Net P&L
$0
Realized · Unrealized
Buying Power
$300,000
100% free
Positions
0
0 long · 0 short
TRAP Snapshots
0
idle
Today's schedule · 0
⭐ COMBINED · multi-strategy conviction · – multi-source
Loading…
🌙 MOC → MOO · Overnight hedged · entry 15:55 ET
Loading…
☀️ MOO → 9:55 (M14) · M10R baseline · entry 09:30, exit 09:55 · (7:1 SHORT pump-fade overlay → see MOO→11:00 card below)
Loading…
🆕 MOO → 11:00 v2 · 10 robust post-audit strategies · entry 09:30, exit 11:00 · 0 picks
—
Advisory only · TPSL +7%/-1% (9) + HOLD→11:00 (1) · $1K BP per (ticker,date) · max 20 concurrent · daily kill -$1K · pre-MOO = PROVISIONAL
Loading…
Skip reasons (last scan):
—
📣 Earnings-day prescan · MOO→9:55 (reporters) & 🔥 PUMP
Loading…
📡 Live Radar · top movers · –
Loading…
⏰ PM → MOO 09:10 · –
—
No scan yet…
🌅 PM → MOO v2 · –
—
No scan yet…
🩸 PM → MOO v3 · pump-fade SHORT · LIVE (manual) –
—
No scan yet…
PUMP Low-float setups · –
—
Scanning universe…
MOO->11:00
SHORT pump-fade · TP+7%/SL-1% · exit 11:00 ET ·
–
—
No scan yet today
Strategies 0
Knowledge · 0
Loading…
Positions · 0
No open positions
Live feed
🌙 MOC → MOO Research v3 — overnight strategy
Entry MOC 16:00 day D → exit MOO 09:30 next day. PM exits underperform (avoid).
Universe: mcap≥$2B, price≥$5, $vol≥$5M. Score: 6 factors (composite).
Train: 36 days (2026-03-31 → 2026-05-21)
| Side | Tier | N | WR% | avg % | Sharpe |
|---|---|---|---|---|---|
| LONG (green close) | A+ | 16 | 81.2 | +0.722 | +10.4 |
| LONG (green close) | A | 1260 | 61.2 | +0.654 | +3.44 |
| COUNTER (LONG on red) | A+ | 6 | 100.0 | +2.099 | +21.9 |
| COUNTER (LONG on red) | A | 843 | 73.5 | +0.833 | +7.06 |
OOS test (daily-only factors): 124 days 2025-10 → 2026-03
| Tier | Train Sharpe | OOS Sharpe | OOS WR | Verdict |
|---|---|---|---|---|
| LONG A | +3.44 | +0.81 | 47% | Edge weaker but survives |
| COUNTER A | +7.06 | -0.08 | 49% | Disappears without intraday features |
Key insights
- Best DOW for LONG: Tue (Sh 6.99) | worst: Fri (-0.39)
- Best DOW for COUNTER: Wed (Sh 11.2), Mon, Tue
- FOMC days: torgovat ONLY COUNTER (Sh +9.51); LONG fails on FOMC
- VIX: edge only при VIX mid/high (VIXY >35). Avoid VIX low.
- SHORT mirror: НЕ работает — red closes отскакивают overnight
- Best single factor: late-day volume ratio >1.5x avg (institutional accumulation)
⚠ Trading note: Train edge stronger than OOS — likely regime-specific. Start with small size, stop after 2 negative months, monthly re-validate. See
research_moc3/AUDIT.md for look-ahead audit.
🧭 Earnings Strategies · 0
Loading…
📣 Reporting TODAY · 0
Loading…
📅 Upcoming 7 days · 0
Loading…
🧪 Active Edges (earnings_edges) — Last scans
No edge scans yet. Enable earnings_edges to activate.
📚 Edge Registry · 0
Loading…
🏷 Earnings-tagged signals (last 3h) · 0
No earnings-tagged signals yet.
📊 MOO→11:00 · R-/R+ Earnings Advisor · 0
Loading research…
🆕 MOO→11:00 v2 · 10 robust strategies · 0 picks
—
No scan yet today.
