EthPerpDailyFundingConfirmedTrendConfluenceLS

ETH Perp Daily Funding-Confirmed Trend (Single-Instrument USD-M, Long-Short, Carry-Momentum Confluence, Flat on Disagreement)

A long-SHORT, SINGLE-INSTRUMENT, single-venue directional strategy on ETHUSDT.BINANCE USD-M perpetual futures using DAILY bars, that takes a position ONLY when price trend and the funding rate AGREE in direction (funding used as a confirmation SIGNAL, never credited as a cash flow — P&L is fully price-based and creditable). Thesis: pure price-trend strategies overfit and decay because they trade EVERY trend, including spot-driven or short-covering moves that are not backed by leveraged-positioning demand and therefore reverse. Funding is a structural readout of that demand: a price uptrend CONFIRMED by positive funding (longs willing to pay to hold) is a genuine, demand-backed trend likely to persist, whereas a price uptrend with negative/flat funding is a hollow rally to avoid. The strategy goes long only when price momentum is up AND funding is positive, short only when price momentum is down AND funding is negative, and stays FLAT whenever the two disagree. This AND-intersection is a fundamentally different, more selective, lower-exposure return stream than a price-only trend or a funding-only regime signal, and the flat-on-disagreement behavior de-correlates it from the portfolio's existing long-biased trend book (it can also go net short in funding-confirmed downtrends). It uses ONLY the two feeds confirmed to span the full history (OHLCV + funding) — sidestepping the recurring supplementary-data-coverage wall that rendered the OI, liquidation, and options-IV strategies untestable this session — on the clean single-instrument USD-M MARGIN path. It avoids every other structural blocker: NOT BINANCE_SPOT/CASH (metric/fill bug), NOT COIN-M (not backtestable), NOT cross-venue/Hyperliquid (basis artifact), NOT a multi-instrument pair (extra-leg feed bug), NOT funding-carry (no uncreditable cash flow). It is distinct from all four pending funding strategies (momentum-follow, crowding-veto, settlement-clock, regime-persistence-funding-sign-alone) by requiring CONFLUENCE of price and funding. Kept to 3 parameters and low turnover to resist the overfitting that killed the pure-trend strategies.

Implements the AND-confluence thesis directly: a price uptrend confirmed by positive funding (real leveraged-long demand) is a demand-backed trend to ride, while an uptrend on negative/flat funding is a hollow rally to skip — and symmetrically for downtrends. The funding series is parsed once into a sorted (ts,rate) index and queried with bisect (O(log n)/bar, timeout-safe) using the documented dual-key {ns:rate} format; only the funding SIGN over a trailing window is used, never added to returns, so PnL stays creditable. Flat-on-disagreement (both entry gate and exit) yields a selective, lower-exposure, de-correlated stream versus the long-biased trend book and lets it go net short in funding-confirmed downtrends. Stays on the clean OHLCV+funding feeds (both dense across ETH history) and the single-instrument USD-M MARGIN path, avoiding spot/COIN-M/cross-venue/multi-leg/funding-carry blockers. Leverage is left at 1.0 (the MARGIN account allows the short leg at 1x without consuming margin), so sizing references no leverage and avoids the inert-leverage gate; kept to 3 core parameters with light validity-only sanitization (no plateau clamps) to resist the overfitting that decayed the pure-trend strategies.

Failed the decisive multiple-testing and out-of-sample gates, and optimization did not rescue it. Deflated Sharpe 0.1596 (far below 0.95), is_significant=FALSE, PBO 0.7883, and the optimized Sharpe 0.468 is BELOW the expected-max-under-noise bar 0.842 (indistinguishable from best-of-225 luck); sharpe_ci_low -0.37 straddles 0. Walk-forward is_overfitted=TRUE (avg IS 0.65 -> avg OOS 0.199, recent window NEGATIVE -0.77), and optimization left the Sharpe flat (0.49->0.47) with recent-year decay (2025 -19.6%). The headline is ~46% unrealized (end_unrealized_pct 124.9 of 269.6% total) and the book is substantially bull-beta (beta 0.24, benchmark_correlation 0.67), so the funding-confirmation filter adds little deflation-surviving alpha. The 'holdout passed' (ratio 1.247) is a red herring — it is a ratio of two near-zero Sharpes (0.25 vs 0.20), not a real edge. NOT iterate: this is the opposite of the ADA/SOL siblings I iterated (those cleared the luck bar with all-positive OOS and just needed parameter-surface shrinkage). Here DSR is 0.16, the Sharpe is BELOW the luck bar, the recent OOS window is negative, PBO is 0.79, and the strategy is ALREADY a 5-parameter low-surface model with sensitivity passing (0 cliffs) — there is no oversized search to shrink and nothing to tune toward; the edge is genuinely marginal, not over-searched. NOT revise_hypothesis: the funding-confirmation-of-trend filter does not generate a real edge beyond ETH bull-beta (corr 0.67) and decays out-of-sample; there is no promoted sibling of this exact mechanism to redirect, so this is a weak-edge overfit, not a proven mechanism stranded on a dead target. FAILURE PATTERN: single-asset daily 'funding-confirmed trend' (price-trend AND funding-sign confluence) on ETH is mostly bull-beta with a thin, fat-tail-driven edge (base Sharpe 0.49, kurtosis 254, ~46% of headline unrealized) that fails deflation (DSR 0.16), shows is_overfitted=TRUE with a negative recent OOS window, and does not improve under optimization; the funding filter does not convert decaying single-asset trend into a deflation-surviving alpha.

