Weekly report — work week 4 (2026-W36)

Weekly report — work week 4 (2026-W36)

• Period: 2026-08-31 ~ 2026-09-06
• Stage: Connors RSI(2) time-frame conversion pilot
• Status: cost normalization review complete — hold (provisional until actual CREON account commissions are confirmed)

This week's single goal

Normalize existing EXP-002~EXP-006 to actual per-product trading costs, and re-judge the daily results and the A005930 and A069500 5-minute results.

Done this week does not mean finding high-return parameters. It means separating a pure 5-minute conversion from combining it with a prior-day daily trend filter, and deciding from results that passed data, session, execution and accounting gates whether collecting additional symbols' 5-minute bars is worthwhile.

The capital contract is fixed per standalone experiment at an initial 3,000,000 KRW, reinvesting 95% of current cash just before entry, integer quantities and no leverage. It is sequential compounding, using the balance after net P&L and costs from each exit as the base for the next trade, and capital across different symbols and algorithms is not combined unless it is a portfolio experiment.

Evidence carried from last week

• EXP-005 daily baseline: 87 trades, +32.33%, MDD -20.83%, profit factor 1.499
• The -7% hard stop reduced drawdown but also lowered returns and profit factor
• The 5-bar time exit, at -10.66%, removed the baseline's positive result
• 5-minute bars secured: about 96 thousand rows each for A005930 and A069500, 2021-07-16~2026-08-26
• 1st 5-minute run complete: A005930·A069500 × tracks A/B × 7 variants × 3 costs, 42 locked OOS results

Completion gates

1. Session, gap and OHLC audit and SHA-256 records for both 5-minute sources — done
2. Synthetic tests verifying next-5-minute-bar open fills and session boundaries — done
3. Accounting tests for initial 3,000,000 KRW, compounding with 95% of current cash and no leverage — done
4. Run track A (200 5-minute-bar filter) and track B (prior-day daily 200-day filter) — done
5. Costs 0/40/80bp, separate train/validation/locked OOS — done
6. Equity curves, return comparison charts and trade ledgers recorded — done
7. advance / hold / rejected verdict and decision on collecting more symbols — hold (expansion to more symbols on hold after cost normalization)
8. Cost-normalization rerun of earlier experiments, charts and ledgers — done, EXP-007

The canonical design is fixed in 0 . The cost normalization review contract is fixed in 1.

Non-goals

• Collecting intraday bars for all symbols
• Adopting the highest-return parameters
• Modifying daily EXP-005
• Connecting a real account or automated orders

Cost normalization review verdict

In EXP-007, existing results were rerun with per-product costs without overwriting them.

• A069500 Connors daily baseline: earlier EXP-005 +32.33% → normalized actual assumption +82.56%
• BNF EXP-003: earlier -94.70% → normalized actual assumption -92.08%; lower costs alone do not resolve the structural problem
• Connors 5-minute: 0 of 14 locked OOS cells positive under the normalized actual assumption
• The normalized commission is provisional until actual CREON account commissions are confirmed, and slippage is 0bp.

Detailed results are in 0 and the 1.

1st verdict

The earlier 40bp baseline included an excessive sell tax for the ETF, so it is not confirmed as "Connors RSI(2) rejected entirely". However, even after normalizing per-product costs, all 5-minute locked OOS results were negative, so expanding the current 5-minute signal/exit/3,000,000 KRW compounding contract to more symbols is on hold.

Detailed results and charts are in 0 and EXP-006-connors-rsi2-m5-review.html (original supporting file). The normalization review results are in 2.

Additional validation of the Bollinger Band candidate — EXP-008

The two documents and the Markdown for candidate 5 supplied by the owner were read only as research input, and simple lower-band mean reversion was excluded from the experiment contract. 60/125/250-day contracts attaching the squeeze as an auxiliary Turtle filter and a pure upper-band breakout + ATR stop + 2R diagnostic group were additionally run on the 7 quality-passed core symbols.

• Accounting: initial 3,000,000 KRW, 95% of cash just before entry, integer quantities, no leverage, sequential compounding
• Run: 10 contracts × ATR 1.0/1.5/2.0 × 4 cost profiles × 2 risk policies, 6,720 result rows
• Results: 43,945 trade-ledger records, 23,072 signal-ledger records (1 per symbol and contract), HTML charts included
• Key observation: in the provisional actual-assumption locked OOS, the 60-day filter had a median CAGR of about 9.5% and a mean MDD of about -16%, but missed the 15% CAGR screening criterion and period consistency. The 125/250-day filters and the pure BB diagnostic group also did not pass the robustness criteria.
• Verdict: hold. Worth further validation as an auxiliary filter, but not grounds for adopting a strategy or moving to a real account.

The design is in the 0, and the results and charts in the 1 and the EXP-008 visual report (original supporting file).

5-minute Connors RSI(2) confirmation — EXP-009

Without overwriting the existing EXP-006/007, a confirmation experiment kept the 5-minute signals and separated per-product costs and one-way slippage of 0/1/2/5bp. A005930·A069500 × 7 variants × 5 profiles = 70 combinations, with 70 trade ledgers plus summaries and charts generated.

• actual_proxy: 0 of 14 OOS cells positive, mean -30.23%
• One-way slippage 1bp: 0 positive, mean -34.63%
• One-way slippage 2bp: 0 positive, mean -37.18%
• One-way slippage 5bp: 0 positive, mean -47.24%

Changing slippage from 0bp to 2bp did not change the direction of the conclusion. Actual CREON execution commissions and quote-based slippage are not yet confirmed values, so it is recorded as hold, not a permanent rejection of the whole strategy. Expansion of the current 5-minute contract to more symbols and promotion to live trading are on hold.

The canonical design is the 0, the results the 1, and the verdict the 2.

Connors RSI(2) daily, generalization across symbols — EXP-010

To check whether the single-ETF result generalizes to other symbols, the quality-passed EXP-008 basket of 7 symbols was fixed before checking results. A069500 is the control, and A000660·A012330·A035420·A051910·A055550·A068270 were added. 504 period rows were run with the daily baseline, four exit variants, 3 cost types and 2 risk policies.

• Baseline at provisional baseline costs: 6 of 7 positive, mean OOS return +12.07%, median CAGR +6.79%, 39 total OOS trades
• Baseline at 2bp one-way slippage: 6 of 7 positive, mean OOS return +11.86%, median CAGR +6.55%
• Excluding the control, 5 of 6 generalization targets were positive, but with only 4~10 OOS trades per symbol the sample is small.
• The account stop policy was separated from strategy edge. Under account_halt, 368 new-entry blocks occurred, reducing OOS trades to 17.

The conclusion is hold. RSI(2) daily did not fail immediately like the 5-minute version and showed a positive direction on other symbols too, but grounds for strategy promotion are still insufficient. Detailed results and charts are in the 1, the design in the 2, and the verdict in the 3.