📋 Strategy roster (10 unique)
| ID | Name | Tier | Dir | Exit | Holdout Sh | Notes |
|---|---|---|---|---|---|---|
| Q003 | opening_cross_dev | ELITE | SHORT | TPSL | 4.50 | 09:30 only (open_px gap up ≥0.5%) |
| I012 | pt_ratio_low | ELITE | SHORT | TPSL | 4.21 | analyst tgt ≤95% prev_close |
| Q004 | opening_cross_dev | ELITE | LONG | TPSL | 3.95 | 09:30 only (open_px gap dn ≤-0.5%) |
| J015 | pm_high_break | ELITE | LONG | TPSL | 2.87 | pm_close ≥99% pm_high |
| K014 | prev_5d_ge015 | ELITE | SHORT | TPSL | 2.55 | prev_5d_ret ≥15% |
| K013 | prev_5d_ge010 | ELITE | SHORT | TPSL | 2.40 | prev_5d_ret ≥10% (=K004) |
| K012 | prev_5d_ge005 | STRONG | SHORT | TPSL | 2.10 | prev_5d_ret ≥5% |
| K015 | prev_5d_ge020 | STRONG | SHORT | TPSL | 1.85 | prev_5d_ret ≥20% |
| K016 | 3d_pump_exhaust | STRONG | SHORT | TPSL | 1.60 | prev_3d_ret ≥15% |
| G006 | tlt_5d_neg | STRONG | SHORT | HOLD | 1.50 | TLT 5d ret ≤-2% (macro, not per-tk) |
⚙️ Skip reasons (last scan)
—
🪞 Déjà-vu Advisor
Target: —
K=— analogs
advisory only — never auto-executes
LONG signals
—
— unique tickers
SHORT signals
—
— unique tickers
Best similarity
—
top analog date
ETF overlays
—
triggered tickers
Top-K analog days
Loading…
Signal breakdown by window
⏱ PM 7–8 ET → MOO backtest —
Entry: VWAP 07:00-07:59 ET · Exit: 09:30 OPEN · cost 15 bps · $1K/signal · liq ≥10K shares
Cumulative PnL (12 mo, per $1K size)
Monthly breakdown — last 12 mo
| Month | N (all) | WR % (all) | PnL $ (all) | N (S+M) | WR % (S+M) | PnL $ (S+M) |
|---|
Signals —
HOLD hidden by default · CI 95% bootstrap
| Ticker | Window | Decision | Tier | N | Mean % | CI 95% [lo … hi] | ETF overlay |
|---|
📊 ETF pre-market divergence (07:00 ET) · advisory · BORDERLINE
7 ETF pairs · |div| ≥ 0.5% triggers significance · 3 borderline patterns
(raw p<0.05, BUT 0/50 BH-FDR pass on 2y OOS) · advisory per Apr 29 directive ·
see docs/research/etf_divergence_premarket_2026-05-07.md
Today's pair states (07:00 ET PM ret vs prev close)
| Pair | Date | ETF A · ret | ETF B · ret | Divergence | Leader | Interpretation |
|---|
🤖 Agent feed · nightly commentary + drift watchdog · advisory only
Source:
broker/data/agent_feed/*.md · runs nightly 22:00 ET via schtask · LLM never touches order path (Apr 29 directive)
Drift history (last entries)
| Date | Max level | dejavu | moo1100_v2 | earn_pead |
|---|
Recent commentary
Node status
Idle
Active (pulse)
Human-in-loop
Healthy
Orders · last 7 days
Loading...
⚙ Nightly Pipeline
Steps
Data freshness · critical tables
Strategy recalibration · drift vs monthly baseline
loading…
⚡ PM-MOO v22 — Multi-window accumulator
post-Mar31 only
strict MOO exit
1:1 QQQ hedge
unlimited positions
backtest 2026-03-31 → 2026-05-22, 34 days, ~87 trades/day
🕐 PM session timeline — 7 entry checkpoints + MOO exit
now: —:— ET
04:00
PM open
PM open
5:00
+$637
+$637
6:00
+$127
+$127
6:45
+$716
+$716
7:30
+$467
+$467
8:00
+$563
+$563
8:45
+$90
+$90
9:10
+$296
+$296
9:30
MOO exit
MOO exit
● entry checkpoint
● 8am-context (mom-aligned + β-excess fire)
● MOO auction exit (single clearing price)
P&L = per-window contribution в replace-low бэктесте ($1K/pos)
⏳
Preflight: loading…
—
—
📡 Data freshness
loading…
⚠️ Risk snapshot
—
loading…
🎯 TAKE NOW — live entry candidates
prescan/pmmoo + radar
loading…
🛡 Intraday cover — open positions (RTH-aware)
heuristic — cover_v3 = PM only
loading…
Heuristic: TP at +3% / SL at −3% (≥15min hold) / SOFT TP at +1.5% (≥2h hold) / SOFT SL at −1.5% (≥2h hold). For PM-tagged positions cover_v3 takes precedence (panel above).
