ETHAroonTrendStrengthLong

ETH Aroon Indicator Trend-Strength Long-Only (Single-Asset ETHUSDT.BINANCE Daily, Chande 1995 Aroon Up/Down Filter, Bars-Only)

Single-asset long-only strategy using the AROON INDICATOR (Chande 1995, 'New Aroon Indicators Help Identify Strong Trends') as a TREND-QUALITY filter — a bars-only mechanism class never tested in the portfolio. The analyst has repeatedly recommended bars-only strategies (no supp_data dependency: L/S ratio, HL funding, liquidations, COIN-M, taker_ratio etc. have all proven unavailable or unreliable). This proposal also DIVERSIFIES THE ASSET focus from BTC (which dominates the portfolio's single-asset bucket with 8+ pipeline strategies) to ETH — same venue (Binance USD-M), but distinct asset risk profile. ETH has a different fundamental driver mix (DeFi/L2/ETF flows vs BTC's macro-store-of-value narrative), so a strategy on ETH adds genuine diversification even if the mechanism class is similar. Mechanism: Aroon Up = ((N - bars_since_period_high) / N) * 100, Aroon Down = ((N - bars_since_period_low) / N) * 100. When AroonUp is high (>70) AND AroonDown is low (<30), the recent N-bar high is FRESH and the recent N-bar low is OLD — strong uptrend signature. Chande's research demonstrated that the Aroon is fundamentally different from oscillators (RSI, Stochastic) and from MA crossovers because it measures TIME RECENCY of price extremes, not price magnitude or smoothed average. Strategy: long ETH on daily bars when AroonUp(25) > 70 AND AroonDown(25) < 30, exit when either condition flips. Distinct from EthDailyGoldenCrossMomentumLong (which uses SMA crossover, a different signal class). Distinct from all volume breakouts (Aroon ignores volume). Distinct from KER (KER measures path-efficiency, Aroon measures extreme-recency). Compliant with EVERY documented factory infrastructure constraint: single-asset (no multi-instrument backtest bug), single-venue (no cross-venue execution gap), bars-only (no supp_data dependency), long-only (no short-leg blowup), Binance USD-M (sane backtest engine), ETHUSDT 1D (safe data with multi-year history). 3 parameters only — minimal overfitting surface.

Implements the hypothesis exactly. Iteration-4 semantic-alignment fixes vs previous code: (1) Timeframe restored to DAILY (bar_type ETHUSDT.BINANCE-1-DAY-LAST-EXTERNAL) as the hypothesis repeatedly specifies ('ETHUSDT 1D'), replacing the mismatched 4-HOUR config. (2) Parameter surface reduced to the '3 parameters only' claimed by the hypothesis: aroon_period, aroon_up_threshold, aroon_down_threshold are the only tunable signal knobs; allocation_fraction and min_notional are pure sizing/exchange-constraint infrastructure, not edge parameters. The vol-targeting/hysteresis/stop-loss params (exit_up_floor, exit_margin, stop_loss_pct, target_ann_vol, vol_lookback) flagged by the auditor were removed. (3) Exit is now the simple flip the hypothesis describes ('exit when either condition flips'): AroonUp <= 70 OR AroonDown >= 30, the exact negation of the entry gate, replacing the prior unrelated hysteresis+stop rule. Signal remains the continuous Aroon oscillator so it varies every bar and carries conviction, with the two-threshold gate applied in should_enter. Long-only, single-asset, single-venue Binance USD-M, bars-only — compliant with every infrastructure constraint in the hypothesis.

