Elasticity-Reversal (ER) trades the divergence between an FX pair's return and what its equity driver predicts. The signal is built from the cumulative beta-adjusted residual of each FX pair against an equity-index driver; entry timing comes from an EMA crossover on that residual. The current production version operates on hourly bars and runs continuously rather than session-restricted. ER 1h was revalidated under v3 doctrine in May 2026 and is one of the three Tier-1 admits running in shadow on the per-system attribution architecture. It is not on live capital yet — re-arm is held until the FV-Fast 24-hour clean window finishes and ER passes its own forward gates. The earlier ER variant (ER 15m on 1-minute data, US-session-restricted, with Monte-Carlo-validated regime gating) was killed and is no longer the canonical version.
For each FX pair, a rolling beta against an equity-index driver (ES futures / S&P 500) is estimated from log-return covariance over a long lookback. The residual at each bar is the FX return less beta times the driver return — the portion of the FX move unexplained by equity behaviour. Cumulative residuals across time produce a running measure of flow that has accumulated in the FX leg independent of equity coupling.
Two exponential moving averages — a fast one and a slower one — are applied to the cumulative residual. The signal is the fast minus the slow EMA, standardised by its own rolling standard deviation. The strategy enters when the standardised signal exceeds a per-pair entry threshold and exits via a 5× ATR initial stop trailed to 3× ATR once unrealised profit reaches 1R, or via a flip when the signal sharply reverses.
The mechanism is unchanged in principle from earlier versions of ER — what changed in v3 is the cadence, the operating window, and the gating discipline.
The pre-v3 ER variant (referred to internally as "ER 15m" or "ER 1-minute") had three properties that v3 either kills or replaces:
1. 1-minute bars on the US session only. v3 ER runs on hourly bars 24/7. Sub-hourly precision did not survive the deflation gauntlet — the apparent Sharpe lift of trading at 1-minute resolution was eaten by holdout degradation. The 24/7 operating window also outperformed the US-session-only restriction in walk-forward, because the residual-accumulation mechanism does not actually require equity-cash-session activity to function; it requires equity *price* data, which futures provide overnight.
2. Per-pair Monte-Carlo regime gating (the ROBUST / VIABLE / FRAGILE / NOISE labels assigned to 120 pair-regime-direction slices in the pre-v3 version). This gating failed PBO when audited under v3 — most of the slice-specific lift was selection on noise. The current ER trades pooled across regimes.
3. Headline Sharpe 4.18. That number came from the regime-gated 1-minute version, evaluated on a substrate that had not yet been cleaned of the USDCNH minute pollution. It overstated genuine edge. The post-cleanup, post-regime-cancellation v3 ER 1h Sharpe is materially lower but holds up under walk-forward.
The ER 15m configuration was formally killed 2026-05-13 when the canonical paper shadow was switched to ER 1h. The 15m state is preserved under tier1_shadow_state/_deprecated_er_15m_pre_2026_05_13/ for audit history only.
The driver is an equity index — historically S&P 500 cash via continuous CFD data, currently ES futures from Interactive Brokers. The driver comparison study on the December 2025–March 2026 overlap period found ES matched or exceeded the legacy SPX driver on seven of nine testable pairs, with slightly fewer trades due to ES's coarser tick at 1-minute (irrelevant at 1-hour). v3 uses ES live.
ER 1h is calibrated for ten USD pairs where equity-FX coupling is strong enough for the residual to carry information: EURUSD, GBPUSD, USDJPY, AUDUSD, USDCAD, USDCHF, NZDUSD, USDMXN, USDPLN, USDCNH-excluded. Pair-specific entry thresholds and EMA spans are calibrated per pair; the variation reflects genuine differences in how each pair responds to equity dynamics — USDJPY's safe-haven role requires wider EMA windows and a lower entry threshold; high-carry pairs like USDMXN require tighter windows but higher entry thresholds to filter noise.
Like FV-Fast, USDCNH is excluded — the same 2026-05-03 substrate audit found polluted minute bars there.
Re-run under v3 doctrine on the post-cutover frozen substrate (post-USDCNH-cleanup). The four core deflation gates:
| Gate | Threshold | ER 1h |
|---|---|---|
| Walk-forward Sharpe (annualised, multi-fold) | All folds > +1.0 | Passes |
| Deflated Sharpe Ratio | DSR > 0.95 | Passes |
| Probability of Backtest Overfitting | PBO < 0.20 | Passes |
| Holdout replication | Holdout SR > 0.6 × WF SR | Holdout SR ≈ +2.1 |
On the 3-year Tier-1 backfill ER 1h logs ~2,500 trades with a hit rate near 46% and asymmetric payoff (average win ≈ 1.6× average loss) — positive expectancy through tail behaviour, not through high win frequency.
ER's per-pair Sharpe stratification differs from FV-Fast: there is no EM amplification in ER. Cross-asset-driver residual mean reversion appears to derive its alpha from a more universal mechanism (equity-FX risk-coupling reversal) rather than from pair-specific dealer-inventory dynamics. This is consistent with the project's observation that not every Tier-1 mechanism amplifies in EM the way FV-Fast does.
ER 1h trades through the same central order router as FV-Fast. Orders carry orderRef='er_1h:{signal_id}'. The reconciler treats ER as an independent system; ER's positions on a pair cannot be inadvertently flattened by FV-Fast's reconciliation logic on the same pair, and vice versa. This is the property that allows ER to be re-armed on a separate schedule from FV-Fast.
tier1_paper_shadow.py on mini at the canonical :15 cadence. Trades are recorded to a shadow ledger but no orders are placed.ER is not a momentum or trend-following strategy. The signal is a flow-divergence indicator — when the FX leg has accumulated movement that the equity driver does not justify. ER does not condition on macro releases, sentiment, or news. It does not use news-latency arbitrage as a signal source. The strategy expects to lose to genuinely-news-driven moves and accepts that as the cost of trading the inertial flow accumulation that follows them.
ER is one of three Tier-1 strategies under the project's two-tier architecture: FV-Fast (residual MR on multi-driver OLS of USD-pair returns) and Range Trading (Donchian-style fade) are the other two. The three trade their own signals through the same execution router. Portfolio overlay across the three sleeves is currently planned as HRP-rolling once the paper shadow has produced enough continuous history (estimated mid-July); a Tier-3 meta-labeler is also research-tracked but lost the 2026-05-11 A/B against HRP and is not on the deployment path. ER and FV-Fast have near-zero return correlation on daily resampling — they trade fundamentally different residuals on different timescales.