Passive-hold comparison
Passive-hold comparison · 2026-08-05
Does this system have an edge, or is it long beta minus costs?
2026-08-05. Kimi K3, consulted at maximum reasoning budget, put the whole project on trial with one claim:
“This system has no demonstrable timing edge on crypto at all — it has long beta during up-legs, minus 1% round-trip costs. BTC ‘works’ OOS because BTC’s OOS window was a 5–6x up move. SOL’s OOS window was range → one terminal pump → persistent downtrend. Same engine, different underlying drift, opposite signs.”
It named the decisive test: compare the strategy against simply holding the same notional over the same window. That test has now been run.
The benchmark
A passive position of 25% of the portfolio — the strategy’s own MAX_POS_PCT, so it
can never take more than this — bought at the start of the window and held to the end.
Same 80,000 capital base, same 0.5%/side commission. This is the honest comparison; an
earlier version of this test scaled a 100%-capital buy-and-hold by time-in-market, which
flattered the benchmark and produced a “1% capture” figure that was not a fair fight.
Result — the strategy loses badly on return and wins on risk
Out-of-sample (last 30% of each history), cooldown removed:
| Symbol | Strategy % | Strategy DD | S Net/DD | Passive % | Passive DD | P Net/DD |
|---|---|---|---|---|---|---|
| NVDA | 49.4 | 6.2 | 7.95 | 836.7 | 42.3 | 19.78 |
| AMD | 22.1 | 22.1 | 1.00 | 6389.5 | 63.3 | 101.01 |
| GOOGL | 19.1 | 7.8 | 2.43 | 129.3 | 20.4 | 6.33 |
| META | 38.7 | 10.9 | 3.55 | 62.4 | 20.4 | 3.05 |
| BTCUSD | 7.2 | 6.1 | 1.17 | 36.7 | 32.6 | 1.13 |
| PLTR | 8.9 | 7.4 | 1.20 | 24.2 | 22.3 | 1.08 |
| ARM | 14.4 | 6.4 | 2.24 | 22.9 | 24.7 | 0.93 |
| COIN | 14.8 | 4.8 | 3.08 | −5.3 | 28.2 | −0.19 |
| TSLA | 17.1 | 5.8 | 2.98 | −3.1 | 19.5 | −0.16 |
| SOLUSD | −7.5 | 5.9 | −1.27 | −10.7 | 29.2 | −0.37 |
On raw return, passive beats the strategy on 8 of 10. Not narrowly — NVDA passive made 836.7% where the strategy made 49.4%. Kimi’s core claim is confirmed: this system captures a small fraction of the move it is riding.
On risk-adjusted return, the strategy wins 6 of 10 and its drawdowns are three to seven times smaller (NVDA 6.2% vs 42.3%, BTCUSD 6.1% vs 32.6%).
Both statements are true, and neither alone is the answer.