EthMacroEventImpulseContinuationLS

ETH Scheduled-Macro-Event Impulse Continuation — Long-Short, Single-Instrument: Trade the Direction of ETH's Initial Reaction to US CPI/FOMC/NFP Releases and Ride the Intraday Continuation, ATR-Trailed, Flat Otherwise (BINANCE ETHUSDT USD-M, 15m Bars, Embedded Event Calendar, 2-Parameter)

An EVENT-DRIVEN, single-instrument, intraday directional strategy on ETHUSDT.BINANCE USD-M perp that is deliberately NOT in any of the saturated dead classes (OHLCV trend-confluence 0/213, single-perp funding/OI gates L70/L71, cross-venue carry L69, rotation baskets L68, aggressor-flow, vol-squeeze breakout, mean-reversion). It trades ONLY in a narrow window around SCHEDULED US macro releases — CPI, FOMC rate decisions, and Non-Farm Payrolls — which are the highest-conviction, mechanically-recurring liquidity/volatility events in crypto. The event calendar is KNOWN HISTORY and fully backtestable today (explicitly blessed for scheduled-macro-event strategies), requiring NO un-injectable supplementary feed (the failure mode that killed the liquidation-cascade and cross-venue ideas). The economic basis: at a scheduled release, a large cohort of participants must re-price ETH simultaneously against a surprise they could not position for in advance; the initial 15-30 minute impulse reflects the aggregate surprise and, because information diffuses and leveraged positioning unwinds over hours, that impulse tends to CONTINUE intraday rather than instantly mean-revert. The strategy stands completely flat on all non-event bars, so it accrues almost no fee drag and takes only high-expected-move trades. Long-SHORT (fills the starved long-short bucket, 14.3% vs ≤55% long-only cap) and event-driven (a genuinely new tag family for the portfolio). Only 2 tunable parameters to resist the overfit graveyard (338 deaths); the event list is fixed, not fitted.

Implements the hypothesis directly and avoids every named failure mode: one instrument, one venue, pure OHLCV plus an embedded calendar, so there is no supplementary feed to inject, no extra leg, no basket, and no funding/OI gate. (1) Calendar correctness was the main implementation risk and was handled explicitly: dates are stored as civil dates and converted with pure integer arithmetic (Hinnant days-from-civil plus the second-Sunday-March / first-Sunday-November DST rule), so no datetime import is needed; I verified the conversion against zoneinfo for eight sample releases across both DST halves and all of them matched exactly, and because CPI/NFP release at 08:30 and FOMC at 14:00 every event lands on a :00 or :30 boundary that coincides with a 15-minute bar edge. Expanding the calendar against the real ETHUSDT 15m catalog matched 213 decision bars over 2020-2026, confirming the alignment end to end. (2) No lookahead: the anchor is the close of the bar ending AT the release instant and the decision is taken at the close of the bar impulse_bars later, so only post-release information is used and the entry executes at that same moment. (3) Continuous signal: calculate_signal returns (close - close[-impulse_bars]) / ATR on EVERY bar, which on a decision bar happens to equal the event impulse; the event gate lives in should_enter, so the signal series never freezes. (4) Sample and fee viability, measured on real data before submitting rather than assumed: at the defaults the rule fires 109 times over the full 2020-2026 history and 13-16 times per year, and a crude replay of entry-to-trailing-stop net of 0.10% round-trip fees gives roughly +0.1% to +0.2% average per trade. That clears the futures fee floor only marginally, and with a per-trade standard deviation near 2.5% the standard error on the mean is about 0.24%, so the edge is NOT yet statistically separable from zero at this sample size; I am flagging that up front because it is the honest read and because it is intrinsic to a ~30-events-per-year design, not a coding choice. For the same reason I deliberately set trail_atr_mult to the middle of its range (5.0) rather than the value that maximized my sweep, since the sweep differences are well inside noise and picking the argmax would be in-sample fitting of the one parameter that matters most. (5) Risk: sizing risks ~0.8% of equity across the trailing-stop distance with gross notional capped at 50% of equity, leverage 1.0 and read in position_size, one position at a time, and a hard 8-hour time stop so the book is flat intraday. (6) Per-bar cost is O(1): the calendar becomes a set of integer timestamps at start-up (constant-time membership test), the ATR is an incremental bounded deque, and nothing rescans history, so 15-minute bars over the full history stay well inside the wall-clock budget.

event_impulse_no_edge_subfloor_perTrade: ETH scheduled-macro-event impulse-continuation shows no risk-adjusted edge over a well-measured 108-trade / 6.7-year sample — Sharpe -0.001 with CI [-0.66, 0.63] straddling zero, profit_factor 1.094, information_ratio -0.79, win_rate 0.50. avg_trade_return_pct 0.1455% is below the 0.15% futures fee floor (L22), so the per-trade edge is inside costs, and modeled market impact (impact_cost_pct 37.86%) consumes another ~38% of gross PnL. Returns are outlier-driven (kurtosis 58.5) with a -5.8% 2022. The design is novel and cleanly built, but the impulse-continuation premise is empirically refuted net of costs; with only 2 tunable params the base Sharpe is the ceiling (L63) and a ~225-trial optimization on 108 trades would curve-fit noise into a DSR death rather than reach the promotion floor. Not worth the 2-hour run.

