject signature (huge return, >50% drawdown, only 55 trades) that will not clear the promotion drawdown floor, and 2 parameters cannot fix a regime rule that rode the bear down.

A long-only, single-instrument regime-timed exposure strategy on BTCUSDT.BINANCE_SPOT (CASH account, daily bars, 2 parameters) that held ~100% BTC while price was in a confirmed bull regime above its moving average and moved fully to cash when the regime broke via a symmetric hysteresis band — an exit-to-cash drawdown-capping overlay whose whole promised edge was beating buy-and-hold on drawdown and information ratio.

The backtest (BTCUSDT.BINANCE_SPOT 1D, 3275 data days, 2017-08 to 2026-08) returned +2230% over 55 trades with Sharpe 0.75 (CI-low 0.22) and profit factor 2.14, but it failed on its own defining metrics: max drawdown 60.7% (CI to 83.2%) — it sat through the 2022 bear (-48.3%) — and information ratio -0.11 versus buy-and-hold, with beta 0.54 and benchmark correlation 0.74.