💰 Sizing indicator — what to allocate per grade
[H7 — UNIFORM locked, tested 2026-05-24]
A+
$500
avg ret +0.481% · WR 57% · $5.98/pick
44% P&L from 27% picks
A
$500
avg ret +0.394% · WR 56% · $4.55/pick
17% P&L from 14% picks
B+
$500
avg ret +0.198% · WR 59% · $2.47/pick
39% P&L from 59% picks (diversifier)
H11 finding: A+/A edge is BACK-LOADED (56-69% in last 25min-MOO) — cover too early destroys alpha. B+ edge is FRONT-LOADED (97% in first 25min) — cover faster could lock gains. Per-grade cover thresholds (H1) tested but Δ within noise; UNIFORM p90 stays locked until 30+ OOS days.
📈 Per-grade scorecard — today vs backtest expected
GRADE
N picks
WR%
avg ret%
vs expected
Expected (H7 backtest baseline, $1K/pos):
A+ avg +0.481% WR 57% $5.98/pick · A +0.394% WR 56% $4.55 · B+ +0.198% WR 59% $2.47
🪣 Cover bucket fill-rate map (1-min)
loading…
N ≥ 50 ·
10 ≤ N < 50 ·
N < 10 (low-confidence fallback)
📋 EOD performance card — auto-refresh after 09:35 ET
NET P&L
—
vs backtest expected: —
TRADES · WR
— · —
A+: — A: — B+: —
BEST / WORST
—
—
COVERS · SLIP · KILL
—
—
Kill switch: not triggered
📊 LOCKED DEFAULT v2.3 — unlimited + $1M $-vol + mcap + ADV + earnings±1d [2026-05-22]
$+9,931
Total P&L (34d)
11.88
Sharpe
+8.48
CI lower
79%
WR-days
-$243
MaxDD
70
avg trades/day
2,371
Total trades
1,961 / 410
SELL / BUY
Filter cascade: no-filter Sh 12.85 PnL $11,850 → +$500K $-vol Sh 12.24 → +$1M $-vol +mcap +ADV +earn±1d → Sh 11.88 PnL $9,931 (LOCKED). Filtered out: 598 trades (-20%, $-1,919 PnL) — primarily earnings-day blackouts. Trade-off: чуть меньше Sharpe но event-safety + cleaner liquidity.
🎛 Filter spec — LOCKED v2.3 (2026-05-22)
| Gate | Threshold | Logic | Status |
|---|---|---|---|
| Price (entry close) | ≥ $5.00 | no penny stocks | ACTIVE |
| $-PM-volume (soft floor) | ≥ $1,000,000 | price × pm_vol_cumul at entry bar | ACTIVE |
| Market cap (if known) | ≥ $300M | finviz; leveraged ETFs / new listings pass-through | ACTIVE |
| ADV (avg daily volume) | ≥ 500K shares | finviz; if NaN — pass | ACTIVE |
| Earnings exclusion | ±1 calendar day | skip if next/prev earnings date within 1d | ACTIVE |
| MIN_GRADE | B+ (score ≥ 0.25) | scoring_pm_moo_v2 weighted-vote normalized | ACTIVE |
| ETF tier sizing | skip 0.5x calm | max(|SMH-QQQ|,|ARKK-QQQ|) gate | ACTIVE |
| MAX positions/day | ∞ (unlimited) | BP-bounded only, no slot cap | ACTIVE |
| Unknown-ticker policy | PASS | leveraged ETFs (SOXS/SOXL/TQQQ/...) — keep, $-vol gate их режет если illiquid | CONFIGURED |
| News blackout (FDA/M&A/lawsuit/...) | — TBD — | TTN integration pending; see roadmap below | DESIGN |
| Borrow availability (SELL leg) | — TBD — | need broker API (IBKR / Webull / ...) | DESIGN |
💰 P&L at different BP / per-position sizes — LOCKED v2.3
| Per-pos size | BP req (Reg-T 50%) | BP req (PM ~25%) | Daily exposure (avg) | Total P&L (34d) | Avg per day | Best day | MaxDD | Monthly (21d) |
|---|---|---|---|---|---|---|---|---|
