BtcCmInversePerpMonthlyMomentumLong

BTC COIN-M Inverse Perp Monthly Momentum Long

A long-term, BTC-margined momentum strategy on the BTCUSD_PERP COIN-M inverse perpetual using daily bars. The strategy is intentionally minimal — a single regime filter, no supplementary data, no pairs, no extra timeframes — to maximize verifiability and minimize overfitting risk. It is structurally distinct from every recent failure (which were 1D/4H/1H USDT-margined directional or contrarian strategies) and from every strategy already in the pipeline (which target USD-M, spot, or pairs). By trading on COIN-M with ~0.06% round-trip taker cost (lowest available among non-options venues) and holding positions for weeks-to-months, the strategy is highly fee-tolerant: only ~3-8 round-trips per year are expected. The thesis is the persistent BTC long-momentum effect documented in cross-asset research, harvested in BTC-denominated terms — useful for portfolio diversification because PnL is naturally hedged against BTC price swings (gain compounds in BTC).

Implements the hypothesis at exactly its stated scope and no more: one signal, one regime filter, three tunable parameters (mom_lookback, regime_sma, entry_thresh) with clamps mirrored in _param_bounds, long only, no supplementary data, no pairs, no extra timeframes, and no stop/target layer beyond the regime exit the hypothesis itself specifies. calculate_signal returns the CONTINUOUS trailing return in natural fractional units every bar (never gated to a constant) and is thresholded in the same units; the regime test is a cached boolean used in should_enter. I measured the gate on the real BTCUSD_PERP daily history (2,106 bars, 2020-08 to 2026-05): the chosen defaults produce 28 entries, about 4.8 round trips a year, which sits inside the hypothesis's stated 3-8/year expectation and gives the sandbox a comfortable trade count while keeping fee drag trivial (28 round trips over ~6 years at ~0.10% is well under 3% total). I deliberately picked regime_sma=100 rather than 200 for this reason - a 200-day filter cut the count to ~2.4/year, thin for statistics. Inverse sizing uses contracts_for_notional (whole $100-face contracts), never the linear equity/price formula that mis-sizes inverse contracts; $50k notional is 500 contracts, far above min notional. leverage stays 1.0 and gross notional never exceeds 0.5x equity, so there is no compounding blow-up path - the COIN-M venue is chosen because the BTC-denominated payoff IS the thesis, not for leverage. State is fully incremental (a bounded deque for the momentum lookback plus a running-sum SMA), so the daily backtest is trivially inside the smoke budget. Honest note: with roughly 5 trades a year the strategy's statistics will always rest on a modest sample, and its performance is essentially a leverage-free bet on BTC's trend persistence, so I would expect the analyst to weigh it against a simple buy-and-hold benchmark.

coding_stuck_3_timeouts

Long-only, long-horizon momentum on the BTC COIN-M inverse perpetual (BTCUSD_PERP.BINANCE_CM, daily bars, pure OHLCV, no supplementary data, no second instrument or timeframe). Every day it computes the trailing 90-day return as its signal and applies a single regime filter - the close must be above its 100-day SMA. It goes LONG when the 90-day return is at least +5% while the close is above the SMA, and exits on the single condition that the close falls back below the SMA, which serves as both the trend exit and the risk control. Positions are held for weeks to months, so turnover is a handful of round trips a year and fee drag is negligible against the COIN-M taker cost. Sizing is a fixed 0.5x equity of gross notional floored to WHOLE $100-face inverse contracts via contracts_for_notional, so PnL accrues in BTC - the BTC-denominated momentum premium the hypothesis targets.

BtcCmInversePerpMonthlyMomentumLong aimed to harvest the persistent BTC long-momentum premium in BTC-denominated terms on the low-fee COIN-M perp, kept intentionally simple — one momentum signal, one SMA regime filter, a few round-trips per year — to maximize verifiability and resist overfitting. Despite that design discipline, it never generated a backtest: the coding stage stalled and, after three timeouts across two iterations, the run was abandoned as 'coding_stuck_3_timeouts'. With no completed code there were no verification, backtest, optimization, or analysis results. The abandonment reflects a coding-process failure rather than any conclusion about the momentum thesis.

Even a deliberately minimal, low-turnover, fee-tolerant design is worthless if it cannot clear the coding stage — repeated coding timeouts terminate a candidate before its edge is ever tested, so implementation must be reliably completable within the stage's time budget.

It was abandoned for 'coding_stuck_3_timeouts': the coding stage failed to complete after three timeouts across two iterations, so the strategy never reached verification or any downstream stage. This is an infrastructure/coding-process failure, not a verdict on the mechanism.

A long-only, single-instrument, pure-OHLCV long-horizon momentum strategy on the BTC COIN-M inverse perp (BTCUSD_PERP.BINANCE_CM, daily bars) that went long when the trailing ~90-day return cleared a threshold and the close was above a ~100-day regime SMA, exiting only when the close fell back below that SMA — deliberately minimal (three parameters, weeks-to-months holds) to harvest BTC-denominated momentum with minimal turnover and fees.

No performance data exists — the strategy never produced a backtest. It stalled in the coding stage, so no verification, backtest, optimization, or analysis metrics were generated.

Verification failed (Layer 4 — QA review) [class=no_edge]: - NO EDGE OVER BUY-AND-HOLD + UNMEASURABLE. Net positive (+55.6%, PF 2.18) but that is BTC trend beta, not skill: alpha -0.0123 (negative), information_ratio -0.713, beta 0.234, benchmark_correlation 0.65 — it UNDERPERFORMS a simple BTC hold risk-adjusted, the exact comparison the developer flagged. Sharpe 0.386 on CI [-0.25, 1.03] straddling zero (not distinguishable from no-skill, below the 0.5 OOS floor), return thin-tail dependent (win_rate 0.296, avg_win $12,166 driven by a $37,922 largest win). Only 27 trades over 5.8 years (~4.8/yr): the holdout carries ~5, so it cannot pass the hard >=10-trade holdout gate and optimization would die there. Single-instrument long-only OHLCV momentum where base Sharpe is the ceiling; no 3-param change lifts 0.386 past the floor or turns negative alpha positive. - The BTC-denominated 'diversification' framing doesn't hold: the book is 65% correlated to BTC with positive beta and negative alpha — not hedged against BTC swings, just long BTC trend with worse risk-adjusted return than holding BTC.