EXP-013 — RSI(2) shared-capital portfolio

EXP-013 — RSI(2) shared-capital portfolio

• Status: design complete · not run, to run after EXP-014
• Contract: common contract
• Question: when the same signals are pooled into one account, do after-cost returns and cash utilization improve?

Base form and changed factor

• Universe: the fixed 30 symbols of EXP-011. Symbols are not removed or added by their return order.
• Signal/exit: day t close>SMA200 & RSI2<5 → enter at the next open; RSI2>70 → exit at the next open.
• Single total capital of 3,000,000 KRW, K=5, 19% budget per symbol, no add-on buys or hard stops.
• Same-day candidates in ascending RSI2 order, ties by symbol code. Held symbols and symbols being exited are excluded.
• The buy list is fixed in priority order from all signals confirmed the previous day. Orders that cannot trade that day or lack cash record the reason; no new candidates are searched that day and earlier signals are not reused the next day.
• Unfilled exits of existing positions keep occupying slots. A scheduled sale alone does not create an assumed empty slot.
• Sale proceeds are not used for buys at the same open. P&L or unrealized gains of new positions do not enlarge the budget for orders at the same open.

The changed factor in this experiment is capital sharing. The base exit is not replaced by searching the EXP-014 return paths for the best day. A fixed 5-day O1→O6 exit is fixed now as a separate diagnostic form. It is not mixed with the existing max_holding_bars=5 exit at the 5th close.

Comparison groups — all with total capital of 3,000,000 KRW

scaled_market uses only calibration weight → history_A and history_A weight → history_B. sleeves30 leaves symbols it cannot buy even 1 share of in cash and publishes the failure count. Because the shared account could win for this reason, a 30,000,000 KRW auxiliary account checks whether the direction is the same, to separate the interpretation.

Preregistered matrix

• 3 slot settings {3,5,10} × exit {RSI70, 5-day open} × 3 periods × 4 costs = 72 strategy cells.
• 1st specification: K5/RSI70/history_B/proxy_slip2. If K changes, the per-symbol budget is 95%/K.
• 30,000,000 KRW account: K5/RSI70 × 3 periods × 4 costs = 12 diagnostic cells.
• Execution delay that postpones only entry by one more trading day: K5/RSI70 × 3 periods × 4 costs = 12 diagnostic cells.
• 96 strategy cells in total. The first 4 fixed comparison groups' 48 cells and scaled_market's 8 cells are in separate ledgers. The best cell is not promoted to the base form.

Verdict

The 1st statistic is the daily log-return difference of the K5 shared account − sleeves30. Research promotion conditions:

1. EXP-014 supported; common data/accounting gates passed.
2. history_B shared account completed trades ≥100, total union of entry days ≥50, months with trades ≥12.
3. The account's own after-cost total return >0, lower CI bound of the 1st excess log-return >0.
4. Under both cost stresses, the own total return and the point estimate of the excess effect vs sleeves30 are both positive.
5. Own total return positive at baseline costs in 2 or more of the 3 periods. Only if the shared−independent excess effect is also positive in the 30,000,000 KRW diagnostic is it interpreted as a 'cash sharing' effect; otherwise it is limited to a small-account integer-quantity effect.
6. The total stays positive after separately excluding the realized net P&L of the single top symbol and the single top year. State explicitly that this is a contribution diagnostic, not a rerun with reallocation.

A 20% average capital deployment rate, MDD≤25% and CAGR≥15% go in a separate operational-suitability table. If exposure rises but MDD grows because of simultaneous entries in market crashes, it is not described as an after-cost risk-adjusted improvement. The number of concurrent positions on the largest loss days, P&L correlation across symbols and cash/slot rejection rates must be reported together.

Synthetic accounting gates

• On every date NAV = cash + Σ(quantity × marked_price), cash≥0, held symbols≤K.
• 8 simultaneous buy signals → limited to 5 with rank/tie handling; reversing the fill order gives the same fixed-rank result.
• Sale + buy at the open → no same-day reuse of sale proceeds; unfilled exits keep their slot.
• Cases: cannot afford 1 high-priced share, fees exceeding the balance, trading halts, splits/dividends, unclosed at the end, 1-day delay.
• Reduce to a single symbol/K1 and compare with v2. Contracts that changed, such as new-order validity, corporate actions and end valuation, state the differences explicitly, and defects of the old engine are not forcibly reproduced.

Planned outputs

NAV and underwater curves for the account and comparison groups, monthly/yearly returns, signal→order→fill→holding→exit ledger, diagnostics of capital deployment rate, rejection rate and simultaneous losses, cost/slot/exit sensitivity, and a verdict gate table. If the broker's actual buying power and raw-price accounting are unconfirmed, performance is recorded only as exploratory.