| $200 | $16K | $8K | $14K | +$1,986 | +$58 | +$271 | -$49 | +$1,226 |
| $500 | $40K | $20K | $35K | +$4,965 | +$146 | +$677 | -$122 | +$3,065 |
| $1,000 (baseline) | $79K | $40K | $70K | +$9,931 | +$292 | +$1,354 | -$243 | +$6,135 |
| $2,000 | $159K | $79K | $140K | +$19,862 | +$584 | +$2,708 | -$486 | +$12,270 |
| $5,000 | $397K | $199K | $348K | +$49,655 | +$1,460 | +$6,770 | -$1,215 | +$30,675 |
| $10,000 | $793K | $397K | $697K | +$99,310 | +$2,921 | +$13,540 | -$2,430 | +$61,350 |
| $25,000 | $1.98M | $0.99M | $1.74M | +$248,275 | +$7,302 | +$33,850 | -$6,075 | +$153,350 |
BP: Reg-T 50% × (long+short avg notional). Portfolio Margin (TIMS): ~25%, BP в 2× меньше. Avg daily exposure = 70 trades × $pos_size × 1.14 etf_mult.
Линейность: Sharpe 11.88 на любом размере. Limit — BP и market impact (см. audit liquidity).
Прицел при $1K/pos: ~$40-80K BP → Monthly ≈ $6.1K. При $5K/pos: ~$200-400K BP → $30.7K/мес.
Scale-up rule: вверх по успеху — добавлять по +50% pos_size каждые 30 trading-days в плюс с MaxDD < 50% allocated capital.
Линейность: Sharpe 11.88 на любом размере. Limit — BP и market impact (см. audit liquidity).
Прицел при $1K/pos: ~$40-80K BP → Monthly ≈ $6.1K. При $5K/pos: ~$200-400K BP → $30.7K/мес.
Scale-up rule: вверх по успеху — добавлять по +50% pos_size каждые 30 trading-days в плюс с MaxDD < 50% allocated capital.
🎯 Cover engine V3+V4 — backtest validation [CALIBRATED + TESTED]
| Config | Days | Trades | PnL$ | Sharpe | MaxDD$ | WR-d |
|---|---|---|---|---|---|---|
| BASELINE (hold to MOO) | 34 | 2,371 | $+9,931 | 11.88 | -$243 | 79% |
| WITH cover V3+V4 | 34 | 2,371 | $+19,900 | 15.87 | -$50 | 91% |
| Δ (cover impact) | — | 0 | $+9,970 (+100%) | +3.99 | +$193 (5×) | +12pp |
⚠ In-sample caveat: V3 buckets calibrated on тех же 2,371 picks → overfit. Realistic OOS lift ≈ 50% from this = +$5K PnL, +2 Sharpe. Still major.
735
V3 hard cover (31% picks)
avg lift +1.18%
23
V4 β-extinct (1% picks)
avg lift +0.44%
1,613
Hold to MOO (68%)
no cover trigger
Live wiring (TODO): at each post-entry window, для каждой active position recompute paper P&L vs bucket p75/p90 × time_decay → emit COVER signal to broker UI. β-trigger picks also check sigma extinction (V4). Calibration data:
cover_v3_buckets.json, cover_v4_beta_stats.json.
📊 Performance — daily P&L · cumulative · rolling Sharpe · drawdown
Cumulative P&L (left axis) + Rolling 7-day Sharpe (right axis)
Daily P&L (baseline vs with cover)
Underwater drawdown — distance from peak equity
Hit rate (days)
base 79% cover 88%
Best day
base $+1,354 cover ~$+2K
Worst day
base -$179 cover ~-$80
Calmar (PnL/MaxDD)
base 40.0 cover 248
🎯 Cover roadmap — V3 (bucket-based) + V4 (β-extinction) [LOCKED design, calibration pending]
Принцип: early-cover не по hard target +2.5%, а по типу стака. Каждый pick относится к бакету; cover когда paper P&L достиг ожидаемой границы для этого бакета. Плюс для β-сигналов отдельный спец-cover когда сигнал «потух».
