#48 — Multiscale shock Phase 10 + joint timing-direction model

Hypothesis — what it tests

Can wavelet-detected multi-scale shocks be monetized — either by gating FIFO-Trends entries with D3 shock timing, or by a joint timing+direction model where coordinated elevation across scales predicts continuation vs reversal?

Description

The final phases of the Multiscale Price Shock (MSP) wavelet study. Phase 10 tried to use DWT D3 shock flags as an entry filter on FIFO-Trends, but ran at the wrong physical scale — M5-resolution D3 captures ~40-minute oscillations so rare (0.06% of bars) that 99.9% of trades were filtered away, and the simplified baseline P&F kernel had a re-entry bug (5% WR); closed as a design mismatch, with the correct S5-shocks-aligned-to-M5 join left as a future fix. The joint model then combined 11 Phase-9 direction features with 16 Phase-7 shock z-score features (27 total) at each D3 shock event to test the cascade hypothesis: if coarser scales (D4/D5) are elevated too, the shock continues; if only fine scales fire, it reverses. On EUR_USD (4,708 shocks) the joint GBM lifted AUC from 0.615 to 0.665 with top-decile continuation 77.3% and all four walk-forward folds >= 0.60 for the first time — but post-spread EV averaged only +0.49p/trade (fold range -0.18 to +1.87), about 85% eaten by the 3.2-pip round-trip spread, with only ~79 shock events per fold. EUR_JPY overfit and GBP_JPY had too few events. Verdict: real structure confirmed, insufficient edge; revisit with ~5yr of EUR_USD S5 data.

Key result

Indicators

Algorithms

Code