From Kill Decision to Candidate Identification: A Systematic Alternative Search After Strategy Failure¶

Introduction¶

When a core strategy is killed—rejected at the acceptance gate with a decisive negative result—the next question is not "Can we salvage this?" but rather "What should we build instead?"

This article walks through a structured process for identifying alternative candidates after a kill decision. The case study: a failed Intraday Volatility Breakout strategy (Sharpe -1.09) that left a gap in the portfolio—specifically, a missing US-session equity index strategy. Rather than immediately green-lighting the next idea, we followed a systematic candidate identification process that resulted in the Donchian breakout strategy covered in Article 3.

The Kill Context: Understanding the Vacuum¶

Before identifying alternatives, you need to understand what the killed strategy was supposed to do:

Failed strategy: Intraday Vol Breakout on USA500 + DAX - Intended role: US-session equity index coverage (no existing US-session sleeves in the portfolio) - Kill reason: Tight-stop design caused whipsaws and spread bleed; Sharpe -1.09 (well below > 1.0 gate) - Original hypothesis: volatility breakouts on intraday equity prices would capture directional moves - Actual failure mode: stops were too tight, causing premature exits on noise

Portfolio gap created: - Current LIVE sleeves: DAX Bollinger Band (European), EURUSD Bollinger Band (European), USDJPY Bollinger Band (European) - No US-session index coverage - No trend-following strategies (all current sleeves are mean-reversion)

The key is recognizing the kill didn't eliminate the need—it only eliminated one proposed solution.

The Systematic Candidate Identification Framework¶

Rather than brainstorming randomly, we used a structured framework:

Step 1: Diagnostic Analysis of the Kill¶

What went wrong with the original approach?

The Intraday Vol Breakout failed due to three factors: 1. Stop placement: Tight VWAP trails generated whipsaws on intraday noise 2. Frequency: 20–30 round trips created spread arithmetic problems 3. Regime-dependence: No filter for trending vs. choppy conditions

The key insight: The problem wasn't equity index breakouts per se, but rather the execution design. The question becomes: what alternative execution design on equity indices might avoid these failure modes?

Step 2: Define the Candidate Space¶

We mapped out a 30-instrument × 6-strategy-archetype matrix:

Instruments: - US equities: USA500, USA100, US-small-cap indices - FX: GBPUSD, AUDUSD, USDCAD, EURUSD, USDJPY - Commodities: Oil, Gold, Natural Gas - Crypto: BTC, ETH (newer additions) - Indices: Volatility (VIX), Rates (Treasuries)

Strategy archetypes: 1. Trend-following: (momentum, Donchian, moving average cross) 2. Mean-reversion: (Bollinger Bands, RSI extremes, percentile reversion) 3. Carry/structural: (roll-based strategies, volatility term structure) 4. Volatility-driven: (vol expansion breakouts, vol compression fades) 5. Correlation: (correlation mean-reversion, index vs. component spread) 6. Hybrid: (combinations of above)

Cross-tabulating: 30 instruments × 6 archetypes = 180 potential candidate combinations.

Step 3: Filter by Portfolio Gap Criteria¶

Not all 180 candidates are equally valuable. We filtered by:

After filtering: - Trend-following on US equities: Donchian, moving average cross, Keltner channels - Mean-reversion on GBPUSD/AUDUSD: RSI extremes, Bollinger Reversion - Different breakout archetypes: Risk-on/risk-off breakouts, volatility-weighted entries

Shortlist: 8–10 viable candidates

Step 4: Estimate Feasibility and Data Requirements¶

For each shortlisted candidate, we estimated:

1. Implementation effort: How hard is it to code and backtest?
2. Data availability: Do we have 6+ years of high-quality data?
3. Execution feasibility: Can we execute the logic at reasonable costs?
4. Backtest runtime: How long does discovery take?

This prioritizes candidates that are low-effort, high-speed validatable.

Step 5: Define Success Criteria for Testing¶

Before testing any candidate, we defined what "success" meant:

Kill gate: Discovery Sharpe < 0.5 → reject, move to next candidate Advancement gate: Discovery Sharpe > 1.0 AND correlation < 0.3 → approve for live-paper sizing Borderline pass: Sharpe 0.7–1.0 AND other metrics strong → escalate for portfolio decision

This prevents analysis paralysis: each candidate either passes, fails, or escalates.

The Donchian Candidate: Why It Was Prioritized¶

From the shortlist, Donchian channel breakout on USA500 was selected as the highest-priority candidate because:

1. Quick feedback: 0.25 days of implementation effort
2. Clear data: 6+ years available, no data gaps
3. Execution simplicity: Standard Donchian indicator, no exotic logic
4. Portfolio fit: Trend-following on US equities (opposite of existing mean-reversion on EUR/JPY)
5. Spread efficiency expectation: Longer holding periods (days/weeks) vs. tight-stop (hours) should improve spread arithmetic by 10–50×

Testing and Results: The Systematic Path¶

We tested Donchian on USA500 first (fastest cycle) and, if it passed, planned to test USA100 as a parallel variant.

