Strategies

SOL USD-M Intraday Volatility-Expansion Breakout, Long-Short (Single Instrument SOLUSDT.BINANCE Perp, 15m Bars, Trade the CONTINUATION of a Genuine Range-Expansion Breakout — Bracketed ATR Target/Stop, Selective Firing, 3-Parameter)

Outcome: Abandoned

SolIntradayVolExpansionBreakoutLS

Outcome Summary

SolIntradayVolExpansionBreakoutLS was the third iteration of an OHLCV-only single-name intraday breakout idea, reworked to ride winners with a chandelier trailing stop after earlier versions showed a real but sub-fee continuation edge. Across a decisive 1,396 trades the edge remained negative — profit factor 0.73, -0.046% per trade, -$88.8 expectancy — and a broken sizing that ran position size to 132% of equity against a claimed 0.5x cap drove the account to liquidation on 2023-11-23, wiping out 100% of capital with impossible daily returns like -1820%. The analyst abandoned it at the pre-optimization gate, concluding that no parameter region flips the negative expectancy and fixing the sizing bug would only slow a guaranteed loss. Optimization, risk review, and later stages were never reached.

Strategy report

Hypothesis

A LONG-SHORT, single-instrument INTRADAY MOMENTUM-breakout on SOLUSDT.BINANCE (USD-M perp, 15m bars) that trades the CONTINUATION of genuine range-expansions — the deliberate OPPOSITE polarity of the intraday VWAP/z-score reversion FADES that keep dying (the ETH VWAP port confirmed negative on 6,130 trades; my OI fade confirmed negative on SOL). Fades of intraday deviations have no edge on crypto perps; this instead trades WITH the breakout, betting on short-horizon momentum continuation once volatility genuinely expands. It fills the under-represented SHORT horizon bucket (short_1m_15m 8.8% vs >=10% target) with a NON-fade mechanism, and is distinct from my pending daily/4h momentum strategies (TSMOM, breakout trend-ride) by operating on 15m intraday bars. It is NOT reversion (opposite sign), NOT a basket (single name — avoids L12 multi-leg fee death), NOT supplementary-fed (pure OHLCV — avoids the taker-flow timeout and OI/liquidation data walls), NOT options/carry. The key discipline against the fee_edge death that kills intraday strategies: it fires ONLY on a volatility-EXPANSION gate (few, high-conviction signals per day, ~1-3 max), targets a wide ~1.5x-ATR continuation move that clears the ~0.10% USD-M round-trip several times over, and uses a bracketed target/stop so losers are cut fast. SOL is chosen precisely because its large intraday % range makes breakout targets big relative to fees. Only 3 parameters; O(1) incremental rolling stats to avoid any smoke-test timeout.

QA's Layer-4 diagnosis was precise: the continuation edge is real but small (~+0.089%/trade gross), and it failed to clear the ~0.10% taker round-trip because the fixed 2.5-ATR target CAPPED winners while most breakouts reverted into the tight 1-ATR stop first (realized reward/risk only 1.66:1). A fixed target is structurally wrong for a breakout system whose edge lives in the fat right tail the hypothesis explicitly claims to monetize. The smallest change that attacks this exact weakness is an EXIT-only redesign: remove the fixed ATR target and ride the position with a chandelier trailing stop (exit trail_atr ATR back from the favourable extreme since entry), so the rare large continuations run uncapped and lift avg_win, while an initial protective stop still cuts losers and a time cap backstops. I widened the initial stop 1.0->1.5 ATR so the ordinary post-breakout wiggle stops out fewer would-be winners (directly addressing 'most revert into the 1-ATR stop'), and raised selectivity (expansion_mult 2.0->2.5, breakout_len 30->40) to cut turnover and thus fee drag on a per-trade edge that is fee-sensitive. The entry/expansion signal, the continuous breakout score, the iter-2 risk-control guards (flat-only entry, total-gross cap at 0.5x equity), and the sizing formula are all unchanged, so Layers 1-3 stay green and leverage remains 1.0 with sizing that never references it. The 3rd tuned param target_atr is repurposed as trail_atr (with a backward-compatible fallback read of 'target_atr'), preserving the 3-parameter design. This gives the mechanism its best structural shot at clearing fees by capturing the tail; if the trailing exit still cannot lift avg_trade_return above the futures floor, the mechanism is genuinely exhausted, but this is the correct, minimal fix for the specific capped-winner failure QA identified.