V3 — Bucket-based expected move
| Dimension | Tiers | Source |
|---|---|---|
| Market Cap | mega (>$100B) / large ($10-100B) / mid ($2-10B) / small ($300M-2B) / micro (<$300M, filtered out) | finviz Market Cap |
| Volatility (ATR14%) | low (<2%) / mid (2-5%) / high (5-10%) / extreme (>10%) | finviz Volatility, или compute из daily bars |
| PM-gap size at entry | small (0.3-2%) / medium (2-5%) / large (5-10%) / extreme (>10%) | stock_ret_cumul at entry bar |
| Grade | A+ vs A vs B+ | scoring_pm_moo_v2 |
5 × 4 × 4 × 3 = 240 buckets. Для каждого bucket — empirical p50/p75/p90 of historical |ret_to_moo| на post-Mar31 + 6m extended.
Cover triggers:
• paper P&L ≥ p75 of bucket → soft cover (manual confirm)
• paper P&L ≥ p90 of bucket → AUTO-cover (signal exhausted)
Time-decay overlay (V5): threshold ×= (1 - 0.5 × hold_pct), где hold_pct = (now - entry) / (9:30 - entry). На 8:00 после 5:00 entry (60% held) → cover-threshold = 70% × p75.
Cover triggers:
• paper P&L ≥ p75 of bucket → soft cover (manual confirm)
• paper P&L ≥ p90 of bucket → AUTO-cover (signal exhausted)
Time-decay overlay (V5): threshold ×= (1 - 0.5 × hold_pct), где hold_pct = (now - entry) / (9:30 - entry). На 8:00 после 5:00 entry (60% held) → cover-threshold = 70% × p75.
V4 — β-excess normalization (signal extinction) — для β-сигналов
Применяется ТОЛЬКО для picks где основной trigger был
• На entry:
• На каждом окне после entry: recompute
• Когда
Why: Strategy зашла в short overshoot. Когда overshoot нормализовался (β-сигма в нуле) — мы получили что хотели, держать дальше = unconditional drift к MOO. Закрываемся.
beta_excess_2sig (β-excess |σ|>2). Логика:• На entry:
beta_excess_sigma_entry = +3.2σ (например)• На каждом окне после entry: recompute
beta_excess_sigma• Когда
|beta_excess_sigma_now| < 0.5σ → signal extinguished → AUTO-coverWhy: Strategy зашла в short overshoot. Когда overshoot нормализовался (β-сигма в нуле) — мы получили что хотели, держать дальше = unconditional drift к MOO. Закрываемся.
Combined cover decision
for position in active_book:
bucket = classify(position)
p75_threshold = bucket_stats[bucket]['p75'] * (1 - 0.5 * hold_pct)
p90_threshold = bucket_stats[bucket]['p90'] * (1 - 0.5 * hold_pct)
if position.paper_pnl_pct >= p90_threshold:
COVER_AUTO # hard target hit
elif position.paper_pnl_pct >= p75_threshold:
COVER_SOFT # show as candidate, 1-click confirm
# V4 overlay для β-signals
if position.signal_type == 'beta_excess_2sig':
if abs(position.current_beta_excess_sigma) < 0.5:
COVER_AUTO # signal extinguished
bucket = classify(position)
p75_threshold = bucket_stats[bucket]['p75'] * (1 - 0.5 * hold_pct)
p90_threshold = bucket_stats[bucket]['p90'] * (1 - 0.5 * hold_pct)
if position.paper_pnl_pct >= p90_threshold:
COVER_AUTO # hard target hit
elif position.paper_pnl_pct >= p75_threshold:
COVER_SOFT # show as candidate, 1-click confirm
# V4 overlay для β-signals
if position.signal_type == 'beta_excess_2sig':
if abs(position.current_beta_excess_sigma) < 0.5:
COVER_AUTO # signal extinguished
Status: design locked. Calibration script — собрать buckets stats из бэктеста, тогда live-cover.
Estimated lift: +0.3-0.8 Sharpe + улучшение DD (фиксация прибыли до post-open volatility spike).