Single-instrument, long-short directional strategy on ETHUSDT.BINANCE USD-M perpetual (daily bars) that opens a position ONLY when price trend and funding-rate sign AGREE: long when close>SMA(trend_period) AND trailing funding>0, short when close<SMA AND trailing funding<0, and FLAT whenever they disagree. Funding is a confirmation signal only — never credited as cash flow, so P&L is pure price. Exits on confluence-break (price/funding stop agreeing) or a loose 15% safety stop. Three core parameters, low turnover, leverage 1.0.

Trades match the hypothesis: selective funding-confirmed long-short (80 long / 22 short), 102/102 submitted, ~12-day holding; funding used as signal only (P&L price-based), OHLCV+funding only (dense feeds), clean single-instrument USD-M path

Differentiated, lower-exposure return stream with a genuine short side; positive alpha (0.084), profit_factor 1.45, asymmetric wins (avg_win/avg_loss 3.6:1)

Adequate sample (102 trades, ~6yr) with low parameter count (3 core)

Weak risk-adjusted edge: Sharpe 0.49 with sharpe_ci_low -0.40 (CI straddles 0), vol 71.5%, max_dd 31% — high DSR/PBO failure risk after 225-trial selection

Headline inflated: end_unrealized_pct 188 of 475 total (~40% is an un-exited open ETH long); judge on realized ~287%

Extreme outlier concentration (kurtosis 254, skew 13.7) — record carried by a few large trade-close PnLs

Bull-concentrated with recent decay: 2022 -17%, 2023 -10%, 2026 -13%; benchmark_correlation 0.68 (confirm funding filter beats price-only trend); avg_position_pct 73% vs intended 50% (verify not sizing inflation)

marginal

80L/22S

has short side

Clean execution: genuine long/short with a real short leg (126 long / 19 short optimized), funding used as signal only (P&L price-based), OHLCV+funding only, no artifacts (metrics_reliable=true)

Sensitivity passes (0 cliffs) — the 5-parameter surface is internally smooth

Holdout ratio technically passes (1.247) — but on near-zero Sharpes (see weaknesses)

Deflated Sharpe 0.1596 (<<0.95), is_significant=FALSE, PBO 0.7883; optimized Sharpe 0.468 BELOW the expected-max-under-noise bar 0.842; sharpe_ci_low -0.37 (straddles 0)

Walk-forward is_overfitted=TRUE: avg IS 0.65 -> avg OOS 0.199 with a NEGATIVE recent window (-0.77)

Optimization did not improve it (Sharpe 0.49->0.47); recent-year decay (2025 -19.6%, 2026 -0.1%)

Headline ~46% unrealized (end_unrealized_pct 124.9 of 269.6% total); substantially bull-beta (beta 0.24, benchmark_correlation 0.67) — funding filter adds little real alpha

'Holdout passed' is a ratio of two near-zero Sharpes (0.25 vs 0.20), not evidence of a real edge

artifact of near-zero Sharpes

0.468 vs 0.842

> expected-max

EthPerpDailyFundingConfirmedTrendConfluenceLS proposed that funding sign confirms demand-backed trends, taking ETH perp positions only on price-and-funding confluence and staying flat on disagreement — a selective, dense-feed, low-parameter design that earned an optimize verdict on a 1.45 profit factor and a genuine short side. But it failed the generalization gates after optimization: deflated Sharpe 0.16, PBO 0.79, an optimized Sharpe (0.47) below the luck bar, and a negative recent out-of-sample window, with ~46% of the headline unrealized and a 0.67 benchmark correlation revealing it as mostly bull-beta. The analyst noted the 'holdout passed' was illusory (a ratio of two near-zero Sharpes) and distinguished this from the iterable ADA/SOL siblings: here the model is already low-surface with passing sensitivity, so there is no over-search to shrink and the edge is simply marginal. It was abandoned as a weak-edge overfit — the funding-confirmation idea adds little alpha beyond decaying ETH trend, with no promoted sibling to redirect toward.

A funding-confirmation filter on single-asset ETH trend does not convert decaying, bull-beta momentum into real alpha — the edge is genuinely marginal (Sharpe below the luck bar, DSR 0.16, negative recent OOS) rather than merely over-searched, so a low-parameter surface with passing sensitivity cannot be salvaged by iteration; and a holdout 'pass' built on two near-zero Sharpes is not evidence of an edge.

After optimization the analyst abandoned it for failing the decisive multiple-testing and out-of-sample gates: deflated Sharpe 0.16, PBO 0.79, optimized Sharpe 0.468 below the expected-max-luck bar of 0.842, sharpe_ci_low -0.37, walk-forward is_overfitted=TRUE (avg IS 0.65 → avg OOS 0.199 with a negative recent window -0.77) — and the 'holdout passed' was a red herring (a ratio of two near-zero Sharpes), with the book substantially bull-beta so the funding filter added little deflation-surviving alpha.

A long/short, single-instrument daily strategy on ETHUSDT.BINANCE USD-M perp that takes a position only when price trend and funding-rate sign AGREE — long when momentum is up and funding positive, short when down and negative, flat on disagreement — using funding purely as a confirmation signal (never credited as cash flow) on the dense OHLCV+funding feeds, with 3 core parameters to resist overfitting.

The base backtest showed a headline +475% (but ~40% unrealized open ETH long; realized ~287%) at a weak Sharpe 0.49 with sharpe_ci_low -0.40, profit factor 1.45, ~3.6:1 asymmetry over 102 trades, 30.8% drawdown, extreme fat tails (kurtosis 254), and substantial bull-beta (correlation 0.68); optimization left it flat (Sharpe 0.468, +269.6%, ~46% unrealized).