Account LIQUIDATED (2023-11-23): total_return -100%, max_drawdown 100%, and every post-blowup daily return is 0.0 (only ~3 of 6 years executed). The underlying edge is decisively negative on a well-sampled 1,396 trades — avg_trade_return_pct -0.046% (below the 0.10% USD-M round-trip), profit_factor 0.73, expectancy -$88.8/trade — so the range-expansion continuation mechanism has no fee-clearing edge on SOL 15m. A sizing/leverage bug compounds it (avg_position_pct 132% vs the claimed 0.5x cap; daily returns of -522% and -1820% as equity went negative and sizing exploded, return_kurtosis 1551), but fixing that cannot create edge — it would only slow a guaranteed loss. This is iteration 3 of the OHLCV-only single-name intraday breakout/fade class that has died repeatedly this session; iters 1-2 showed a sub-fee edge and iter-3's rework blew the account up. No parameter region flips negative expectancy and no reframe of intraday SOL continuation clears costs, so abandon rather than iterate or revise.

Implementation

Long-short intraday range-expansion breakout continuation on SOLUSDT.BINANCE (USD-M perp), 15m bars, pure OHLCV. Enters only while flat when a bar both breaks the prior 40-bar high/low AND expands (true range >= 2.5x ATR): close above prior high -> long, below prior low -> short. Exit is now a chandelier trailing stop that rides the winner into the fat tail (exit trail_atr=3.0 ATR back from the highest-high/lowest-low since entry), with an initial 1.5-ATR protective stop and a 64-bar (~16h) time cap; the fixed profit target was removed so continuations run uncapped. Sizing risks 2% of equity over the initial stop distance, bounded so total gross exposure never exceeds 0.5x equity (leverage 1.0). 3 tuned params (breakout_len, expansion_mult, trail_atr).

Verification Results

Verification failed (Layer 4 — QA review): - Confirmed high-turnover FEE-EDGE DEATH — the QA fee-viability block. On a well-sampled 532 trades (metrics_reliable=TRUE) the strategy nets avg_trade_return_pct -0.011% (NEGATIVE), profit_factor 0.745, total_return -40.9%, Sharpe -0.93. Arithmetic: net -0.011%/trade + ~0.10% USD-M taker round-trip (market entries AND exits) => gross edge ~+0.089%/trade — a real but TINY continuation edge that does NOT clear the ~0.10% round-trip. With turnover 333 it bleeds fees every rotation (avg_win 726 vs avg_loss 438 = only ~1.66:1 realized, short of the 2.5:1 target because most breakouts revert into the 1-ATR stop before the 2.5-ATR target). This iteration's explicit goal was to fix this (expansion_mult 1.5->2.0, breakout_len 20->30, target_atr 1.5->2.5, trades 1125->532), yet per-trade edge stayed negative. The 15m breakout-continuation mechanism on SOL is now empirically falsified across two iterations; the gross edge is structurally below fees.

Analyst: apply the 0.15% floor — 0.123% with PF 0.954 is below it and net-negative; abandon unless walk-forward robustly lifts avg_trade_return_pct above 0.15% OOS. Don't sink a full optimization into a breakeven mechanism after three iterations without real sensitivity headroom.

Edge is now above the QA round-trip bar but BELOW the analyst's futures floor and still a net loser in-sample — analyst should scrutinize and most likely ABANDON. avg_trade_return_pct +0.1233% clears the ~0.10% round-trip (iter-2 fee-death resolved), but profit_factor 0.954 (<1), total_return -4.16%, dollar expectancy -$17.65: equal-weighted % positive while dollar-weighted negative means the larger-sized trades are the losers. 0.123% is below the 0.15% futures floor. On 237 well-sampled trades this is marginal-to-negative, and by the developer's OWN exhaustion criterion the mechanism did not clear the floor.

Analyst/risk: weight impact heavily — $23k capacity with 206% impact is effectively un-deployable at size; a strong argument for abandonment over optimization.

SEVERE capacity/impact red flag: impact_cost_pct 206.3%, capacity_usd only $23,486, total_impact_usd $8,116 (impact_folded=TRUE). For liquid SOL a ~$23k capacity means the edge is so thin realistic impact erases it, and turnover 173 compounds it. Even if optimization nudges per-trade return above 0.15% frictionless, this impact/capacity profile suggests it won't survive real execution.

