v004 shorted altcoins that held up while Bitcoin fell, betting they would catch down. Across 349 unseen trades it lost 0.46% each.
· 2 min read · 342 words · updated
What the record says
Premise: When Bitcoin falls and an altcoin holds up, that coin is lagging and should catch down.
Unseen test: 349 trades at −0.46% net per trade after costs.
Costs: 16 bps per round trip on futures.
Live paper: The first 8 live paper trades averaged −0.84%.
Direction problem: The correlation between Bitcoin's move and the trade's gross return was −0.72. The signals were generated exactly when Bitcoin fell, which is when the trade looked best on paper.
Replay baseline: A 37-day replay produced 127 trades at +0.8 bps per trade, with a 95% confidence interval of −42 to +43 bps. The interval includes zero.
Two-sided replay: Allowing both directions produced 244 trades at +1.5 bps per trade, profit factor 1.02.
Smaller targets: Halving the target made it worse: 144 short-only trades at −3.1 bps, and 291 two-sided trades at −7.3 bps.
Hedging: Hedging the Bitcoin exposure cost more than it saved: −24.8 bps per trade.
Stop clustering: Two stop-outs carried by one Bitcoin impulse cost 1069 bps inside a run that lost 295 bps in total.
Forecast quality: The engine's own expected edge was negatively correlated with realised net at −0.32, so the trades it liked most were not the better trades.
Verdict: Every variant fails the 16 bps cost hurdle. A successor needs a stationarity gate and an exit that matches the lag horizon, not a retune.
Where the live record stands
This note covers a frozen or retired result, and no live engine reports on it.
How to check it
Every figure above is taken from the frozen test outputs and the live paper record.
The changelog dates every version and its result, and the strategy catalog lists the rules as they were frozen.
Positions described here are paper positions; no order is ever placed.
How this note was written
It is generated from a fixed fact sheet built from the frozen test outputs and the live paper record, then checked so that every figure in the text appears in that sheet. Fact sheet fb500e8f4d9e5cd4.
Paper research only. Not investment advice, and not a signal. Read the commitments for what stays public and the signal log for every signal as it was published.
