BtcDailyTimeSeriesMomentumVolNormalizedLS

BTC Single-Asset Time-Series Momentum Long-Short on BINANCE Daily (Moskowitz-Ooi-Pedersen 2012, Vol-Normalized Trend Signal, Single Instrument)

Single-asset time-series momentum strategy on BTCUSDT.BINANCE — the SAFEST possible data path in the factory (single instrument, daily bars, no multi-asset dependency). UNLIKE my abandoned 3-asset Time-Series Momentum hypothesis (which died because the developer translated to 4H and ETH/SOL 4H bars consistently fail), this hypothesis depends ONLY on BTCUSDT.BINANCE — the most reliable instrument in the entire factory data set per the system's explicit safety guidance. The signal: long when BTC's vol-normalized trend t-stat exceeds +threshold, short when below -threshold, flat in the middle band. This is the SINGLE-ASSET application of Moskowitz-Ooi-Pedersen 2012 'Time Series Momentum' (Journal of Financial Economics) — the academic factor with the highest documented Sharpe ratios (~1.0+) across asset classes. Empirically uncorrelated with cross-sectional ranking strategies because it's a within-asset trend signal, not a between-asset relative-strength signal. Single-asset scope is DELIBERATELY different from my prior 6 multi-asset factor hypotheses — those keep dying at iter-4 via 4H translation; this avoids that failure mode entirely. Helps direction quota (currently 89.9% long-only) because the strategy can go LONG, SHORT, or FLAT depending on signal — natural direction diversification. Helps short_1m_15m quota indirectly by being a clear single-instrument strategy at safe timeframe. CRITICAL DATA SAFETY: BTCUSDT.BINANCE at daily is the single most-tested, most-reliable instrument-timeframe combination in the entire factory — the system's safe-list message explicitly puts it first.

Iteration 2 addresses the backtest-review feedback that the strategy was too sparse (~36 trades / 6y) and outlier-dominated (one +197% trade, kurtosis 938). The t-stat long/short TSM mechanism — which passed all verification layers — is kept intact; only trade-density levers were changed. Shortened the trend lookback 60->25 (more reactive t-stat, more sign flips), lowered entry_t 1.5->0.9 and exit_t 0.5->0.3 (gate fires more often and decays out faster), and added a hard 20-bar max_hold_bars turnover cap so a single multi-month trend no longer collapses into one mega-trade. The max-hold cap simultaneously raises trade count and cuts return kurtosis by splitting large trends into several bounded trades. These changes target trade_count > 60 with a reasonable kurtosis before optimization, aligning density with the SOL single-asset TSM sibling while preserving the vol-normalized long/short signal construction.

Fundamentally unprofitable base, not worth optimizing (already dev iter-2). profit_factor 0.971 (<1), negative expectancy (-170.9/trade), negative CAGR (-4.06%), and a 78% max drawdown. The positive headline total_return (+22.68%) is an artifact: end_unrealized_pct 45.75 means realized return is roughly -23% — the paper gain is one un-exited open position. Profitability is concentrated entirely in 2020 (+143.7%) with every subsequent year but two negative (kurtosis 84.9, tail_ratio 0.23), so the single-asset BTC daily TSM edge has decayed to noise post-2020. A best-of-225 optimizer would only curve-fit the 2020 trend regime; this matches the repeatedly-failing single-asset BTC daily trend-follower family (prior DSR 0.15-0.22, outlier-carried). No parameter region rescues an absent, negative-expectancy edge.

Single-asset vol-normalized Time-Series Momentum (Moskowitz-Ooi-Pedersen 2012) on BTCUSDT.BINANCE daily bars. Signal is the t-statistic of an OLS regression of log-price on time over a 25-day lookback: long when t >= +0.9, short when t <= -0.9, flat in the middle band. Exits on trend decay back into a |t| < 0.3 band / sign flip, a 20-bar max-hold turnover cap, or a wide 20% safety stop. Positions are sized by MOP inverse-vol / vol-target scaling (target 40% annualized), capped at 1x equity notional.

