BtcMultiWeekTrendFollowingLongDaily

BTC Multi-Week Trend Following Long-Only (Weekly-Aggregated Daily Bars, Long-Horizon Trend Capture, OHLCV-Only, BINANCE USD-M)

A long-only, single-instrument, single-venue, MACRO-HORIZON trend-following strategy on BTCUSDT.BINANCE 1D bars. Aggregates daily bars into weekly-equivalent metrics (8-week return, 26-week return, 52-week return) and trades only when all three weekly-scale momentum signals are simultaneously positive — capturing multi-month / multi-quarter BTC bull trends with HOLDING PERIODS OF WEEKS TO MONTHS. EXPLICITLY ADDRESSES the macro_1w_plus horizon bucket which is at 0.1% (1 in 703 — the second-most-under-represented horizon after short_1m_15m). FOLLOWS THE VERIFICATION-SURVIVAL RECIPE confirmed by 3 consecutive landed OHLCV-only daily strategies (BtcVolatilityRegimeConditional, EthBtcRatioMeanReversion, BtcEthSolCrossSectionalMomentumRotation): (1) OHLCV-ONLY (no supp_data → no developer pivot to sub-daily); (2) EXPLICIT daily bar type — 'BTCUSDT.BINANCE-1-DAY-LAST-EXTERNAL'; (3) single venue (BINANCE USD-M). The weekly metrics are COMPUTED FROM daily bars in-code — no actual weekly bar type is requested, eliminating the bar-availability failure mode entirely. ORTHOGONAL TO ALL MY LANDED STRATEGIES: BtcVolatilityRegimeConditional trades 5-10 day regime swings; EthBtcRatioMeanReversion trades 5-14 day basis reversion; BtcEthSolCrossSectionalMomentumRotation trades 1-week rotations; THIS strategy trades MULTI-MONTH trends. Holding period 4-12 weeks median. Expected trade count 5-10/year — radically below the parameter-count-squared overfitting threshold the analyst identified in 8 ADA 4H failures. FOUR PARAMETERS — the absolute minimum that still constitutes a working strategy. Mechanism: Moskowitz-Ooi-Pedersen (2012) canonical daily-TSMOM ON A LONG-HORIZON: long when all of 8-week, 26-week, 52-week returns are positive AND price is above 50-day SMA. Exit when 8-week return turns negative OR price breaks below 50-day SMA. Captures the multi-quarter BTC bull-cycle PROFILE that's been documented across every halving cycle (2013, 2017, 2021, 2025) and is robust to choppy weeks within the broader trend.

Implements the canonical long-horizon MOP (2012) time-series-momentum the hypothesis specifies, addressing the under-represented macro_1w_plus horizon, and follows the verification-survival recipe: OHLCV-only (no supp_data), explicit daily bar type, single venue, weekly metrics computed from daily bars so there is no weekly-bar-availability failure mode. Two engineering judgments were required. (1) VENUE: I routed to BINANCE_SPOT rather than the proposal's USD-M perp. The strategy holds for weeks-to-months and is unlevered with a price-trend (not funding) edge, which is exactly the venue rule's long-only-no-leverage->spot case; on futures a multi-month long pays funding every 8h (persistently positive in BTC bull trends, so the long pays ~1-2%/month), a drag that on ~3-5 trades/year dwarfs spot's higher per-trade fee and that spot avoids entirely. All the OHLCV-only/daily-bar/single-venue verification-survival properties are preserved on spot. (2) EXIT HYSTERESIS: the hypothesis demands BOTH 5-10 trades/year AND 4-12-week holds 'robust to choppy weeks', but a real-data sweep on full BTC daily history showed the literal 'exit when 8-week return turns negative or price < 50-SMA' produces a 3-day median hold (the 8-week return is noisy and crosses zero often inside uptrends) — whipsaws that also drag avg-trade-return toward the 0.15% spot abandonment gate. I therefore added a small exit_threshold (default 3%) so exits require a decisive rollover; this delivers the stated intent (median ~13-day holds, 7 holds > 8 weeks capturing the big trends) and ~22 trades over 6.4 years (above the >=1 Layer-3 gate and reasonable for walk-forward, acknowledging trades cluster in bull regimes as is inherent to macro long-only trend). Verified locally: Layer 1 static (clean), Layer 2 synthetic (all 6 scenarios pass, non-frozen 8-week-return signal at min_bars=369), and a real-data backtest of the entry/exit logic on clean BTC daily closes (read via the NT catalog) confirming the ~3-4 trades/year, multi-week hold distribution. BTCUSDT spot daily OHLCV has full multi-year history, so no data-gap risk.

