BtcSpotWeekendHoldingCalendarLong

BTC Spot Weekend Holding Long — Time-Based Calendar Strategy (Spot)

A LONG-ONLY CALENDAR-EFFECT strategy on BTCUSDT spot: BUY BTC at Friday 22:00 UTC close, HOLD across the weekend, SELL at Monday 08:00 UTC open. RATIONALE driven by the now-EXHAUSTIVE session finding (24+ strategies, 7 mechanism classes ALL FAILED including: trend-following, channel breakouts, oscillators, cross-asset pair MR, cross-sectional momentum, sub-daily intraday MR, basket strategies). The analyst's final calibration explicitly states 'every mechanism class theoretically promising at session-start has been empirically falsified' and recommends 'STOP generating hypotheses' or 'wait for regime change.' This proposal explores a GENUINELY UNTESTED MECHANISM CLASS: time-based calendar effects — specifically the weekend-holding anomaly in BTC spot. KEY STRUCTURAL DIFFERENCES from all 24+ failed strategies: (a) NO SIGNAL — entries/exits are PURELY time-based, no indicator, no oscillator, no MR threshold, no rank, no spread; (b) SIMPLEST POSSIBLE STRATEGY — only 2 parameters (entry time, exit time), both FIXED at canonical calendar boundaries; (c) NOT a directional bet — it's an exploitation of a STRUCTURAL CALENDAR effect; (d) BTC SPOT venue eliminates funding-rate complications; (e) NOT cross-asset, NOT cross-venue, NOT cross-sectional — pure single-instrument calendar; (f) LOW TRADE FREQUENCY (~50/year) keeps fee drag minimal at BINANCE_SPOT 0.20% RT. EDGE THESIS: BTC has documented WEEKEND POSITIVE DRIFT in the post-2024 ETF era specifically because: (1) BTC ETFs (IBIT, FBTC) trade ONLY during US weekday market hours — there's no institutional ETF selling pressure during weekends; (2) Retail flows dominate weekend BTC volume and historically tilt LONG; (3) Weekend news cycles (China policy, macro events) often resolve favorably for BTC by Monday open; (4) Multiple Bitcoin academic papers (Aharon-Qadan 2019, Kaiser 2019) document the persistent BTC weekend effect with ~0.3-0.5% average weekend return historically. CRITICAL ACKNOWLEDGEMENT: every recommendation I've made this session has been falsified. This proposal's confidence should be LOW (not high). It's genuinely-untested in this session and is structurally different from the 24 failed mechanism classes, but the analyst's pattern of mis-calibrated recommendations means we should temper expectations. If pre-opt Sharpe < 0, abandon immediately — no iteration.

Iteration-2 fix for the Layer-2 TypeError: `_bar_timestamp_ns` was a `@staticmethod` with signature `(bar)`, but the synthetic verifier's strategy proxy invokes helpers as bound instance methods, passing `self` as an extra positional arg → 'takes 1 positional argument but 2 were given'. Converted it to a regular instance method `def _bar_timestamp_ns(self, bar)`. This is the smallest possible change: all imports, class structure, signal, entry/exit and sizing logic (which already passed Layer 1) are unchanged. The strategy exploits the untested time-based weekend-drift calendar effect on BTC spot, keeping it long-only on the CASH BINANCE_SPOT venue with ~50 trades/year to minimise fee drag.

Pre-optimization abandon per the hypothesis's own hard kill-switch (pre-opt Sharpe < 0 -> abandon, no iteration). The BTC weekend-holding calendar effect has NO edge net of fees: Sharpe -0.461, total_return -49.4%, profit_factor 0.909, negative expectancy, and NEGATIVE alpha (-0.102) / information_ratio (-0.748) vs BTC buy-hold (benchmark_meaningful=true) — it is strictly worse than holding BTC. The thesis of post-2024 ETF-era weekend positive drift is directly falsified in the target period (2025 -3.2%, 2026 -25.5%; losing in 6 of 9 years). Nothing to optimize: the only parameters are the fixed Friday-22:00/Monday-08:00 calendar boundaries that define the hypothesis, so tuning them would merely curve-fit noise, and the sign of the edge is wrong. FAILURE PATTERN: a pure time-based BTC-spot weekend-holding calendar long has no exploitable drift after 0.20% RT spot fees — a documented academic weekend effect does not survive as a tradeable, fee-net edge, and its structural novelty (no signal, 2 fixed params) does not rescue a strategy that loses money with negative alpha vs simply holding the asset.