Event-driven long-short intraday strategy on ETHUSDT.BINANCE USD-M 15-minute bars. It is flat on every bar except around scheduled US macro releases from an EMBEDDED calendar: CPI and Non-Farm Payrolls at 08:30 America/New_York and FOMC decisions at 14:00 America/New_York (NFP generated by the exact BLS first-Friday rule; CPI and FOMC from stored release-date lists; all converted to UTC nanoseconds with an integer days-from-civil routine plus the US DST rule, verified against zoneinfo). At the close of the impulse window (release + impulse_bars 15-minute bars) it measures the reaction as (close - pre-release anchor close) / ATR and, if the magnitude clears entry_atr_mult, enters in the DIRECTION of that impulse to ride the intraday continuation. Exits are an ATR trailing stop at trail_atr_mult x ATR from the best price since entry, or an 8-hour intraday time stop. Sizing is risk-first (equity * risk_pct / trailing-stop distance) capped at 50% of equity notional. Exactly two tunable parameters; the calendar, impulse window, ATR period and risk settings are fixed construction.

Low statistical power / small sample — the central risk, honestly flagged by the developer. The event gate fires ~13x/year (13 trades over the 363-day sandbox; ~109 over the full 2020-2026 history). Sandbox Sharpe 0.588 with CI [-0.991, 2.116] straddling zero, return_kurtosis 90.97, and avg_loss ($712) > avg_win ($598); +expectancy rests on a 0.69 win rate over 13 trades. Per-trade edge (~0.1-0.2% net) is not statistically separable from zero at this cadence. The full-history holdout will carry only ~17-20 trades — clears the >=10 gate but barely.

NFP generated by the exact first-Friday rule is right in most months but has rare holiday-shift/second-Friday exceptions, so a few NFP timestamps across 2020-2026 may be off by a week. Cannot introduce lookahead and affects at most a few of ~30 NFP events; only slightly dilutes signal purity. Optionally use a stored BLS release list as done for CPI/FOMC.

Genuinely novel class — scheduled-macro-event, long-short, event-driven; not in any saturated dead bucket (OHLCV trend, funding/OI gates, rotation baskets)

Fully backtestable with an embedded event calendar — no un-injectable supplementary feed

Disciplined 2-parameter surface, flat between events (low fee drag by design)

Well-measured sample: 108 trades over 6.7 years — the no-edge conclusion is statistically supported, not noise

Sharpe -0.001 with 95% CI [-0.66, 0.63] straddling zero — no risk-adjusted edge before optimization

avg_trade_return_pct 0.1455% is below the 0.15% futures fee floor (L22) — edge sits inside trading costs

profit_factor 1.094 (marginal, <1.2); information_ratio -0.79; win_rate exactly 0.50

impact_cost_pct 37.86% — market impact eats ~38% of gross PnL

Returns are outlier-driven (kurtosis 58.5, skew 1.85) with losing years (2022 -5.8%)

abandoned

EthMacroEventImpulseContinuationLS opened a genuinely new front for the factory: a scheduled-macro-event, long-short strategy that traded only around CPI, FOMC, and NFP releases using a fully embedded, backtestable event calendar — deliberately outside every saturated dead class and requiring no un-injectable feed. The design was disciplined (2 parameters, flat between events, low fee drag) and the sample was well-measured, but the edge simply was not there: over 108 trades and 6.7 years the Sharpe was -0.001 with a CI straddling zero, the per-trade return of 0.1455% sat below the futures fee floor, and modeled market impact consumed ~38% of gross PnL, with the thin headline return driven by a few outlier days. The analyst abandoned it at the backtest-review gate, concluding the impulse-continuation premise was refuted net of costs and that two parameters left nothing to tune toward the promotion floor — a clean, novel idea that the data did not support.

A genuinely novel, cleanly-built event-driven class is not automatically an edge — the intuitive premise that ETH's initial macro-release impulse continues intraday did not survive net of fees and market impact, and with the per-trade capture sitting inside the cost floor a 2-parameter design has no room to be tuned into viability.

The analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate: over a statistically adequate sample the impulse-continuation premise was empirically refuted net of costs — Sharpe indistinguishable from zero, per-trade edge inside the fee floor, and ~38% of gross PnL eaten by market impact. With only 2 tunable parameters the base Sharpe was the ceiling, so optimization on 108 trades would curve-fit noise into a DSR death rather than reach the promotion floor. It never reached optimization, post-opt analyst review, or risk review.

An event-driven, long-short, single-instrument intraday strategy on ETHUSDT Binance USD-M perp (15m bars) that stayed flat except in a narrow window around scheduled US macro releases (CPI, FOMC, NFP), taking the sign of ETH's initial impulse over the first few post-release bars and riding the intraday continuation with an ATR trailing stop and time stop, using a fully embedded (backtestable, no supplementary feed) event calendar and only 2 tunable parameters.

Over a well-measured 108-trade sample across 6.7 years (61 long / 47 short, exactly 50% win rate) it showed no risk-adjusted edge: total return 3.19%, Sharpe -0.001 (95% CI [-0.66, 0.63]), profit factor 1.094, information ratio -0.79, avg_trade_return_pct 0.1455% (below the 0.15% futures fee floor), with impact_cost_pct of 37.86%, outlier-driven returns (kurtosis 58.5, skew 1.85), and a losing 2022 (-5.8%).