EXP-008 verdict — Turtle + Bollinger squeeze

EXP-008 verdict — Turtle + Bollinger squeeze

• Run date: 2026-09-04
• Execution model: gpt-5.6-luna / high
• Verdict: hold
• Scope: 7 core CREON daily symbols, 3,000,000 KRW, 95% compounding contract

Verdict

The Bollinger candidate is not promoted to a live strategy. Simple mean reversion at the lower band was excluded from testing from the start; only the form using the squeeze as a Turtle entry filter was explored.

In the provisional actual-assumption locked OOS, the 60-day squeeze filter showed a median per-symbol CAGR of about 9.5% and a mean MDD of about -16%, a signal of lower risk than the Turtle control. But it did not pass the preregistered criteria of a CAGR of 15% and consistent direction across train/validation/OOS. The 125-day and 250-day filters suffered from trade sparsity and low median CAGR, and the pure upper-band breakout + 2R diagnostic group also had a median CAGR of about 3~5%, too little to justify promotion.

Under 2x costs and 5bp/one-way slippage sensitivity, the direction of the 60-day filter did not change much, but that is not a robustness pass; it is the limited observation that "the result does not vanish from costs alone". Survivorship bias in the current-basket, current-listing data means a final adoption decision cannot be made.

Next conditions

Before lifting hold, add a point-in-time universe and delisting history, fix trading-value and price filters and a simultaneous multi-symbol portfolio, then revalidate. Confirm commissions from actual CREON fills, and decide by a separate contract whether the account peak-drawdown stop is permanent or resumes monthly.

The canonical results are the 0, and the design is the 1.