A great strategy with broken execution wrecks live PnL fast. The execution layer is the last defense between strategy output and real accounts — it decides how to slice orders, when to circuit-break, and how to handle anomalies. This post covers the core discipline (methodology only).
Directly market-ordering a large parent order causes two problems: 1. Market impact: Large lots move price against you on entry — actual fill far worse than mid 2. Signaling: Counterparties see your size and front-run
Algorithmic orders slice the parent into child orders by rule: - TWAP — Time-weighted, uniform pace - VWAP — Volume-weighted to market distribution - POV — Percent of Volume, track market's actual pace - IS — Implementation Shortfall, optimize impact vs timing risk
China A-shares add constraints: T+1, daily limits, matching rules. US-style algo implementations cannot be copy-pasted to A-shares.
| Dimension | Trigger action |
|---|---|
| Daily loss cap | → Stop opening new positions today |
| Single-stock drawdown | → Stop-out that name |
| Portfolio volatility | → Reduce gross exposure |
| Limit-order price deviation | → Reject if >X% off mid |
| Anomalous PnL jump | → XDXR-misalignment fake gap, don't stop |
| Data staleness | → Pause new trades |
The last two are especially landmine: overnight positions hitting XDXR (adjustment article) and fake signals from stale data — execution must have explicit checks, not blindly execute strategy output.
If any link in strategy or execution chain anomalies (data feed dead, model timeout, order ack abnormal, consecutive losses exceed threshold) → immediately stop new trades, wait for human confirmation. "Better not trade than wrong-trade" is the execution-layer rule — especially under licensed-advisor compliance, the risk floor is non-negotiable.