Tuatara Headline Signal — Hourly Return Curve

Equal-weight directional return of the Tuatara vector basket, sampled every hour for 72 hours after a news headline, aggregated across 321 tradeable headlines. Causal entry at the headline timestamp; BTC is the benchmark.

Tuatara crypto basket hourly return curve, 0 to 72 hours, vs BTC benchmark

Where the alpha lives. The signal’s edge over the market (Tuatara − BTC, green) is front-loaded: it peaks at +0.78% around 13 hours and is positive only in the ~6–20 hour band, then decays to −2.5% by 72h as BTC rallies and the basket lags. For market-neutral / long–short use, the optimal hold is ~12–18 hours, not days — the daily backtest’s “2–3 day” best hold was capturing BTC beta, not signal.

Return by hold time

Hold (h)	Basket mean %	vs BTC %	Win %	n
1	−0.13	−0.08	39.4	193
6	−0.37	+0.15	44.3	192
12	−0.29	+0.64	43.8	192
13 ◄ peak alpha	−0.26	+0.78	43.8	192
18	−0.43	+0.56	45.8	192
24	−0.62	−0.01	43.6	202
36	−0.61	−1.00	42.7	192
48	−0.73	−2.09	41.6	202
72	−0.03	−2.50	46.5	202

Method

Asset class crypto (the only universe with dense intraday history): prices from historical_prices (~1-min cadence), 24/7 trading so the curve has no market-hours gaps. Baskets from the Tuatara vector engine asset_deep_discovery (namespace CMDB-crypto-v01), top-12 by score. Direction = sign of FinBERT headline sentiment (neutral band → no trade). Per leg, hour h: dir × (px(t₀+h)/px(t₀) − 1), equal-weight, split/data-artifact-guarded, minus round-trip cost. Benchmark = BTC (undirected).

Stocks are not shown: no intraday stock history is available (daily bars only), so an hourly stock curve would need a paid intraday feed — the curve logic is identical once one is wired in.

Download CSV Download JSON Download chart (PNG)

Cymetica · Tuatara vector research · signal/research only, not investment advice · cymetica.com