Cumulative P&L
—
— days traded
Latest Signal (US Date)
—
—
Avg Slippage
—
— days measured
60D Sharpe*
—
Weak Days
—
30D α
—
エッジ算出日
—
Trading Reference Card ▼ click to toggle
Losing Streaks (11yr, regime)
| Days | Freq | Avg Loss | Worst | Verdict |
|---|
| 1d | 128x | -1.8% | -17.6% | routine |
| 2d | 49x | -4.0% | -16.7% | common |
| 3d | 20x | -4.9% | -10.7% | ~1/month |
| 4d | 11x | -9.4% | -22.7% | ~1/2mo |
| 5d | 5x | -10.5% | -15.0% | ~1/6mo |
| 8d | 1x | -12.2% | -12.2% | 1x/11yr |
| 10d | 1x | -10.4% | -10.4% | 1x/11yr |
Winning Streaks (11yr, regime)
| Days | Freq | Avg Gain | Best |
|---|
| 1d | 90x | +2.2% | +7.5% |
| 2d | 52x | +4.6% | +15.1% |
| 3d | 28x | +6.2% | +15.5% |
| 4d | 24x | +8.5% | +21.2% |
| 5d | 10x | +8.2% | +14.9% |
| 6-8d | 10x | +15.3% | +26.3% |
| 12d | 1x | +26.3% | +26.3% |
Win Rate
57.9%
514/887 days
Monthly Win
75%
42/56 months
Max Streak Loss
-22.7%
4d, 2025-04
Max DD
-31.0%
avg recovery ~9d
RULES:
3連敗 = 月1で普通 /
5連敗 = 半年に1回 /
6連敗以上 = 年1回以下の稀なイベント（最悪10連敗） /
累積-23%超 = 過去にない異常値 /
HALT中は従う（過去に無視してたら-49%）
Backtest Cumulative Return (PCA PLAIN)
—
Nikkei 225 DD from peak
Entry Levels:
Strategy: Conservative (DD -12/-15/-18/-21%, 25% each) ·
Exit: +10% profit target / 180d max hold ·
Target: 1321.T (Nikkei 225 ETF) ·
BT: 16 rounds, 94% win, +9.74%/yr ·
Repo
Individual Stock Strategy
PIPELINE NOT BUILT
Strategy Configuration
SignalPCA PLAIN (L=120, K=3, λ=0)
GapFadeOFF (harmful for stocks)
Gap Opposite1.0%, x2.0
MA Deviation20d, 5%, x1.3
SPY Trend<MA20 → x0.8
VIX Filter≥25 → 0x (full cut)
RegimeETF Dispersion (shared)
HoldingIntraday (open→close)
TOP6 Stock Mapping (BT-based selection)
| Sector ETF |
Stock |
AR% |
SR |
ADV (B¥) |
| 1623.T Steel | 5401.T Nippon Steel | +4.89% | 0.53 | 70 |
| 1617.T Food | 2503.T Kirin HD | +3.90% | 0.63 | 39 |
| 1618.T Energy | 1605.T INPEX | +3.13% | 0.43 | 58 |
| 1620.T Materials | 4063.T Shin-Etsu | +2.21% | 0.44 | 20 |
| 1633.T REIT | 8801.T Mitsui Fud. | +1.89% | 0.27 | 17.6 |
| 1622.T Auto | 7203.T Toyota | +1.56% | 0.32 | 84.5 |
Key Findings (v1-v25 Research)
• GapFade is harmful for individual stocks (SR 0.72→0.27). Gap Opposite replaces it.
• ETF Dispersion regime works for stocks too (PCA signal sourced from ETF correlations).
• Correlation ≠ performance: MUFG (r=0.778) yielded ±0%, while Nippon Steel (r=0.583) topped at +4.89%.
• IS/OOS stable: SR 1.19 (IS) / 1.31 (OOS). No overfitting detected.
• 2023: zero TRADE days (full-year HALT). Capital idle risk exists.