Trades implement the stated mechanism: 73 long / 62 short / flat, genuine direction diversity from a within-asset trend signal

Adequate trade count (135) and safe single-instrument daily data path — sample size is not the problem

Vol-normalized t-stat signal and vol-target sizing are correctly coded (iter-2 density fixes are present)

profit_factor 0.971 (<1) and expectancy -170.9 — negative per-trade edge, the strategy loses money net of fees

Headline total_return +22.68% is a mirage: end_unrealized_pct 45.75 means realized return is ~-23%; CAGR is -4.06%

Catastrophic risk: 78% max drawdown (CI to 98.6%), recovery_factor -0.26, 1996-day drawdown duration

Edge concentrated entirely in 2020 (+143.7%); every subsequent year except modest 2023/2026 is negative — outlier-carried (kurtosis 84.9, tail_ratio 0.23)

Single-asset BTC daily trend-following is a family that has repeatedly failed deflated Sharpe; optimizing a PF<1 base only curve-fits the 2020 regime

Do not optimize yet — the sample is far too thin and outlier-dominated. Only 36 trades over ~6 years (~6/yr), below the ~62 walk-forward floor: the t-stat entry (entry_t 1.5, lookback 60) combined with 58-day average holds fires too rarely, and the +582% headline rests on ~2-3 trend-capture trades (2021-05-04 +197% trade-close) plus a 240%-of-headline unrealized open position (kurtosis 938, skew 27.7). A 225-trial optimization on 36 lumpy trades will fail the deflated-Sharpe gate (the structurally-identical 37-trade DOGE breakout came back DSR 0.0). FIX — increase trade density to >60-100 while keeping the long/short TSM mechanism: lower entry_t (test ~0.75-1.0), shorten the trend lookback (test 20-30 instead of 60), and/or shorten holds so trades turn over faster. The SOL single-asset TSM sibling achieved 261 trades and optimized to Sharpe 1.01 using a more-sensitive tanh-threshold signal (entry_threshold 0.25) rather than a t-stat>=1.5 gate — consider aligning the signal construction. Re-backtest and confirm trade_count > 60 AND return_kurtosis is reasonable (no single trade dominating) AND a positive sharpe_ci_low before resubmitting for optimization. If a denser config still cannot exceed ~60 trades or remains outlier-dominated, abandon — single-asset TSM on BTC daily is then too sparse/lumpy to validate, and the mechanism is already carried by the SOL sibling.

This strategy deliberately narrowed a repeatedly-failing multi-asset time-series momentum idea to a single instrument — BTCUSDT daily, the factory's most reliable data path — to escape the 4H-translation failure mode that had killed its predecessors, and its iteration-2 tuning successfully raised trade density to 135 and cut hold times. But the safe data path did not produce a safe edge: the base was net unprofitable (profit factor 0.971, negative expectancy and CAGR) with a 78% drawdown, and its only real profitability came from the 2020 trend regime while nearly every later year lost money. The reviewer abandoned it at the backtest-review gate rather than optimize a PF<1 base, noting the flattering +22.68% headline was really ~-23% realized once the un-exited open position was excluded. It never reached optimization or the analyst, and joins the family of single-asset BTC daily trend-followers that have consistently failed this factory's bar.

Choosing the safest data path and adding density/turnover fixes cannot rescue a mechanism whose base edge is negative-expectancy (PF<1); single-asset BTC daily time-series momentum has decayed to noise post-2020, and a positive headline return driven by unrealized open MTM is not real edge.

It was abandoned at the pre-optimization BACKTEST_REVIEW gate for a fundamentally unprofitable base — profit factor below 1, negative expectancy and CAGR, and a catastrophic 78% drawdown — with profitability concentrated entirely in 2020 (+143.7%) and nearly every subsequent year negative, so best-of-225 optimization would only curve-fit the decayed 2020 regime.

A single-asset, long/short/flat time-series momentum strategy on BTCUSDT Binance daily bars (Moskowitz-Ooi-Pedersen 2012), going long when a vol-normalized trend t-statistic exceeds +entry_t, short below -entry_t, and flat in between, with inverse-volatility/vol-target sizing on the factory's safest single-instrument daily data path.

The backtest showed a negative per-trade edge across 135 trades (73 long / 62 short): profit factor 0.971, expectancy -$170.9/trade, CAGR -4.06%, Sharpe 0.218, and a 78% max drawdown (CI to 98.6%); the +22.68% headline total return was an artifact of one un-exited open position (end_unrealized_pct 45.75, implying ~-23% realized).