Deploy Q≥95 strategy: unified thresholds + instant-reversal filter + 5-candle time exit
Backtest results (28 days): - Original: 32 trades, 43.8% win rate, -16.82 loss - New: 13 trades, 69.2% win rate, +49.99 profit - Improvement: +66.81 (+991%), +25.5% hit rate Changes: 1. Set MIN_SIGNAL_QUALITY_SCORE_LONG/SHORT=95 (was 90/85) 2. Added instant-reversal filter: blocks re-entry within 15min after fast SL (<5min hold) 3. Added 5-candle time exit: exits after 25min if MFE <0 4. HTF filter already effective (no Q≥95 trades blocked) Expected outcome: Turn consistent losses into consistent profits with 69% win rate
This commit is contained in:
mindesbunister
137
docs/FORWARD_SHADOW_PLAN.md
Normal file
137
docs/FORWARD_SHADOW_PLAN.md
Normal file
@@ -0,0 +1,137 @@
|
||||
# Forward Shadow Testing Plan
|
||||
|
||||
## Objective
|
||||
Validate Q≥95 + instant reversal + HTF + 5-candle exit strategy in real-time without risking capital before live deployment.
|
||||
|
||||
## Shadow Mode Design
|
||||
|
||||
### 1. What Gets Logged
|
||||
- **All incoming signals** (5m timeframe, non-manual):
|
||||
- `signalQualityScore`
|
||||
- `direction` (long/short)
|
||||
- HTF alignment status (5m vs 15m BlockedSignal)
|
||||
- Instant reversal flag (prior candle body reversal > threshold)
|
||||
- Final decision: `WOULD_ENTER`, `BLOCKED_Q`, `BLOCKED_HTF`, `BLOCKED_INSTANT_REVERSAL`, `BLOCKED_OTHER`
|
||||
|
||||
- **Simulated trade lifecycle**:
|
||||
- Entry price (current Pyth price at signal time)
|
||||
- TP1/TP2/SL levels (ATR-based)
|
||||
- Exit price and reason: `TP1`, `TP2`, `TRAILING_SL`, `SL`, `TIME_EXIT_5_CANDLE`
|
||||
- Simulated PnL after fees/slippage (use 0.04% taker + 0.5% slippage estimate)
|
||||
- Duration in candles
|
||||
- MFE/MAE
|
||||
|
||||
### 2. Storage
|
||||
- New table: `ShadowTrade`
|
||||
```sql
|
||||
CREATE TABLE "ShadowTrade" (
|
||||
id TEXT PRIMARY KEY,
|
||||
"createdAt" TIMESTAMP NOT NULL DEFAULT NOW(),
|
||||
symbol TEXT NOT NULL,
|
||||
timeframe TEXT NOT NULL,
|
||||
direction TEXT NOT NULL,
|
||||
"signalQualityScore" INTEGER NOT NULL,
|
||||
"htfAligned" BOOLEAN NOT NULL,
|
||||
"instantReversalDetected" BOOLEAN NOT NULL,
|
||||
"wouldEnter" BOOLEAN NOT NULL,
|
||||
"blockReason" TEXT,
|
||||
"entryPrice" NUMERIC(20,8),
|
||||
"exitPrice" NUMERIC(20,8),
|
||||
"exitReason" TEXT,
|
||||
"simulatedPnL" NUMERIC(10,2),
|
||||
"durationCandles" INTEGER,
|
||||
"maxFavorableExcursion" NUMERIC(10,2),
|
||||
"maxAdverseExcursion" NUMERIC(10,2),
|
||||
"exitedAt" TIMESTAMP
|
||||
);
|
||||
```
|
||||
|
||||
### 3. Implementation Steps
|
||||
|
||||
#### Step 1: Add Shadow Mode Flag
|
||||
- ENV: `SHADOW_MODE=true`
|
||||
- When true, `check-risk` and `position-manager` log decisions but do **not** place orders.
|
||||
|
||||
#### Step 2: Instant Reversal Detection
|
||||
- In `check-risk`, before quality gate:
|
||||
- Fetch last completed 5m candle (from Pyth or cache).
|
||||
- Compute body size: `abs(close - open)`.
|
||||
- Compute reversal: if signal is long and prior candle body was strong down (close < open and body > k·ATR), flag `instantReversalDetected=true`.
|
||||
- Similarly for short.
|
||||
- Suggested k=0.5 initially; tune based on results.
|
||||
|
||||
#### Step 3: HTF Alignment Check
|
||||
- In `check-risk`, after quality gate:
|
||||
- Fetch most recent 15m `BlockedSignal` direction.