Result: Discovery Sharpe 0.80 (per-trade, annualized)

This was a borderline pass — below the > 1.0 ideal but above the 0.5 kill gate. Key supporting metrics: - Spread absorption: 0.6% (100× better than failed Intraday Vol) - Correlation with existing sleeves: -0.04 to -0.09 (excellent) - Portfolio Sharpe lift: +0.066 (+5.4%) - No overfitting signature

The systematic identification process paid off: we found a candidate that was different enough from the failed approach to avoid its failure modes, simple enough to implement and test quickly, and valuable enough to advance as a portfolio component.

Why Systematic Beats Intuitive¶

This structured approach outperformed intuitive brainstorming in several ways:

1. Prevents Anchoring Bias¶

Intuitive: "The last strategy failed because X, so let's avoid X." Systematic: "The last strategy failed because X, Y, Z. What candidates have different X, Y, Z profiles?"

The failed Intraday Vol Breakout had: - Tight stops (X) - High frequency (Y) - No regime filter (Z)

The Donchian candidate has: - Loose stops (opposite of X) - Low frequency (opposite of Y) - Trend regime naturally filters (addresses Z)

2. De-Risks Implementation via Parallelization¶

Systematic framework allowed us to test multiple candidates in parallel: - Donchian on USA500 (0.25d implementation) - Donchian on USA100 (parallel, same 0.25d) - MA cross variants (parallel, 0.5d)

Rather than testing each sequentially (would take 2+ weeks), parallel testing delivered a result in 5 days.

3. Creates Documented Audit Trail¶

Each candidate in the shortlist has: - Explicit rationale (why this was prioritized) - Estimated effort (feasibility) - Success criteria (kill gate, advancement gate) - Test results (pass/fail/escalate)

If the Donchian strategy later underperforms, we can ask: "Was this a random failure, or did the expected regime conditions fail to materialize?" The documented hypothesis makes that question answerable.

Common Pitfalls Avoided¶

Pitfall 1: "Let's Try a Completely Different Asset Class"¶

After the USA500 + DAX failure, a tempting instinct is to pivot entirely away: "Maybe equity breakouts don't work; let's try crypto or commodities."

Systematic approach: No. The US-session gap is still the portfolio gap. The problem wasn't US equities; it was tight-stop breakout designs on US equities. Change the design, not the asset class.

Pitfall 2: "Let's Add a Regime Filter to the Failed Strategy"¶

Another temptation: tweak the original Intraday Vol Breakout by adding a regime filter, running it again, and hoping for better results.

Systematic approach: The failure was multi-dimensional (stops, frequency, regime). Adding one filter doesn't address all three. Better to switch to a fundamentally different execution model (longer holding periods, wider stops).

Pitfall 3: "This Looks Promising; Ship It"¶

The Donchian discovery Sharpe (0.80) was below the > 1.0 ideal gate. A non-systematic approach might have rushed to live deployment: "Good enough!"

Systematic approach: Define clear gates upfront (kill < 0.5, advancement > 1.0, borderline 0.7–1.0). Let each candidate fall into its category. For the borderline, escalate with supporting evidence (correlation, portfolio lift) and let decision-makers choose.

Candidate Identification as Iterative Process¶

This isn't a one-time decision; it's a framework that repeats:

1. Kill a strategy → identify the portfolio gap
2. Systematically generate candidate list
3. Test highest-priority candidates first
4. Document results and decision
5. If a candidate advances: deploy and monitor
6. If the deployed candidate later underperforms: return to the candidate list and test the next option

The process is robust because it: - Doesn't waste time on low-priority candidates - Parallelize fast tests (low effort) - Escalates borderline cases to decision-makers with full evidence - Creates an audit trail for post-mortems

Conclusion: From Kill to Rebuild¶

When a core strategy fails, the instinct is to move fast: get the next idea live as soon as possible. But sustainable strategy development moves fast in the right direction—which requires a moment of systematic thinking before testing.

This framework—diagnostic analysis, candidate space definition, filtering, feasibility estimation, and pre-test success criteria—ensures that the next strategy you develop doesn't replicate the failure modes of the one you just killed.

In this case, the systematic approach took a Sharpe -1.09 failure (Intraday Vol Breakout) and replaced it with a Sharpe +0.80 borderline success (Donchian). The process wasn't perfect, but it was faster and more reliable than intuitive guessing.

That's the real value of systematic candidate identification: speed and reliability, not spectacular returns.

Data sources: Portfolio composition data (current sleeves, session alignment), instrument and strategy candidate matrix, backtest results for Donchian and alternative candidates, feasibility estimates.

Note on methodology: This article describes the decision-making framework for strategy candidate prioritization after a kill decision. No proprietary models, live sizing details, or undisclosed instruments are referenced. The framework is applicable to any quantitative trading context.