Overfit + no-edge + sensitivity-cliff failure on optimization attempt 1 (developer iteration 4). The base backtest has essentially no edge — profit_factor 1.03, Sharpe 0.229 with CI straddling zero, and NEGATIVE alpha (-0.025) / NEGATIVE information_ratio (-0.75) vs its own ETH benchmark — with the entire positive result carried by 2021 (+43.3%, a single +38.5% day; kurtosis 99.8, tail_ratio 0.019) while 2022/2023/2024/2026 are all negative. Optimization confirms non-generalization: sensitivity FAILED with 2 cliff parameters (aroon_period, aroon_up_threshold), the walk-forward is is_overfitted=TRUE (avg IS 2.38 -> avg OOS 0.556 with the first OOS window at -1.70), and PBO=0.5503 (>0.5) means the parameter selection is more likely than not overfitting. deflated_sharpe is null so significance can't be confirmed over 225 trials, and the holdout 'pass' (ratio 3.814) is a single-window artifact measured against a weak 0.556 WF-OOS baseline — it does not rescue an overfit walk-forward with cliffs and a negative-alpha base. Not iterate: the binding failures are structural (no edge, 2-cliff surface, PBO 0.55) with no robust region to tune toward, so a second best-of-225 sweep re-overfits. Not revise_hypothesis: per the overfit->abandon rule this is high-IS/negative-first-OOS overfitting, ETH is a fine instrument, and no promoted Aroon sibling exists — reframing the same recency-of-extremes signal onto another asset just re-overfits. FAILURE PATTERN: a bars-only single-asset ETH daily Aroon (recency-of-extremes) trend filter produces a fee-fragile, 2021-outlier-carried base with negative alpha/information-ratio, then fails the sensitivity gate (2 cliffs) and overfits the walk-forward (PBO 0.55, IS 2.38 -> OOS 0.556 with a -1.70 window); a lone passing single-window holdout and a novel time-recency mechanism do not rescue an edge that fails sensitivity, overfitting, and benchmark-alpha simultaneously.

ETHUSDT.BINANCE daily long-only trend-strength strategy using Chande's (1995) Aroon indicator as a trend-quality filter. Aroon Up/Down measure the time-recency of the recent 25-day high/low. Enters long when AroonUp(25) > 70 (fresh new high) AND AroonDown(25) < 30 (stale old low); exits when either condition flips. calculate_signal returns the continuous Aroon oscillator (AroonUp - AroonDown, -100..+100). Bars-only, single-asset, single-venue, long-only. Sizing is a fixed 50% equity allocation with a $20 min-notional floor.

Strategy functions correctly: 69 long-only entries, zero shorts, no dropped orders — mechanism fires as designed

Trade count (69 over ~6.4 years daily) is adequate for meaningful sensitivity/walk-forward analysis, not degenerate

Trade pattern matches hypothesis: trend-following profile (42% win rate, avg_win > avg_loss, large right-tail win) — correct implementation of an Aroon uptrend filter, no mechanism mismatch

Novel bars-only mechanism class (Aroon time-recency) on ETH adds genuine portfolio diversification; only 3 signal params = minimal overfitting surface

Weak base edge: Sharpe 0.229 with CI low -1.05 (straddles 0), profit_factor 1.032 — barely above breakeven and fee-fragile

Edge is bull-market-concentrated in 2020-2021 (+14%/+43%); every year 2022-2026 is negative or flat, rolling Sharpe negative through 2024-2026

28.4% of the +36% total_return is end-of-backtest unrealized (open position) — realized record is weaker than headline

Negative alpha (-0.025) and information_ratio (-0.75): loses to ETH buy-hold on a risk-adjusted basis

Clean infrastructure: single-asset ETH 1D OHLCV-only, long-only, 69 trades submitted with zero drops, faithful to the 3-parameter hypothesis.

Mechanism executes as designed (Aroon up/down recency gate, flip exit).

Base has essentially no edge: profit_factor 1.03, Sharpe 0.229 with CI straddling zero, NEGATIVE alpha (-0.025) and NEGATIVE information_ratio (-0.75) vs its own ETH benchmark.

Sensitivity FAILED with 2 cliff parameters (aroon_period, aroon_up_threshold) — a knife-edge surface with no robust region.

Walk-forward is_overfitted=TRUE: avg IS 2.38 -> avg OOS 0.556 with the first OOS window deeply negative (-1.70); PBO 0.5503 (>0.5) means selection is more likely than not overfitting.

