#40 — Multiscale Price Shock (MSP) propagation

Hypothesis — what it tests

Do volatility shocks at fine time scales propagate to coarser scales with a predictable lag (Granger causality), and can that cascade be converted into a tradeable timing and/or direction signal net of spread?

Description

A full research pipeline testing whether volatility shocks propagate up a scale hierarchy in FX, adapted from the academic multiscale-microstructure framework. S5 bars were decomposed into rolling time-window features (5s to 1h TR-rate, MAD z-score shock flags) and DWT wavelet bands D1-D8; phases covered descriptive lift tables, Granger causality and transfer entropy, walk-forward predictive models, and a trading test. The headline positives: all 15 fine-to-coarse band pairs pass Granger causality at p<0.0001 with 50-100s lags, and a 13-feature model predicts D5 (80-160s scale) shock occurrence 110 seconds ahead at AUC 0.73-0.88 with 5-7x top-decile lift. The headline negative: direction models plateau at AUC 0.60-0.665, and the best joint timing+direction model (EUR_USD GBM) nets only +0.49p per trade after the ~3.2p round-trip spread — real structure, insufficient edge. A Phase-10 attempt to gate FIFO-Trends entries with the shock signal failed on a timescale mismatch (M5-resolution DWT filtered 99.9% of trades) plus a P&F kernel bug; the final Session-071 verdict was 'signal real but too sparse — not deployable standalone, best used as a volatility-regime filter'.

Key result

Indicators

Algorithms

Code