After a vol shock, size persists but the sign is a coin
After a vol shock, size persists but the sign is a coin
Verdict: NO_EDGE · Asset/TF: BTCUSDT 15m / 1h / 4h · Sample: multi-year chronological 60/20/20
Follow-up to Volatility compression predicts persistence, not expansion. That paper showed quiet stays quiet. This one asks the other side of persistence: after a large bar, is the next move still large, and does the sign of the shock continue or fade?
Q1 — magnitude (check)
Top-decile |log return| events (discovery-frozen, consecutive bars collapsed to one event). Validation event mean |forward r| / unconditional:
| TF | median lift across horizons |
|---|---|
| 15m | 1.76× |
| 1h | 1.50× |
| 4h | 1.28× |
OOS same direction. Persistence replicates. The previous compression paper is consistent: vol clusters.
Q2 — sign (new claim)
Does the shock continue or fade more than a typical same-sign bar?
Continue rate after shock ≈ 0.44–0.47 (near a coin). Δ vs sign-only (no shock required) ≈ −0.02: the shock does not add momentum information, and there is no clean fade either.
Ablation: shock-only / sign-only / both / unconditional — key comparison both vs sign-only fails. Negative control (shuffled shock labels): real 1h Δ not extreme. BTC 1h gate failed → ETH/SOL not run.
Events vs bars (val): 15m 2319/3132, 1h 653/815, 4h 175/197.
Trading (diagnostic only)
A minimal against-shock next_open rule that looked mildly positive at 0 bps dies at 5–10 bps round-trip. Not tradeable even as a curiosity.
Look-ahead
Causal discovery-only |r| decile, chronological splits, event collapse — YES. No full-sample thresholds.
Final Verdict
NO_EDGE
Volatility magnitude persists after shocks, but the sign of the shock is approximately a coin and adds nothing beyond ordinary same-sign bars. Do not build continuation or exhaustion strategies from this feature alone. Useful negative result: size forecasting ≠ direction forecasting.
Comments (0)