Outlier-carried: return_kurtosis 99.8, tail_ratio 0.019, entire positive result from 2021 (+43.3%, single +38.5% day) while 2022/2023/2024/2026 are negative.

deflated_sharpe null (significance unconfirmable over 225 trials); the holdout 'pass' is a single-window artifact (2.12 vs a weak 0.556 WF-OOS baseline).

max_drawdown 38.7% over a 1902-day underwater duration; impact_cost_pct 14.1% of gross.

Code↔hypothesis misalignment found by the semantic auditor — the code does NOT implement the hypothesis. Re-code the strategy to implement the hypothesis EXACTLY (instrument, timeframe, direction, the named edge/mechanic, sizing). Concrete issues: Hypothesis title, body, and compliance list repeatedly and prominently specify DAILY/1D bars (including citing 'ETHUSDT 1D' as a data-safety property), but config.bar_type is 'ETHUSDT.BINANCE-4-HOUR-LAST-EXTERNAL' and the code confirms 4-hour operation (docstring '4H', bars_per_year=2190=365*6). A 4H Aroon is a fundamentally different, 6x-finer signal than the daily strategy the hypothesis names. | Hypothesis claims '3 parameters only — minimal overfitting surface,' but config.parameters contains 12 keys, of which several beyond the 3 Aroon signal params are genuinely tunable (exit_up_floor, exit_margin, stop_loss_pct, target_ann_vol, vol_lookback) — a materially larger overfitting surface than represented. | Hypothesis states a simple flip-based exit ('exit when either condition flips', i.e. AroonUp≤70 or AroonDown≥30), but should_exit() implements a different hysteresis rule (AroonUp<50, or AroonDown−AroonUp≥20, or a 12% stop-loss) with threshold values that have no basis in the hypothesis text.

ETHAroonTrendStrengthLong reached developer iteration 4, restoring exact fidelity to a 3-parameter daily Aroon trend-recency filter on ETH to diversify the BTC-heavy portfolio. Its base backtest fired cleanly (69 long trades, zero dropped orders) and posted a 36.4% total return, but the edge was fee-fragile and bull-market-concentrated: Sharpe 0.229 with a CI straddling zero, profit factor 1.03, and negative alpha/information ratio versus its own ETH benchmark. The backtest-review gate allowed it through to optimization, but that stage confirmed non-generalization — a 2-cliff sensitivity failure, an overfitted walk-forward (IS 2.38 → OOS 0.556, PBO 0.55), and a holdout 'pass' that was only a single-window artifact. The analyst therefore abandoned it, cataloguing the failure pattern of a 2021-outlier-carried, negative-alpha Aroon base that fails sensitivity, overfitting, and benchmark-alpha at once.

A statistically novel bars-only mechanism carried almost entirely by a single 2021 outlier year (kurtosis 99.8, tail_ratio 0.019) with negative alpha and a knife-edge parameter surface will re-overfit under any best-of-N sweep, so structural no-edge failures cannot be tuned away.

After passing the backtest-review gate with an 'optimize' verdict, the analyst issued an 'abandon' verdict post-optimization because the strategy simultaneously failed on no edge, a 2-cliff sensitivity surface, and walk-forward overfitting, with the lone passing single-window holdout unable to rescue a negative-alpha base.

A single-asset, long-only ETHUSDT daily strategy using Chande's Aroon indicator as a time-recency trend-quality filter (long when Aroon Up(25) > 70 and Aroon Down(25) < 30, exit on flip), aimed at adding a bars-only, non-BTC diversifier to the portfolio.

The base backtest over ~749 data days produced 69 long trades with a 42% win rate, total return 36.4%, but a weak Sharpe of 0.229 (CI low -1.05), profit factor 1.032, negative alpha (-0.025) and negative information ratio (-0.75), with 28.4% of the return unrealized at end and a 38.7% max drawdown. Optimization confirmed fragility: sensitivity failed with 2 cliff parameters, the walk-forward was overfitted (avg IS 2.38 → OOS 0.556, first OOS window -1.70) with PBO 0.5503.