Activation: deposit ≥ ¥1.9M (ETF max_capital reached) ·
Required: Pipeline build + paper trade ·
Repo:
profit2
Capital Allocation Decision Matrix
Strategy Comparison
|
Net AR |
Sharpe |
MDD |
Win Rate |
Holding |
Turnover |
| NEXUSα ETF |
+48.0% |
1.87 |
14.8% |
57.9% |
Intraday |
Daily |
| Individual Stocks |
+10.6% |
1.23 |
18.5% |
52.3% |
Intraday |
Daily |
| Crash Recovery |
+9.7% |
N/A |
8.9% |
94% |
~103 days |
~1.6/yr |
Growth Phases & Capital Routing
| Phase |
Deposit |
ETF |
Stocks |
CR |
| 1. Ramp-up |
¥30–54万 |
100% |
— |
— |
| 2. Growth |
¥54–153万 |
100% |
— |
— |
| 3. Expand |
¥153–190万 |
100% (17 tickers) |
— |
— |
| 4. Overflow |
¥190万– |
Cap ¥190万 |
Surplus |
Conditional |
When to Route Surplus to Crash Recovery
DEFAULT
Surplus → Individual Stocks (daily compounding, same capital efficiency)
EXCEPTION 1
Surplus < ¥50万: Stock unit constraints reduce efficiency → CR via 1321.T is viable
EXCEPTION 2
Stock strategy in HALT + Nikkei DD ≥ -12% → Route to CR entry
EXCEPTION 3
Nikkei DD ≥ -18% (3rd/4th level) → EV > +10%. Consider partial realloc from stocks at next open
Core Insight: Stocks and CR have similar annual returns (~10%), but stocks compound daily while CR locks capital for ~103 days.
CR only wins when: (1) stocks are halted, or (2) a deep crash (≥-18%) offers >10% EV in a short window.
The 2024/8 crash recovered +11.2% in 11 days — that’s the scenario where CR adds value.
Research: Approach A — Pre-market Mid Adjustment (Look-ahead Proxy BT)
⚠ Look-ahead Bias Notice
この BT は 9:00 実寄値 を「8:59:55 mid の代理」として使用してます。
真の 8:59:55 mid データは Phase 5 (4/13 月〜) で蓄積中。
look-ahead degradation 推定: 0-5% (個人スケール)
— 中央値 gap 143bps に対し 5秒前 mid noise は 5-10bps、ranking flip ほぼ起きず。
アイデア: PCA は「今日 JP がどっち動く」を予測。mid は「8:59 時点で既に動いた分」を観測。
residual = signal_z − w × observed_z で「残りエッジ」を計算し、それで re-rank → 新 long 4 / short 4 を選定。
期間: 2024-01-05 〜 2026-04-09 (556 営業日)、universe 17銘柄 (1629/1633 除外で active 15)、PCA L=120 K=3 λ=0
| weight |
days |
cum P&L |
SR (ann) |
MDD |
win rate |
PF |
vs w=0 |
win vs w=0 |
| 0.0 (baseline) |
556 |
+66.04% |
+2.20 |
14.59% |
55.0% |
1.43 |
— |
— |
| 0.3 |
556 |
+121.17% |
+4.05 |
10.42% |
60.1% |
1.98 |
+55.13% |
273/556 |
| 0.5 |
556 |
+151.11% |
+4.97 |
7.71% |
63.1% |
2.32 |
+85.07% |
340/556 |
| 1.0 ★ |
556 |
+217.30% |
+6.80 |
6.27% |
68.0% |
3.40 |
+151.26% |
366/556 (66%) |
| 2.0 |
556 |
+248.61% |
+7.72 |
5.92% |
68.7% |
3.99 |
+182.57% |
351/556 |
★ w=1.0 が sweet spot:
SR/PF/MDD のバランス最良、win vs w=0 が 66% (Z=7.46, p<0.0001 で統計的に強く有意)。
w=2.0 は cum P&L 微増だが win 率は飽和。
注目: mid 取り込みは MDD も改善 (14.59% → 6.27%)。
攻撃的になるだけでなく リスク制御 にも効く。
個人スケール realistic SR 推定 (look-ahead -5% + cost 2bps × 250日 -19% を考慮):
SR 5.2-5.5
(production 現行 SR* 1.88 monthly = daily-equiv ~2.45 の 2x 改善)
capacity: 個人～小規模 (deposit 1億円以下) で full effect。
機関規模ではセクター ETF の流動性制約で degradation。
詳細設計: docs/research/premarket_mid_design.md /
実装: scripts/research/mid_adjust_approach_a_bt.py --full-historical