Failed deflated Sharpe: DSR=0.0001, expected-max=12.22 over 225 trials — the optimized Sharpe of 5.34 is far below the best-of-N luck bar and statistically indistinguishable from selection noise (is_significant=false, PBO 0.661 > 0.5). Overfit by over-selectivity: walk-forward is_overfitted=true with avg IS 9.73 collapsing to avg OOS 0.0 and ALL three OOS windows exactly [0.0, 0.0, 0.0] — the optimizer thinned the gate (days_per_week 7->3, mom_short_weeks 8->4, mom_long_weeks 52->73) into a config that produces ZERO qualifying trades in every out-of-sample window, and the holdout is a degenerate fail (holdout_sharpe 0.0 against a 0.0 WF-OOS reference, ratio 0). The favorable full-period sharpe_ci_low (0.61) and PSR (0.998) are bootstrap statistics that do not correct for the 225-trial selection that DSR corrects for, and are swamped by the all-zero OOS — a config that produces no out-of-sample trades has not generalized. Sensitivity passing (cliff_count 0) is necessary but not sufficient. The edge is 2021/2024 bull-outlier beta (baseline annual 2021 +314%, 2024 +137%; return_kurtosis 20.8, the single 2021-04-21 +338% per-day attribution), and when forced to generalize the optimizer can only over-select into a non-trading config. This is not a tunable-region case: the 225-trial sweep already produced all-zero OOS over a cliff-free space, so a second pass hits the identical wall. Not revise_hypothesis: the multi-week TSMOM mechanism has no genuinely promoted sibling — every variant tested this session (BNB, ADA, SOL, ETH, and now BTC, the canonical strongest-case instrument) failed with this same DSR<<0.95 / all-zero-OOS / degenerate-holdout signature — so reframing onto another instrument would just re-overfit the same bull-outlier dependence. FAILURE PATTERN: single-asset long-only multi-week TSMOM (8/26/52-week all-positive gate) produces a high in-sample Sharpe carried by bull-cycle outliers but fails DSR even on BTC; the optimizer thins the momentum lookbacks to maximize IS Sharpe and yields zero walk-forward OOS trades + degenerate holdout — the entire family does not clear multiple-testing significance on any single instrument.

Long-only macro-horizon BTC trend-following on daily bars (OHLCV-only), computing weekly-scale momentum from daily closes in-code (no weekly bar type requested). It goes LONG only when the 8-week, 26-week, and 52-week trailing returns are ALL simultaneously positive AND price is above the 50-day SMA, and exits on a decisive macro rollover: the 8-week return drops below -exit_threshold (3%) OR price breaks 3% below the 50-day SMA. The exit_threshold is hysteresis that keeps holds multi-week and robust to choppy weeks within the broader trend (without it the noisy 8-week return whipsaws to ~3-day holds); on real BTC history this yields ~3-4 trades/year with a ~13-day median hold and a long tail of multi-month holds that capture the bull-cycle moves. calculate_signal returns the continuous 8-week trailing return (drives the exit); the discrete all-positive + regime gate is in should_enter. Long-only, single instrument, leverage 1.0, routed to BINANCE_SPOT.

Genuine trend-capture edge with correct shape: profit_factor 2.47, omega 5.7, avg_win $84k vs avg_loss $31k at 48% win rate; canonical Moskowitz-Ooi-Pedersen TSMOM validated across BTC halving cycles

Clean execution (25 signaled/submitted, 0 dropped), realized (end_unrealized 5.3%), on the deepest-liquidity / strongest-trend-persistence crypto

Trade count (25 over ~8.7 years, ~2.9/yr) is thin in absolute terms but by-design for the macro horizon (avg hold 35 days)

Headline Sharpe 3.74 is a per-period-attribution artifact (annualized_vol 1090%, Sortino 79.8, a single +338% day) — must be judged on realized PF/return, not Sharpe

Extreme outlier concentration: 2021 (+314%) and 2024 (+137%) carry the whole result; 2020 (-19.3%), 2023 (+1.2%), 2025 (-13.1%) flat/negative; kurtosis 20.8 — strip the 2021 trade and the edge collapses

