Strategies

ADA 4H Parabolic SAR Trend-Following Long-Only on ADAUSDT.BINANCE

Outcome: Abandoned

AdaFourHourParabolicSarTrendFollowingLong

Outcome Summary

AdaFourHourParabolicSarTrendFollowingLong introduced the portfolio's first Parabolic SAR mechanism — an accelerating stop-and-reverse trend-follower — long-only to match the proven directional asymmetry on the reliably testable ADA 4H path. The mechanism wasn't broken: it was slightly positive gross (PF 1.059, +31.6%), but the edge evaporated after costs to a 0.18 Sharpe with a zero-straddling CI and ~0 alpha. The decisive problem was structural and unfixable: SAR flips constantly, generating 546 trades whose thin per-trade edge is dominated by 28.6%-of-gross market impact at a deployable-infeasible $1.2M capacity, with the small total front-loaded in 2020-2021 and recent years negative. The reviewer abandoned it pre-optimization as a picking-up-pennies churn system that no parameter change can rescue, noting only lower-frequency, positive-base-edge ADA 4H mechanisms survive trading costs.

Strategy report

Hypothesis

A long-only, single-instrument, single-timeframe Parabolic SAR (Stop-And-Reverse) trend-following strategy on ADAUSDT.BINANCE — confirmed testable per the FULL series of 9+ recent successful ADA-4H single-instrument pipeline additions (ADADonchianBreakoutLong4HOnly, ADAWeekendCalendarEffectLong, ADAVolContractionSqueezeLong, ADABullishRSIDivergenceLong, ADAMACDBullishCrossoverWithHistogramAccelerationLong, ADABullishOBVDivergenceLong all backtested cleanly on ADA-4H bars alone — establishing this as THE most reliable testable single-instrument setup this session). Structurally distinct from every prior ADA pipeline strategy: those use static price-channels (Donchian), calendar timing, vol-regime detection, mean reversion (BB), price-momentum divergence (RSI), EMA-gap crossover (MACD), or cumulative-volume divergence (OBV). This is the FIRST PARABOLIC SAR strategy in the portfolio. Parabolic SAR (Wilder 1978 'New Concepts in Technical Trading Systems' — same source as RSI but a fundamentally different indicator class) is a NON-LINEAR ACCELERATING trailing-stop system: it sets a stop-and-reverse point at each bar, accelerating closer to price as the trend extends. The 'AF' (acceleration factor) starts at 0.02 and increases by 0.02 on each new high (capped at 0.20). Entry: when SAR flips below price (trend reversal to up), GO LONG. Exit: when SAR flips above price (trend reversal to down), CLOSE. This is purely LONG-ONLY directional momentum — directly responding to the recent failure log evidence that short-only directional on crypto has negative EV; we capture ONLY the long-side asymmetry. The mechanism is structurally distinct from Donchian breakout (which uses static N-bar high/low) because SAR's acceleration factor adapts to recent move magnitude, producing tighter exits in fast-moving trends and looser exits in slow-grinding trends — adaptive without explicit parameter selection. Documented edge: SAR is one of the oldest systematic trend-following indicators (Wilder 1978), validated in equities (Pring 1991, Murphy 1999) and crypto-specific (Day-Glass-Wang 2020). Uses ONLY ADAUSDT.BINANCE 4H bars. Fills critical gaps: (1) FIRST PARABOLIC SAR strategy in portfolio (entirely new signal class — non-linear adaptive trailing-stop), (2) Long-only matching the proven directional asymmetry, (3) Single-instrument simplicity, (4) Confirmed-testable instrument. Position sizing: 18% per trade. Risk per trade: ~1.5% capped by SAR-driven stop.

Iteration 2 fixes the Layer-2 TypeError with the smallest possible change and no regression. Root cause: `_ohlc` was declared `@staticmethod` and called as `self._ohlc(bar)`, but the Layer-2 verification proxy does not preserve the staticmethod wrapper when it executes the strategy, so `self` was passed as an extra positional argument ('takes 1 positional argument but 2 were given') on the very first bar of every scenario. Converting `_ohlc` to a normal instance method (`def _ohlc(self, b)`) makes the call signature correct regardless of whether the decorator is honored, and every call site already uses `self._ohlc(...)`. I verified the fix by replaying the actual synthetic steady_uptrend bars through calculate_signal: 60 bars process without error and produce 59 unique (non-frozen) signal values. The full SAR initialization/advance logic, the signed price-vs-SAR signal, entry/exit thresholds, risk-capped sizing, imports, and parameters are all unchanged, so Layer 1 (static, passes clean) stays green and the strategy now runs through Layer 2.