Estimated lift: +0.3-0.8 Sharpe + улучшение DD (фиксация прибыли до post-open volatility spike).
📰 News blackout — типы новостей для skip [DESIGN — TTN integration pending]
| News type | Why skip | Skip window | Severity |
|---|---|---|---|
| FDA approval / denial / CRL | Discrete event, ±20-50% moves, no mean-reversion | day-of + next 2 days | CRITICAL |
| M&A target (deal announce) | Pinned to deal price, no fade possible; for SELL = guaranteed loss | until deal closes/breaks | CRITICAL |
| M&A acquirer (rumor or announce) | -5-15% gap not fade, real signal | ±1 day | CRITICAL |
| Bankruptcy filing | -50-90% gap; can't borrow for SELL | indefinite | CRITICAL |
| Going concern / delisting | Same as bankruptcy class | indefinite | CRITICAL |
| Drug trial results (Phase 1/2/3) | Biotech, ±30-80% moves, all-or-nothing | day-of + next 1 day | HIGH |
| Guidance pre-announce | ±5-15% gap, real info | ±1 day | HIGH |
| SEC investigation / lawsuit (major) | Trust shock, -5-20% gap, not noise | day-of + next 2 days | HIGH |
| Stock offering / dilution announce | -5-10% gap, sustained | ±1 day | HIGH |
| Cyber-breach / data leak (announced) | -5-15% gap, sustained 1-3 days | ±1 day | HIGH |
| CEO/CFO resignation/firing | -5-15% gap, real signal | ±1 day | HIGH |
| Recall (product/safety) | -3-15% gap, sector dependent | ±1 day | HIGH |
| Analyst major rating change (Sell→Buy or vv) | Per memory: analyst upgrade + gap>10% = WR 12.5% trap | ±1 day if gap>5% | MEDIUM |
| Insider buying large ($1M+) | Bullish signal, fade SHORT not great | ±1 day for SELL only | MEDIUM |
| Activist investor 13D filing | +5-20% gap, real signal | ±2 days | MEDIUM |
| Earnings (already handled by calendar) | Discrete event, IV-collapse PM | ±1 day ✓ ACTIVE | DONE |
| Reverse split | Mostly handled by price filter ($5+ excludes) | implicit | PARTIAL |
Что разрешено (НЕ skip): general market commentary · sector news without name-specific impact · general analyst notes (без rating change) · social media trends without verified source. В эти news fade работает нормально.
Implementation plan:
1. Pull TTN news feed for previous 24h before each PM session start
2. Classify each news item by type (rules + LLM tag where rules ambiguous)
3. Build per-ticker blacklist for the day: any CRITICAL type in last 24h → skip 7 days; any HIGH type → skip 1 day; MEDIUM → SELL-only filter
4. Backtest separately: how many picks would be killed? Estimated 10-20% based on TTN volume during PM-MOO universe
5. Live integration: pre-session pass → blacklist → into accumulator filters
Implementation plan:
1. Pull TTN news feed for previous 24h before each PM session start
2. Classify each news item by type (rules + LLM tag where rules ambiguous)
3. Build per-ticker blacklist for the day: any CRITICAL type in last 24h → skip 7 days; any HIGH type → skip 1 day; MEDIUM → SELL-only filter
4. Backtest separately: how many picks would be killed? Estimated 10-20% based on TTN volume during PM-MOO universe
5. Live integration: pre-session pass → blacklist → into accumulator filters
🎯 Early-cover & flip-to-opposite — концепт (на разработке)
✅ Early-cover — закрыть в плюс не дожидаясь MOO
Идея: позиция уже даёт acceptable plus на середине PM — фиксируем, не рискуем drawdown'ом до 9:30.
Правила (предлагаются):
Cover mechanics:
Expected lift: ~+0.3-0.5 к Sharpe если правильные пороги — большие гэйнеры фиксируются до возможного pullback на open auction (volatility пиковая в первые 5min торгов). Нужен backtest.
Правила (предлагаются):
- Hard target: paper P&L ≥ +2.5% для SELL fade → cover
- Time decay: ≥ +1.5% и до MOO осталось < 30 min → cover (less time, less downside)
- Reversal signal: position SELL и текущий score уже SKIP / direction flipped → cover
Cover mechanics:
- Закрываем stock на bid (если SELL → buy to cover) или ask (если BUY → sell to close)
- Соответствующий QQQ hedge тоже снимаем
- Slippage ~1c per share (vs zero на MOO auction)
Expected lift: ~+0.3-0.5 к Sharpe если правильные пороги — большие гэйнеры фиксируются до возможного pullback на open auction (volatility пиковая в первые 5min торгов). Нужен backtest.