Negative recent regime (2025 -13.1%) where the holdout lands; thin sample (~2-3 OOS trades/window) and sharpe_ci_low -1.59 straddles 0 — same profile that failed DSR on the BNB/ADA/SOL siblings

Clean execution and sensitivity passed (cliff_count 0); canonical Moskowitz-Ooi-Pedersen mechanism on the deepest-liquidity / strongest-trend-persistence crypto

Optimized in-sample profit_factor 2.39 — captures the 2020/2021/2024 bull trends when it fires

Failed deflated Sharpe: DSR 0.0001 (<<0.95), optimized Sharpe 5.34 far below expected-max luck bar 12.22 over 225 trials; is_significant=false, PBO 0.661 > 0.5

Walk-forward is_overfitted=true: avg IS 9.73 collapses to avg OOS 0.0 with ALL three OOS windows [0.0, 0.0, 0.0] — the optimizer thinned days_per_week 7->3 / mom_short 8->4 into a config that produces zero out-of-sample trades

Holdout degenerate fail (holdout_sharpe 0.0 vs 0.0 WF-OOS reference, ratio 0) — no out-of-sample activity to validate

Edge is 2021/2024 bull-outlier beta (baseline kurtosis 20.8, single 2021-04-21 +338% attribution day) that does not generalize; even BTC, the strongest-case instrument, fails

BtcMultiWeekTrendFollowingLongDaily applied the canonical Moskowitz-Ooi-Pedersen multi-week TSMOM to BTC, the deepest-liquidity, strongest-trend-persistence crypto and the family's best-case instrument, clearing the review gate with a genuine trend-capture shape (profit factor 2.47, +603%) and an 'optimize' verdict despite a flagged dependence on 2021/2024 bull outliers. Optimization confirmed the concern: deflated Sharpe collapsed to 0.0001 far below the 12.22 luck bar, PBO hit 0.661, and the optimizer thinned the lookbacks into a config producing zero out-of-sample trades across all three walk-forward windows with only a degenerate holdout. The analyst ruled it overfit bull-outlier beta and abandoned it, noting every multi-week TSMOM variant this session — BNB, ADA, SOL, ETH, and now even BTC — failed identically. It ended after one iteration as abandoned, having reached optimization and analysis but never risk review.

Single-asset long-only multi-week TSMOM fails multiple-testing deflation even on BTC, the canonical strongest-case instrument: a high in-sample Sharpe is carried by bull-cycle outlier days, and when forced to generalize the optimizer over-selects the momentum lookbacks into a non-trading config (all-zero OOS) — every variant this session (BNB, ADA, SOL, ETH, BTC) failed with the same DSR<<0.95 / zero-OOS / degenerate-holdout signature, so the whole family does not clear significance on any single instrument.

Although it passed the backtest-review gate with an 'optimize' verdict, the analyst abandoned it at the post-optimization analyzing stage: it failed deflated Sharpe (0.0001, far below the luck bar, is_significant=false), the optimizer thinned the momentum lookbacks (days_per_week 7→3, mom_short 8→4) into a config that produced zero qualifying trades in all three walk-forward OOS windows plus a degenerate holdout, and the edge was 2021/2024 bull-outlier beta — so it reached optimization and analysis but never advanced to risk review.

A long-only single-instrument macro-horizon time-series momentum strategy on BTCUSDT (daily bars, OHLCV-only) that computed weekly-scale 8/26/52-week trailing returns in-code and went long only when all three were simultaneously positive and price was above the 50-day SMA, exiting on a decisive 8-week-return rollover or a break below the SMA — the canonical Moskowitz-Ooi-Pedersen multi-week TSMOM, routed to BINANCE_SPOT to avoid funding drag on weeks-to-months holds.

The baseline showed a genuine trend-capture shape over just 25 trades (profit factor 2.47, omega 5.7, avg win $84k vs avg loss $31k at 48% win rate, +603% total return) but the headline Sharpe 3.74 was a per-period-attribution artifact (annualized vol 1090%, a single +338% day) with extreme outlier concentration (kurtosis 20.8, carried by 2021 +314% and 2024 +137%, while 2020/2023/2025 were flat/negative); after optimization DSR collapsed to 0.0001 (vs a 12.22 luck bar), PBO 0.661, and walk-forward in-sample 9.73 fell to 0.0 OOS with all three windows producing zero out-of-sample trades and a degenerate holdout (ratio 0).