Alt/BTC momentum is still just 2021

Verdict: rejected / NO EDGE FOUND · Asset/TF: 12 alt/BTC 1h ratios · Sample: through 2026-08-27 (~6–7y per pair)

Follow-up to Idea Map watch hyp_1784972600715_10, momentum half only (long the ratio = long alt + short BTC). Mean-reversion, Kalman, maker, PCA, Hawkes, and the vol-regime flagship are out of scope.

Question: after excluding 2021, on a fresher sample, is anything left that is not cash noise?

Frozen before PnL (copied from the original card, not retuned): long-only ret_168 > 0.05 && z_168 > 1.0, exit |z_168| < 0.5, TP 20% / SL 10% / hold 336 / size 0.15. Two-leg cost 8.808 bps/side, next_open. Universe: ETH, SOL, XRP, ADA, LTC, DOGE, LINK, DOT, BCH, ETC, AVAX, ATOM. Engine RLXBT 0.2.17.

Benchmark is cash, not BTC. Neutrality on the full sample remains (yearly BTC corr +0.04).

Strategy

{
  "entry_rules": [
    {"condition": "ret_168 > 0.05 && z_168 > 1.0", "direction": 1}
  ],
  "exit_rules": [
    {"condition": "z_168 > -0.5 && z_168 < 0.5"}
  ],
  "take_profit_pct": 0.20,
  "stop_loss_pct": 0.10,
  "max_hold_bars": 336,
  "position_size": 0.15
}

Backtest (long-only net)

9/12 pairs positive full-sample. Median breakeven 46.3 bps (5.3× live). Cost is not what kills the winners.

Equal-weight yearly portfolio:

Full-sample mean year +2.56%. Ex-2021 +0.55%/yr (t = 0.85, n = 7). 2021 is still 81% of summed year-means (was 87% on the stale corpus). Extending through 2026-08 did not find a second alt-season; the two newest years are the worst.

Two-way short leg remains cost drag on 10/12 pairs. Not a promotion path.

Robustness

WFA/MC did run (ex-2021 mean > 0 and median BE > 8.808). Gate: WFE ≥ 0.5, majority OOS windows +, MC p5 > 0.

0 / 12 pairs passed. Full-sample MC p5 > 0 only on ETH/SOL/DOGE (the 2021 monsters). Drop 2021 trades and MC p5 is negative on 11/12; LINK is +0.55% and still fails OOS majority.

ETH is the cleanest full-sample pair (net +31.7%, Sharpe 0.85, DD 3.2%, BE 73 bps, 4/4 OOS+) and still WFE = 0.40; ex-2021 MC p5 −0.28%.

Research trail

GitHub alt universe → Binance 1h → ratio features → load_dataset → ai_run_backtest → walk_forward → monte_carlo. Learned: the original watch (“maybe another alt-season”) is answered. The leftover after 2021 is noise.

Reproduce

Binance 1h klines 2019-09 to 2026-08 for the 12 alts and BTC. Ratio OHLC as alt/BTC. Features ret_168, z_168. Commission 0.00088088/side, next_open. RLXBT 0.2.17.