|
||||
- If 5m signal direction == 15m blocked direction AND quality < 85, flag `htfBlocked=true`.
|
||||
- Else `htfAligned=true`.
|
||||
|
||||
#### Step 4: Shadow Entry Decision
|
||||
- Combine checks:
|
||||
- `wouldEnter = (Q >= 95) AND htfAligned AND !instantReversalDetected AND (other risk checks pass)`
|
||||
- If `wouldEnter=false`, set `blockReason` and write to `ShadowTrade` with `entryPrice=null`.
|
||||
- If `wouldEnter=true`, write to `ShadowTrade` with `entryPrice` and start shadow monitoring.
|
||||
|
||||
#### Step 5: Shadow Monitoring
|
||||
- Position Manager polls every 10s:
|
||||
- Fetch current Pyth price.
|
||||
- Check TP1, TP2, SL, trailing SL (same logic as live).
|
||||
- Check 5-candle time: if 25 minutes elapsed and MFE < $30, exit with `TIME_EXIT_5_CANDLE`.
|
||||
- On exit, compute `simulatedPnL` (with fees/slippage) and update `ShadowTrade`.
|
||||
|
||||
#### Step 6: Daily Reports
|
||||
- Cron job or API endpoint `/api/trading/shadow-report`:
|
||||
- Aggregate `ShadowTrade` by day:
|
||||
- Total signals, would-enter count, per-block-reason counts.
|
||||
- Win rate, avg PnL, total PnL, max drawdown.
|
||||
- Per-direction breakdown.
|
||||
- Push to Telegram or email.
|
||||
|
||||
### 4. Validation Criteria (4–6 weeks, ≥100 trades)
|
||||
|
||||
#### Go criteria:
|
||||
- Net simulated PnL > 0 after all costs.
|
||||
- Win rate > 40% or expectancy > $5/trade.
|
||||
- Max drawdown < 20% of starting capital.
|
||||
- Return-to-drawdown > 1.0.
|
||||
- Metrics stable across 2-week rolling windows.
|
||||
|
||||
#### No-go criteria:
|
||||
- Net PnL < 0 or expectancy < 0.
|
||||
- Max drawdown > 30%.
|
||||
- Hit rate < 35% with large tail losses.
|
||||
- Sensitivity tests show parameter brittleness.
|
||||
|
||||
### 5. Dashboard (Optional)
|
||||
- Build Next.js page `/shadow-dashboard`:
|
||||
- Real-time counters: signals today, would-enter today, blocked by reason.
|
||||
- PnL chart: cumulative simulated equity curve.
|
||||
- Table: last 20 shadow trades with entry/exit/PnL.
|
||||
- Filters: date range, direction, block reason.
|
||||
|
||||
### 6. Rollout After Validation
|
||||
- Disable shadow mode: `SHADOW_MODE=false`.
|
||||
- Deploy unified Q≥95, instant reversal filter, HTF enforcement, 5-candle exit to live `check-risk` and `position-manager`.
|
||||
- Monitor live logs for first 48h with small position sizes.
|
||||
- Compare live vs shadow metrics; revert if divergence or losses exceed kill-switch.
|
||||
|
||||
## Files to Create/Edit
|
||||
1. `prisma/schema.prisma`: Add `ShadowTrade` model.
|
||||
2. `app/api/trading/check-risk/route.ts`: Add instant reversal + HTF checks, shadow logging.
|
||||
3. `lib/trading/position-manager.ts`: Add shadow monitoring loop and 5-candle exit.
|
||||
4. `app/api/trading/shadow-report/route.ts`: Daily aggregation endpoint.
|
||||
5. `app/shadow-dashboard/page.tsx`: (Optional) Real-time UI.
|
||||
6. `.env`: Add `SHADOW_MODE=true`.
|
||||
|
||||
## Timeline
|
||||
- Week 1: Implement shadow logging, instant reversal, HTF checks.
|
||||
- Week 2: Deploy shadow mode; verify logging and first results.
|
||||
- Weeks 3–6: Collect ≥100 shadow trades; daily review.
|
||||
- Week 7: Validation decision; if pass, deploy live with small size.
|
||||
- Week 8: Scale up if live matches shadow.
|
||||
|
||||
## Risk Mitigations
|
||||
- Shadow mode ensures no capital at risk during validation.
|
||||
- Kill-switches remain active in live deployment.
|
||||
- Telegram alerts for anomalies (slippage spikes, API errors, sudden drawdown).
|
||||
- Rollback plan: revert to current thresholds (90 long / 85 short) within 5 minutes if loss > $50 in one day.
|
||||
Reference in New Issue
Block a user