EXP-007 — cost normalization review of past experiments

EXP-007 — cost normalization review of past experiments

• Generated: 2026-08-31
• Creation date: 2026-08-31
• Review model: gpt-5.6-luna / high
• Purpose: re-review EXP-002~EXP-006, separating their existing conclusions from their cost assumptions
• Status: 1st cost-assumption normalization complete · provisional until actual CREON account commissions are confirmed

Cost contract

• A069500, a domestic equity ETF, was treated as having a 0bp securities transaction tax on sales.
• A005930, a KOSPI stock, used the 2005~2026 per-period transaction tax schedule. The 2026 rate change follows the statute, and past periods used the published staged rate table.
• Tax sources: KRX ETF tax rules, National Law Information Center, Enforcement Decree of the Securities Transaction Tax Act.
• Source for the staged rate table: National Assembly Budget Office, 2024 Korean Taxation.
• The 1.5bp/one-way in normalized_actual_proxy is a conservative account assumption before checking the CREON fee schedule. It must be re-fixed after checking actual fills.
• Slippage was set to 0bp. With no real quote or fill data, it is interpreted together with separate slippage sensitivity.

Data

• ETF daily: [original local path withheld] · 2002-10-14~2026-08-26 · 5,891 rows · SHA-256 4964c32990e70b3aff09a7d895be7bf81b5d90ea31230e061b3e30f28c7e1a95
• Stock daily: [original local path withheld] · preserved original input of EXP-002/003 · SHA-256 b10f653a2ee6c1f9f462c59e0b74fcc0f10fe1a0367317e6c12dc405cbacbb7d
• 5-minute: the same A005930 and A069500 inputs as the existing EXP-006 were used; originals are not committed to Git.

Summary of review results

Key insights

1. Under the normalized actual assumption for 5-minute bars, 0 of 14 cells were positive. Nothing is adopted on a mean value or a single best value.
2. The earlier fixed 40bp cost results for the ETF may have exceeded actual costs, so the earlier "reject everything" wording for EXP-005 and EXP-006 is narrowed to rejecting the cost contract.
3. For BNF EXP-003, the structural mismatch between the 3R target and the 3-day exit remains even with lower costs. The cost problem and the strategy-structure problem are recorded separately.
4. Even if normalized results improve, nothing is promoted before checking that 0bp, 1.5bp and 2x stress point the same way and that OOS, year and symbol concentration pass.
5. Once the actual CREON fee schedule or fills are secured, normalized_actual_proxy is re-fixed as the new cost baseline.

Charts

The chart below is recalculated from the new ledgers. The 5-minute bars show locked OOS (2025-01-01~2026-08-26) returns.

Reproducibility

• Run date: 2026-08-31
• Model: gpt-5.6-luna / high
• Existing EXP-002~EXP-006 results and ledgers were not overwritten.
• Full summary ledger: reports/ledgers/EXP-007-summary.csv
• Original data and privately supplied material were not included in Git.