DogeCompressionBreakoutExpansionLS

Compression-Breakout Volatility-Expansion Continuation, Long-Short (Single-Instrument DOGEUSDT.BINANCE USD-M Perp, 15m Bars, Trade the Directional Break of a Coiled Low-Volatility Range with a FIXED 2:1 Reward:Risk ATR Bracket, 3-Parameter)

A LONG-SHORT, single-instrument intraday breakout on DOGEUSDT.BINANCE (USD-M perp, 15m bars). The edge is volatility mean-reversion in TIME, not price: periods of unusually tight range (low realized volatility / compression) are followed disproportionately often by a large directional expansion move as coiled positioning unwinds. The strategy waits for a compression regime, then trades the DIRECTION of the range break with a fixed, favorable reward:risk bracket. This is deliberately NOT another single-name mean-reversion fade (the class that repeatedly died this session with inverted reward/risk: tiny TP + wide ATR stop => wins small, loses big). Here the structure is inverted on purpose — the take-profit is 2x the stop distance, so the strategy wins less often but each win is larger than each loss, giving positive expectancy at a modest ~40% hit rate. DOGE is chosen because its 15m realized volatility is high enough that a 2-ATR target represents a ~0.6-1.2% move, comfortably clearing the 0.10% round-trip futures fee, while it is far less saturated in our portfolio than BTC/ETH.

Implements volatility-mean-reversion-in-time: a genuine low-vol coil (recent ATR <= 0.70 x baseline ATR, measured EXCLUDING the current bar so the expansion bar itself doesn't mask the compression) precedes disproportionately large directional expansions, and we trade the break WITH the move. The structure is deliberately inverted vs the tiny-TP/wide-stop fades that repeatedly died with inverted reward/risk: here the take-profit is FIXED at 2x the stop distance (rr_mult=2), so each win is larger than each loss and expectancy is positive at a modest hit rate — the core thesis. DOGE 15m realized vol is high enough that a 2-ATR target is a ~0.6-1.2% move, clearing the ~0.10% USD-M round-trip several times over, and DOGE is under-represented vs BTC/ETH. Risk control incorporates the lessons from the SOL breakout sibling that liquidated: entry fires only while flat (base guard + explicit check, no pyramiding), and position_size bounds TOTAL gross exposure (net signed notional + new order) to max_notional_mult x equity = 0.5x under leverage 1.0, so a stop-out costs ~risk_per_trade of equity and cannot liquidate. Single-instrument uses the base on_bar/dispatch unchanged; calculate_signal returns a continuous channel-position score that varies every bar while caching the compression+breakout booleans and the fixed 2:1 bracket levels, and should_exit runs the entry-anchored target/stop/time bracket seeded via the _was_in_pos pattern. The compression gate plus a 100-bar baseline warmup is a small fraction of the thousands of 15m bars in the window, and compressed regimes (~20-30% of the time) break out frequently, so the trade count is measurable (hundreds). Kept to exactly 3 tuned parameters (compression_ratio, breakout_len, stop_atr) to resist overfitting; the 2:1 ratio, ATR/squeeze windows, hold cap, and risk caps are fixed structural choices." }

Account liquidated (100% drawdown, -100% total return) on 2021-10-25 with negative per-trade expectancy across a decisive 1594-trade sample (avg_trade_return_pct -0.153%, PF 0.31, Sharpe -1.33). The 2:1 compression-breakout thesis is falsified by its own realized numbers: avg_win $152 is SMALLER than avg_loss $186, so the intended favorable bracket never materialized (targets rarely reached, stops filled through) — the same inverted reward:risk that killed the sibling single-name strategies this session, reached by a different route. Optimizing compression_ratio/breakout_len/stop_atr cannot flip negative expectancy or un-liquidate the account; it would only overfit the pre-blow-up window. Abandon per the >50% DD hard-abandon rule and the negative avg_trade_return_pct fee floor rather than spend 2 hours optimizing.

Long-short intraday compression-breakout on DOGEUSDT.BINANCE (USD-M perp), 15m bars, pure OHLCV. It arms only when the market is coiled — mean true range over the prior 20 bars (excluding the current bar) <= compression_ratio (0.70) x the baseline mean TR over the prior 100 bars — then trades the direction of the range break: close above the prior 20-bar high -> long, below the prior low -> short. Each trade uses a FIXED, entry-anchored bracket with a 2:1 reward:risk (target = rr_mult=2 x stop_atr=1 x ATR, stop = stop_atr x ATR) plus a 32-bar (~8h) time cap — wins are twice the size of losses, so the strategy is profitable at a ~40% hit rate. Entry only while flat (no pyramiding); sizing risks 1% of equity over the stop and is bounded so total gross <= 0.5x equity (leverage 1.0, margin buffer). 3 tuned params (compression_ratio, breakout_len, stop_atr).