Long-only BTC spot weekend-holding calendar strategy. Buys BTCUSDT at Friday 22:00 UTC, holds across the weekend, and sells at Monday 08:00 UTC. Entries/exits are purely time-based (no indicator) via a wrap-aware window-membership test on the bar timestamp, robust to missing boundary bars. Runs on BINANCE_SPOT 1-HOUR bars, long-only, no leverage.

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['exit_hour', 'entry_hour', 'exit_weekday', 'min_notional', 'entry_weekday', 'capital_fraction'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: TypeError: BtcSpotWeekendHoldingCalendarLong._bar_timestamp_ns() takes 1 positional argument but 2 were given (bar timestamp: 1735689660000) - steady_downtrend: TypeError: BtcSpotWeekendHoldingCalendarLong._bar_timestamp_ns() takes 1 positional argument but 2 were given (bar timestamp: 1735689660000) - flat_ranging: TypeError: BtcSpotWeekendHoldingCalendarLong._bar_timestamp_ns() takes 1 positional argument but 2 were given (bar timestamp: 1735689660000) - volatility_spike: TypeError: BtcSpotWeekendHoldingCalendarLong._bar_timestamp_ns() takes 1 positional argument but 2 were given (bar timestamp: 1735689660000) - zero_volume: TypeError: BtcSpotWeekendHoldingCalendarLong._bar_timestamp_ns() takes 1 positional argument but 2 were given (bar timestamp: 1735689660000) - price_gap: TypeError: BtcSpotWeekendHoldingCalendarLong._bar_timestamp_ns() takes 1 positional argument but 2 were given (bar timestamp: 1735689660000)

Clean, minimal implementation with correct time-based entry/exit and no lookahead

Adequate trade count (419) and exposure for a meaningful read on the calendar effect

Negative Sharpe (-0.461) with CI straddling well below zero, and negative Sortino (-0.577) — no risk-adjusted edge

Loses money outright: total_return -49.4%, CAGR -8.07%, profit_factor 0.909, negative expectancy

Negative alpha (-0.102) and negative information_ratio (-0.748) vs BTC buy-hold (benchmark_meaningful=true) — worse than simply holding BTC

Weekend-drift thesis is falsified in the very ETF-era it targets: 2025 -3.2%, 2026 -25.5% YTD; losing in 6 of 9 years

Fee drag is heavy (commissions 14.9% of gross), consuming any thin calendar edge on BINANCE_SPOT

BtcSpotWeekendHoldingCalendarLong was a deliberately novel, last-resort experiment after 24+ signal-based strategies had failed in the session — a pure Friday-to-Monday weekend-holding calendar long on BTC spot, motivated by a claimed ETF-era weekend positive-drift anomaly and proposed with explicitly low confidence and a built-in kill-switch. The backtest falsified the thesis in the very period it targeted: across 419 trades it lost 49.4% total with a -0.461 Sharpe, negative expectancy, and heavy fee drag, losing in 6 of 9 years including -25.5% in 2026 YTD, and underperforming buy-and-hold on both alpha and information ratio. The backtest reviewer returned 'abandon,' and because the only parameters were the fixed calendar boundaries that define the hypothesis, there was nothing to optimize — tuning them would merely curve-fit noise on an edge whose sign was already wrong. The strategy was retired before optimization under its own pre-opt-Sharpe-negative rule.

A documented academic calendar effect (BTC weekend drift) does not survive as a tradeable, fee-net edge after 0.20% round-trip spot fees, and structural novelty (no signal, two fixed parameters) cannot rescue a strategy that loses money with negative alpha versus simply holding the asset.

It was abandoned at the pre-optimization BACKTEST_REVIEW gate with an 'abandon' verdict, triggering the hypothesis's own hard kill-switch (pre-opt Sharpe < 0 → abandon, no iteration); optimization and all later stages were never reached.

A pure time-based calendar strategy on BTCUSDT spot with no price signal — buy at the Friday 22:00 UTC close, hold across the weekend, and sell at the Monday 08:00 UTC open — betting on a post-2024 ETF-era weekend positive-drift anomaly using only two fixed calendar parameters.

Over 419 trades from 2018 to 2026 it lost money outright: Sharpe -0.461 (CI low -1.93), total return -49.4%, CAGR -8.07%, profit factor 0.909, 50.4% win rate, 63.8% max drawdown, and negative expectancy. It also delivered negative alpha (-0.102) and information ratio (-0.748) versus BTC buy-and-hold, with commissions consuming 14.9% of gross.