🔄 Flip — развернуться в обратную сторону
Идея: взяли SELL в 5:00, но к 8:00 сигнал на тот же тикер сменился на BUY — закрываемся и разворачиваемся в long.
Правила (предлагаются):
Mechanics:
Expected behavior: в данных у нас 487 SELL / 11 BUY — flip-кандидаты редки. Возможно < 20 flips за 34 дня. Edge от каждого может быть значительный если убирать кат tail-losers (типа CMPX -67%).
Правила (предлагаются):
- Перескор позиции на каждом окне (5:00→6:00→6:45→...)
- Если
new_direction != current_directionANDnew_score ≥ 0.50→ FLIP - OR: paper P&L < -1% AND opposite direction signal A+ → FLIP (stop+reverse)
Mechanics:
- Close existing position (cover SELL / sell BUY) at current bar close
- Open opposite position SAME notional
- Both legs of QQQ hedge нужно перевернуть тоже
Expected behavior: в данных у нас 487 SELL / 11 BUY — flip-кандидаты редки. Возможно < 20 flips за 34 дня. Edge от каждого может быть значительный если убирать кат tail-losers (типа CMPX -67%).
⚠ Trade-off: Early-cover / flip требуют intraday execution и реальный bid/ask. На MOO auction slippage нет (single clearing price). Off-MOO exit ≥ +1c bid-ask spread per share + impact. Нужно посчитать что edge от early-cover > cost.
Реализация next:
1. Бэктест early-cover на 5-min intraday bars (есть в features_pm_v2) — порог +2.5% / +3% / +4%
2. Бэктест flip — re-score каждой позиции на каждом окне
3. Combined: early-cover + flip + unlimited
4. UI: добавить в live-box ещё одну колонку «COVER NOW» и «FLIP» для активных позиций
1. Бэктест early-cover на 5-min intraday bars (есть в features_pm_v2) — порог +2.5% / +3% / +4%
2. Бэктест flip — re-score каждой позиции на каждом окне
3. Combined: early-cover + flip + unlimited
4. UI: добавить в live-box ещё одну колонку «COVER NOW» и «FLIP» для активных позиций
📈 Cumulative P&L — different fixed per-position sizes
🚀 Exponential ramp scenarios — increase pos size with success
| Scenario | Logic | 34d cum P&L | Final pos$ | Bumps |
|---|---|---|---|---|
| Fixed $1K (baseline) | no scaling, linear | $+9,931 | $1,000 | — |
| Conservative ramp (1.5×) | ×1.5 каждый раз когда cum ≥ $3K, $7.5K, $14K, $22K, $32K, $45K | $+16,501 (+66%) | $3,375 | 3 |
| Aggressive ramp (2×) | ×2 каждый раз когда cum ≥ $5K, $15K, $30K, $60K, $100K | $+14,542 | $2,000 | 1 |
| Weekly streak ramp (1.25×) | ×1.25 после каждой серии 5 winning days | $+16,121 | $2,441 | 4 |
| Continuous compound | pos = $1K × (1 + cum/$10K)^0.6 — smooth | $+12,752 | $1,638 | smooth |
Conservative (1.5× @ $3K) — best Δ: +$6,570 (+66%) на 34-day window. Это потому что bumps происходят на дне 9 (cum $3K), 16, 23, 28 — каждый bump активирует усиленный compounding на оставшихся sessions. На 252d (~год) при той же логике: extrapolation ≈ $1,000 → $50K-150K final pos, cumulative ~$500K-$1M (зависит от volatility).
📈 Cumulative P&L by fill mode
📊 Sharpe + bootstrap CI 95% — mode comparison
🔄 Replace-low — что это и зачем
В одну фразу: «Постоянно держим топ-15 текущего лучшего знания о дне. Когда сильнее сигнал приходит позже — меняем слабейшую существующую позицию.»
