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

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

• Run date: 2026-09-04
• Review model: gpt-5.6-luna / high
• Status: exploration complete · hold

| generated_at | 2026-09-04T00:00:00+09:00 |

Purpose and symbol selection

Check whether the RSI(2) result observed on the single daily symbol A069500 reproduces on other symbols. Symbols were fixed to the EXP-008 data-quality-passed basket before seeing RSI results, and A069500 is the control. Only daily results are included; 5-minute results are not mixed in.

• Scope: A069500, A000660, A012330, A035420, A051910, A055550, A068270
• Control: A069500
• Rule: enter at the next trading day's open when RSI(2) < 5 above SMA(200); exit at the next trading day's open when RSI > 70
• Capital: 3,000,000 KRW per symbol and variant, 95% of cash just before entry, integer quantities, sequential compounding, no leverage
• Risk: the strategy-edge diagnosis uses research_unhalted (no blocking), and the account-operation diagnosis is separated as account_halt (daily -3%, monthly -8%, -20% from peak)

Data audit

Cost contract

The actual_proxy commission is a provisional value before checking actual CREON fills. actual_proxy_slip2 is a 2bp one-way sensitivity, not a real fill value.

Locked OOS baseline results

The locked OOS of the baseline (baseline) is 2025-01-01~2026-08-26. The main criterion below is the research_unhalted result, and the account stop policy is separated as a diagnostic. Returns are relative to the NAV just before the OOS start; the whole basket distribution is examined rather than adopting the single best symbol.

• actual_proxy positive symbols: 6/7 · mean 12.07% · median 11.46%
• 2bp one-way slippage positive symbols: 6/7 · mean 11.86%

Full results for all four exit variants

The table below covers every variant and cost profile in locked OOS, including both risk policies. hard_stop_7pct and the time exit are preregistered comparison groups, not best values adopted after the fact.

Interpretation

1. This experiment checks whether the positive result belongs only to the control A069500 or repeats on other symbols. At baseline costs, 6/7 symbols were positive, and no single-symbol adoption conclusion is drawn.
2. Check whether the direction holds under the provisional actual cost and the 2bp one-way stress. Detailed results can be recalculated from the summary and trade ledgers.
3. account_halt can make trades disappear after a peak-drawdown stop, so it is not used as the main result for strategy edge but kept as an auxiliary diagnostic of what the account policy actually does.
4. Not all symbols traded over the same period, and the current-listing basket has survivorship bias. So this result alone does not promote RSI(2) to a live strategy.
5. Even with positive symbols, it moves to the next stage only after passing year and trade concentration, benchmark comparison and point-in-time universe reviews.

Charts

Files

• Run: scripts/backtest/run_connors_rsi2_daily_basket_exp010.py
• Summary ledger: reports/ledgers/EXP-010-summary.csv
• Trade and signal ledgers: reports/ledgers/EXP-010-daily-*-*.csv
• Data manifest: reports/experiments/EXP-010-data-manifest.json
• Existing EXP-005, EXP-007 and the 5-minute EXP-006 and EXP-009 results were not overwritten.