Marginal pre-cost edge that is fee/impact-fragile, uncapacious, and decaying -- not worth optimizing. profit_factor 1.059, Sharpe 0.18 (CI -0.69 to +0.90, straddles 0), expectancy only +30/trade, alpha ~0, CAGR 2.45%. The decisive problem is trading-cost fragility: impact_cost_pct 28.6% of gross (past the ~25% red line), capacity_usd only $1.2M (edge exists only at toy scale; negative at any deployable size), plus 4.24% commission. This is structural to Parabolic SAR: it flips constantly, producing 546 trades with a razor-thin per-trade edge that costs dominate (picking-up-pennies), and the per-trade return is below realistic trading costs at scale -- a profile no parameter change fixes (SAR's stop-and-reverse inherently churns; even turning on the trend filter or widening af cannot remove 28.6% market impact). The edge is also decaying: recent years are negative (2024 -3.9%, 2025 -0.06%, 2026 -7.7%) with the small positive total front-loaded in 2020-2021. A 0.18 Sharpe / 1.06 PF base would be obliterated by deflated-Sharpe after best-of-225 selection. Do not spend optimization compute. FAILURE PATTERN: high-flip-frequency trailing-stop trend systems (Parabolic SAR) on mid-cap alt 4H generate hundreds of trades with a sub-cost per-trade edge -- gross PF barely exceeds 1, but market impact (28.6% of gross) and a toy capacity ($1.2M) consume the edge at any real size; only lower-frequency, positive-base-edge ADA 4H mechanisms clear costs.

Implementation

Long-only Parabolic SAR (Wilder 1978) trend follower on ADAUSDT.BINANCE 4H bars, pure OHLCV. Each bar the SAR trailing stop accelerates toward price (AF 0.02→0.20 on new extremes); the strategy holds long while SAR is below price (uptrend) and exits the moment SAR flips above price (reversal). Position size is 18% of equity notional, capped so the distance-to-SAR risk stays within ~1.5% of equity. Single instrument, single timeframe, leverage 1.0.

Verification Results

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['af_max', 'af_step', 'risk_pct', 'min_notional', 'min_stop_pct', 'notional_pct', 'trend_period', 'exit_threshold', 'entry_threshold', 'use_trend_filter'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: TypeError: AdaFourHourParabolicSarTrendFollowingLong._ohlc() takes 1 positional argument but 2 were given (bar timestamp: 1735689840000) - steady_downtrend: TypeError: AdaFourHourParabolicSarTrendFollowingLong._ohlc() takes 1 positional argument but 2 were given (bar timestamp: 1735689840000) - flat_ranging: TypeError: AdaFourHourParabolicSarTrendFollowingLong._ohlc() takes 1 positional argument but 2 were given (bar timestamp: 1735689840000) - volatility_spike: TypeError: AdaFourHourParabolicSarTrendFollowingLong._ohlc() takes 1 positional argument but 2 were given (bar timestamp: 1735689840000) - zero_volume: TypeError: AdaFourHourParabolicSarTrendFollowingLong._ohlc() takes 1 positional argument but 2 were given (bar timestamp: 1735689840000) - price_gap: TypeError: AdaFourHourParabolicSarTrendFollowingLong._ohlc() takes 1 positional argument but 2 were given (bar timestamp: 1735689840000)

Backtest Review

Sharpe: 0.18
Total return: 31.60%
Max drawdown: 20.85%
Trades: 546
Win rate: 39.2%
Profit factor: 1.06

Clean OHLCV-only SAR implementation, correct long-only routing, adequate sample (546 trades), modest max_drawdown 20.9%, 0 dropped

Slightly positive gross (PF 1.059, total +31.6%) — the mechanism isn't catastrophically broken

Edge essentially gone after costs: Sharpe 0.18 (CI -0.69 to +0.90), expectancy +30/trade, alpha ~0

Fee/impact-fragile and uncapacious: impact_cost_pct 28.6% of gross (>25% red line), capacity_usd only $1.2M, commission 4.24% of gross

Structural high churn: SAR flips constantly -> 546 trades with razor-thin per-trade edge that trading costs dominate (picking-up-pennies)

Decaying edge: recent years negative (2024 -3.9%, 2025 -0.06%, 2026 -7.7%); positive total front-loaded in 2020-2021

deployable scale

Outcome Summary

High-flip-frequency trailing-stop trend systems like Parabolic SAR on a mid-cap alt 4H generate hundreds of trades with a sub-cost per-trade edge — gross profit factor barely exceeds 1, but market impact (28.6% of gross) and toy capacity ($1.2M) consume the edge at any real size — a picking-up-pennies failure that no parameter tuning fixes; only lower-frequency, positive-base-edge mechanisms clear costs.

The analyst abandoned it at the backtest-review gate as a marginal, fee/impact-fragile, uncapacious, decaying edge not worth optimizing: Parabolic SAR's constant flipping produced 546 trades with a razor-thin per-trade edge that market impact (28.6% of gross) dominates, the edge exists only at toy scale ($1.2M capacity) and front-loaded in 2020-2021 — a structural churn problem no parameter change can fix.

A long-only, single-instrument Parabolic SAR (Wilder 1978 stop-and-reverse) trend-follower on ADAUSDT.BINANCE 4H — the portfolio's first SAR strategy — going long when SAR flips below price (uptrend) and exiting when it flips above (downtrend), with the accelerating SAR itself as the trailing stop, ~18% notional per trade and leverage 1.0.

It had a marginal pre-cost edge that costs erased: +31.6% total return but CAGR just 2.45%, Sharpe 0.18 (CI -0.69 to +0.90), profit factor 1.059, +$30/trade expectancy, alpha ~0, over 546 trades at a 39.2% win rate — with impact_cost_pct of 28.6% of gross, capacity only $1.2M, commission 4.24% of gross, and recent years negative (2024 -3.9%, 2025 -0.06%, 2026 -7.7%).

Backtest and paper results are hypothetical. Trading involves risk of loss.