Без replace-low (greedy): что не так
В 5:00 утра трейдер набирает 15 позиций. Среди них — пара marginal picks со score 0.30-0.45 (formally A grade, на пороге). Они проходят фильтр потому что нужно набрать 15 слотов.
Дальше до 9:30: появляются сильнее сигналы на 6:45 / 7:30 / 8:00 — обычно с накопившимся volume, чистым momentum, выраженной β-excess. Score 0.85-1.00 (A+).
Greedy этих новых не возьмёт — «книга полная, мимо». И marginal picks дают катастрофические убытки (см. CMPX -67% / -$673 в 04-27 бэктесте).
С replace-low: как работает
На каждом окне сравниваем новых кандидатов с слабейшим в текущей книге. Если score новый > score слабейший → SWAP.
Слабый pick из 5:00 (CMPX score 0.44) вытесняется сильным A+ из 7:30 (MXL score 0.85). В книге остаётся только высококачественные позиции.
Когда swap происходит
| Условие | Action | Пример |
|---|---|---|
| Δ score ≥ 0.35 | AUTO-swap | drop SNXX(0.37)→add CCC(1.00), Δ=+0.63 |
| score_new ≥ 0.85 AND score_old ≤ 0.50 | AUTO-swap | A+ заменяет marginal A |
| 0.15 ≤ Δ score < 0.35 | Manual (1-click) | RGTI(0.75) ↔ MSTZ(0.48), Δ=+0.27 |
| Δ score < 0.15 | Не показывать (шум) | |
| Книга < 15 позиций | Просто добавить, без swap | free slot, no contest |
Эффект на бэктесте
| Metric | Greedy | Replace-low | Δ |
|---|---|---|---|
| Total P&L | $+2,545 | $+2,896 | +$351 (+13.8%) |
| Sharpe | 8.21 | 9.41 | +1.20 |
| CI lower (2.5%) | +3.57 | +5.97 | +2.40 |
| MaxDD | -$223 | -$167 | $56 меньше |
| WR days | 71% | 79% | +8pp |
| A+ picks | 204 | 361 | +157 (×1.8) |
🪟 Per-window contribution (replace-low)
| Window | Bar | is_8am | Trades | P&L $ | Avg ret % | WR % | Note |
|---|---|---|---|---|---|---|---|
| 5:00 | 04:55 | — | 62 | +637 | +0.92% | 61% | Best edge/trade, но slots сдают позднее |
| 6:00 | 05:55 | — | 46 | +127 | +0.34% | 59% | Слабее 6:45 |
| 6:45 | 06:40 | — | 60 | +716 | +1.07% | 63% | Workhorse — highest edge |
| 7:30 | 07:25 | — | 83 | +467 | +0.45% | 55% | Reliable mid-session |
| 8:00 | 07:55 | ✓ | 60 | +563 | +0.76% | 57% | Mom-aligned signals fire here |
| 8:45 | 08:40 | ✓ | 70 | +90 | +0.09% | 56% | Was -$2 в greedy → +$90 после upgrade |
| 9:10 | 09:05 | — | 117 | +296 | +0.24% | 62% | Самый большой shift — 33→117 picks |
⚙ Strategy config
Entry windows (7)
5:00 → bar 04:55
6:00 → bar 05:55 [NEW]
6:45 → bar 06:40
7:30 → bar 07:25 [NEW]
8:00 → bar 07:55 [is_8am]
8:45 → bar 08:40 [is_8am]
9:10 → bar 09:05
Gates & filters
MIN_GRADE = B+ (score ≥ 0.25)
PMVol ≥ 50,000
Price ≥ $5.00
ETF tier skip 0.5x calm (V7)
MAX_POS = ∞ (unlimited)
PER_POS = $1,000 base
etf_mult ∈ {1.0×, 1.5×}
SELL signals (fade)
vwap_above (+1%) → 7.32
mom_5m_sell (>+0.3) → 9.45
mom_aligned_up (8am) → 7.05
rsi_over (>70) → 2.91
bex_bb_tight → 10.55
BUY signals (re-added)
vwap_below (-1%) → 3.85 [RE-ADDED]
mom_5m_dn (<-0.3) → 4.82 [NEW]
mom_aligned_dn (8am) → 4.91
beta_excess_2sig (8am) → 3.22
bex_bb_fade → 2.11
📋 Daily baskets — picks per window per day (click date to expand)