Abandon at backtest_review; lookbacks+stop can't lift a 32% breakout hit rate above 33.3%, and pure-OHLCV breakout is a dead class.

STRONG ABANDON RECOMMENDATION (not a QA code fail). Decisively negative and well-sampled: 296 trades, total_return -31.73%, Sharpe -2.48 CI entirely below zero [-6.40, -2.01], PF 0.54, win_rate 0.32, avg_trade_return_pct ~0.0019% (below the 0.10% round-trip), turnover 187 / $17,946 impact. Even a perfect 2:1 bracket gives NEGATIVE expectancy at a 32% hit rate (2:1 breakeven needs >33.3%; hypothesis assumed ~40%). Genuine edge failure, no degenerate signature — analyst's abandon call.

If iterated, use actual stop_market + limit bracket orders to enforce 1-ATR risk intrabar — but this won't fix the sub-33% hit rate, so abandon is still likely correct.

The core 2:1 thesis is NOT realized: avg_win $392 vs avg_loss $347 (1.13, not 2.0), extreme left tail (skew -16.1, kurtosis 290, largest_loss 3.7x avg_loss). Cause: should_exit checks the bar CLOSE rather than an intrabar stop order, so a volatile 15m DOGE bar gaps past the 1-ATR stop before close, realizing multi-ATR losses. Levels are structurally correct (not critical), but risk isn't bounded at 1 ATR — undermining the thesis. Consistent with the factory's standard close-based pattern.

Gate the compression/breakout/_last_trap block on self._open_position() is None, matching the BNB fix.

Minor latent smell: the _last_trap write in calculate_signal isn't gated on flat (unlike the BNB sibling that fixed this). If the bar right after entry also registers a breakout before _track_position seeds the bracket, _last_trap can be overwritten and the bracket seeded from the wrong bar. Impact is small here (fills at next open ≈ overwriting bar's levels; reseed only on first in-position bar), but it's a correctness smell.

Well-sampled (1594 trades, 211k bars) so the result is statistically decisive, not noise

Trade direction matches the hypothesis (long-short breakout, 823 long / 771 short, balanced)

Account LIQUIDATED on 2021-10-25: total_return -100%, max_drawdown 100% — hard-abandon territory (>50% DD)

avg_trade_return_pct -0.153% — negative per-trade expectancy, below the fee floor

profit_factor 0.31, Sharpe -1.33, win_rate 0.277 — no edge

The core 2:1 reward:risk thesis is falsified: realized avg_win $152 < avg_loss $186, so wins are smaller than losses (the exact inverted-RR failure the hypothesis claimed to invert)

Catastrophic tails (skew -31, kurtosis 1030, largest_loss -$3,684) drove the blow-up; the risk/sizing model failed to prevent liquidation

4.5 years of the backtest produced zero activity because the account was dead — no meaningful sample for optimization

This strategy deliberately inverted the tiny-take-profit/wide-stop fade structure that had repeatedly died, instead waiting for volatility compression on DOGE 15m bars and trading the directional break with a fixed 2:1 reward:risk bracket. Over a decisive 1,594-trade sample the thesis was falsified on its own terms: realized average wins ($152) came in smaller than average losses ($186), win rate was just 27.7%, and profit factor 0.31 — the account was liquidated on 2021-10-25 for a -100% total return and 100% drawdown, with extreme tails (skew -31, kurtosis 1030) driving the blow-up. The analyst abandoned it at backtest review under the hard >50%-drawdown rule and the negative per-trade fee floor, noting that optimizing the parameters could neither flip negative expectancy nor un-liquidate the account and would only overfit the pre-blow-up window.

Designing a 2:1 reward:risk bracket on paper does not guarantee it in practice — if targets are rarely reached and stops fill through, realized wins can still be smaller than losses; the sizing/risk model must actually prevent liquidation, and negative per-trade expectancy over a decisive sample cannot be optimized away.

The analyst abandoned it at the pre-optimization backtest-review gate under the >50% drawdown hard-abandon rule and the negative avg_trade_return_pct fee floor: the intended favorable bracket never materialized (targets rarely reached, stops filled through), producing the same inverted reward:risk that killed sibling single-name strategies this session, plus catastrophic tails (skew -31, kurtosis 1030) that drove the blow-up.

A long-short, single-instrument intraday breakout on DOGEUSDT.BINANCE USD-M perps (15m bars) that waited for a low-volatility compression regime and then traded the direction of the range break with a fixed 2:1 reward:risk ATR bracket, aiming for positive expectancy at a ~40% hit rate by making wins larger than losses.

It was a catastrophic, decisively-sampled failure: the account was liquidated on 2021-10-25 with total return -100% and 100% max drawdown across 1,594 balanced trades (823 long / 771 short), with win rate 27.7%, profit factor 0.31, Sharpe -1.33, and avg_trade_return_pct -0.153%. The core 2:1 thesis was falsified by its own numbers — realized avg win $152 was smaller than avg loss $186.