Escaping Spread Costs: Why a Borderline Backtest Got Advanced to Live Deployment¶

Introduction¶

Not all winning strategies have a Sharpe ratio above 2.0. Some of the best portfolio additions are borderline-pass candidates that clear the acceptance gates not because of stellar standalone metrics, but because of a single hidden advantage: they do something expensive strategies do poorly—they absorb spread costs efficiently.

This is the story of a Donchian channel breakout strategy on equity indices that achieved a per-trade Sharpe of 0.80 (below the typical > 1.0 discovery gate), yet was advanced to live deployment. The decision hinged on a single metric: spread absorption of just 0.6%, compared to the failed predecessor's 100%.

The Predecessor's Fatal Flaw¶

Before we understand why the Donchian breakout worked, we need to understand what didn't: the Intraday Volatility Breakout strategy tested in our first article was rejected at Sharpe -1.09, primarily because tight stops and frequent exits created a spread cost problem.

• Intraday vol breakout: ~20–30 round trips, spread absorbed ~100% of intended alpha
• Strategy output: negative P&L, unprofitable

The fundamental lesson: holding periods and exit frequency determine spread arithmetic more than entry logic.

The New Approach: Multi-Day Donchian Breakouts¶

The Donchian channel breakout is a classic trend-following indicator: - Entry: when price breaks through the highest (or lowest) close of the prior N days - Exit: either take profit after a threshold move, or exit on a midband cross (opposite extreme) - Key difference from intraday vol: longer holding periods (measured in days or weeks, not hours or minutes)

Strategy parameters (N=55 was the optimized variant): - Instrument: S&P 500 (USA500IDXUSD) and Nasdaq-100 (USA100IDXUSD) - Entry trigger: Donchian channel breakout (55-day lookback) - Exit: Opposite N-day extreme OR ATR-based time stop - Position sizing: equal-weight across two indices - Testing window: 2020–04 through 2026–04 (6.2 years) - Realistic spread assumptions: same IG spread modeling as other research

The Results: Borderline Pass, High Portfolio Utility¶

Discovery Backtest (2020–2026)¶

The per-trade Sharpe of 0.80 is a borderline result—it doesn't hit the typical discovery gate of > 1.0. But the spread absorption metric flagged it as a portfolio diversifier, not a standalone strategy. This is a critical distinction.

Year-by-Year Breakdown: Regime Evidence¶

Key insight: Out-of-sample (2023–2026) outperforms in-sample (2020–2022) (Sharpe 1.03 vs 0.18). This is the inverse of overfitting—no evidence of curve-fitting artifacts.

The Spread Advantage: Root Cause of Advancement¶

Here's where the strategy's true value emerges:

Exit breakdown (23 trades across 6.2 years): | Exit Type | Count | Win % | Total P&L | |-----------|-------|-------|-----------| | Take-profit (target hit) | 8 (35%) | 100% | +£3,015 | | Midband cross (reversal) | 9 (39%) | 67% | +£304 | | Stop loss | 6 (26%) | 0% | -£896 |

The arithmetic of holding periods: - Average hold duration: 45 days - Spread cost per round trip on USA500: ~0.3–0.5 bps (~£0.03–0.05 per £10k notional) - Total spread cost across 23 round trips: ~£15 - Gross alpha before spreads: ~£2,437 - Spread absorption: 0.6% (industry-leading; most intraday strategies are 50–100%)

Compare this to the failed intraday strategy: - Average hold duration: < 1 hour - Spread cost per round trip: same absolute cost, much higher frequency - Round trips: 25–30 - Spread absorption: ~100% of intended alpha

The Donchian strategy's longer holding period inverts the spread/alpha equation. Even though the per-trade Sharpe is modest (0.80), the spread efficiency makes it a genuine portfolio contributor—not a drag.

Correlation: The Second Advantage¶

A portfolio needs strategies that move independently. The existing live portfolio was: - BB (Bollinger Band mean-reversion) on DAX - BB on EURUSD - BB on USDJPY

All three are European-session focused; all three are mean-reversion based. The Donchian breakout on USA500/USA100 is: - US-session focused (near-zero overlap) - Trend-following (opposite of mean-reversion) - Different instrument class (equity indices vs currencies)

Correlation analysis (daily returns, 2020–2026): | Frequency | vs BB-DAX | vs BB-EURUSD | vs BB-USDJPY | Average | |-----------|-----------|-------------|-------------|---------| | Daily | -0.001 | -0.002 | +0.000 | -0.001 | | Weekly | -0.088 | -0.040 | +0.006 | -0.041 | | Monthly | -0.120 | -0.186 | +0.037 | -0.090 |

Average correlation: -0.04 to -0.09 (well under the < 0.4 gate)

This is nearly zero correlation—slight negative tilt suggests mild diversification benefit. A portfolio containing this strategy is less correlated overall.

Composite Portfolio Impact¶

Combining the new Donchian strategy with existing BB strategies:

Adding the Donchian strategy increased portfolio Sharpe by 5.4% without increasing maximum drawdown. The marginal contribution is +£2,423 at only 10% of the existing portfolio risk.

Decision Gate: Why It Was Advanced Despite Borderline Results¶

The advancement decision came down to portfolio considerations, not standalone metrics:

The strategy was regime-conditional alpha: it works when equity indices are trending (2020–2024 style), and it breaks even or slightly loses in mean-reverting, low-vol regimes. But as a portfolio component in a mean-reversion-heavy portfolio, it provided essential balance.

The Caveats: Risk Management at Deployment¶

Advancing a borderline strategy to live requires careful position sizing and monitoring:

1. Small sample risk: 23 trades over 6.2 years means wide confidence intervals (±£106 per trade at 95% CI). Treat the first 6–12 months of live deployment as confirmation, not scaling.

2. Sparse activation: ~3.7 trades per year means the strategy can go 3–4 months without exposure. It's designed as a portfolio sleeve, not a standalone system.

3. Regime dependency: Strong trend environments (2020–2024) gave strong returns. Sideways regimes (2022–2023 mixed) or structural reversals could reverse the thesis quickly.

4. USA500 data dependency: 2018–2019 validation was deferred pending a data fix for earlier years (resolved later with 8-year testing).

Conclusion: When Borderline Is Actually Excellent¶

This case reveals a counterintuitive truth about quantitative trading: a borderline standalone backtest can be the best portfolio addition because it solves a different optimization problem.

The Donchian breakout scored a modest per-trade Sharpe (0.80), but it: - Absorbed spreads 100× better than the failed predecessor - Provided near-zero correlation to existing strategies - Lifted portfolio Sharpe by 5.4% without increasing drawdown

It was advanced to live not because it was outstanding as a stand-alone system, but because it was outstanding as a portfolio component—filling a specific gap (US-session trend exposure) that existing mean-reversion strategies couldn't fill without replicating their risks.

This teaches a core principle: measure strategies against portfolio impact, not just standalone metrics. A borderline strategy that solves a portfolio problem beats an excellent strategy that replicates existing risk.

Data sources: Equity index backtest data (USA500, USA100, 2020–2026), correlation analysis with existing strategy pairs, spread modeling from IG Tradable Instruments.

Note on methodology: This analysis is based on historical backtests of Donchian channel breakouts (a well-established indicator), standard equity index parameters, and realistic spread assumptions. No proprietary position sizing, live deployment details, or internal strategy names are disclosed.