Backtest Review

Sharpe: 0.31
Total return: -100.00%
Max drawdown: 100.00%
Trades: 1396
Win rate: 32.0%
Profit factor: 0.73

Well-sampled (1,396 trades) so the verdict is statistically decisive, and trades match the hypothesized long/short expansion-breakout mechanism (680 long / 716 short).

Account LIQUIDATED on 2023-11-23 — total_return -100%, max_drawdown 100%; only ~3 of 6 years traded before the blowup, all post-liquidation returns are 0.0.

Negative per-trade expectancy on 1,396 trades: avg_trade_return_pct -0.046% (below the 0.10% round-trip), profit_factor 0.73, expectancy -$88.8/trade — no fee-clearing edge.

Position sizing broken: avg_position_pct 132% despite a claimed 0.5x cap; daily returns of -522% and -1820% show equity went negative and sizing exploded (kurtosis 1551, annual 2024 -34,591%).

Sharpe 0.31 with CI [-1.18, 0.71] straddling zero; information_ratio -1.67 — negative active value.

Third iteration of the OHLCV-only single-name intraday breakout/fade class that has died repeatedly; iter-3 rework made results catastrophically worse rather than better.

Iteration History

Verification failed (Layer 4 — QA review): - The account was LIQUIDATED (liquidated=true 2026-06-03, total_return -100.0, max_drawdown 100.0) with avg_position_pct 400.48% — a physics-level contradiction of the stated 'leverage 1.0, gross capped at 1x equity.' position_size() caps a SINGLE entry at 1x, but realized average gross is ~4x, so the per-entry cap is not bounding CUMULATIVE exposure. A genuine 1x long-short with a 1-ATR stop cannot lose 100% of the account, so the liquidation proves positions reach multiples of equity. Most likely cause: should_enter() returns a side purely from the cached break booleans with NO flat check, so consecutive same-direction 15m expansion-breakouts PYRAMID the position (each add ~1x → ~4x gross); the bracket anchor _entry_px is seeded only once via _was_in_pos, so added lots are never re-anchored and the stop/target no longer bound true cost. On a reversal the loss is unbounded relative to the 1-ATR stop and the account is wiped. This is a risk-control defect, not a negative edge. - Fee/edge death — the exact failure the hypothesis claimed to avoid. avg_trade_return_pct 0.051% over 1,125 trades (turnover 1312.9) is BELOW the ~0.10% USD-M round-trip and far below the 0.15% futures floor: profit_factor 0.76, win_rate 0.43, avg_win 756.7 ≈ avg_loss 758.7 (no reward/risk edge), information_ratio -2.07. The realized per-trade move does not clear fees, so the range-expansion-continuation edge is absent at this configuration on SOL 15m. Even with the sizing bug fixed, 0.05%/trade gross across 1,125 fee-paying trades is a structural loser. - Catastrophic tail (return_skew -13.27, kurtosis 231.2, largest_loss -6261.8 vs largest_win 5238.2) is consistent with the over-sizing/stacking diagnosis: a few stacked-position reversals dominate P&L and drove the liquidation, confirming a risk-control/sizing defect rather than ordinary breakout underperformance.

Outcome Summary

An intraday OHLCV-only single-name breakout-continuation mechanism has no fee-clearing edge on SOL 15m, and a position-sizing cap that isn't actually enforced can turn a losing strategy into an account-liquidating blowup — verify realized avg_position_pct against the intended cap before trusting any result.

The backtest-review analyst issued an abandon verdict before optimization: the account blew up mid-backtest (liquidated, -100% return, 100% drawdown), the underlying continuation edge was negative and fee-losing across 1,396 trades, and a compounding leverage/sizing bug drove equity negative — none of which parameter tuning could fix.

A long-short, single-instrument intraday momentum-breakout on SOLUSDT.BINANCE 15m perp bars that traded the continuation of genuine volatility-expansion breakouts — firing only when a bar broke the prior 40-bar channel and its true range exceeded 2.5x ATR, then riding winners with a chandelier trailing stop to monetize the fat right tail while cutting losers on an initial ATR stop.

Over a well-sampled 1,396 trades (680 long / 716 short) the edge was decisively negative — avg_trade_return_pct -0.046% (below the 0.10% round-trip), profit factor 0.73, expectancy -$88.8/trade, Sharpe 0.31 with CI [-1.18, 0.71] straddling zero. The account was liquidated on 2023-11-23 for a -100% total return and 100% max drawdown, with a broken sizing that pushed avg_position_pct to 132% despite a claimed 0.5x cap and produced impossible daily returns like -522% and -1820%.

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