Execution algorithms

Definition

Execution algorithms split a parent order into child orders scheduled over time to reduce market impact — the price movement caused by one’s own trading. The canonical frame (Almgren-Chriss) states the trade-off exactly: executing fast pays impact, executing slow pays timing risk (the price drifts while you wait); every standard algo — VWAP, TWAP, POV, implementation shortfall — is a point on that curve. For the platform, the relevance is double: sizing simulations honestly requires an impact model, and the flash crash documents what a badly parameterized algo can do.

How it works / structure

• The standard menu (engine-parameterizable):
  • TWAP: equal slices over a time window — simple, predictable, gameable if detected.
  • VWAP: slices proportional to the historical intraday volume curve (indicator-vwap as benchmark and schedule) — the institutional benchmark standard.
  • POV (participation): trade a fixed percentage of live volume — adapts to activity, UNBOUNDED in price (the flash-crash seller’s documented configuration: volume-tied, price-insensitive — episode-flash-crash-2010).
  • Implementation shortfall (arrival price): front- loaded schedule minimizing expected total cost vs the decision price (the Almgren-Chriss solution shape); the honest benchmark because it charges for delay.
• The impact model (simulation-relevant): documented empirical regularity — impact scales roughly with the square root of order size relative to volume (labeled: robust stylized fact, exact form debated); temporary vs permanent impact decomposition; spread cost as the floor (ms-slippage-friction parameters).
• Retail translation: the same logic at small scale — slicing entries (mgmt-scaling), marketable limits, and avoiding predictable time-of-day patterns capture most of the value without infrastructure.

When it applies

Any order large relative to typical volume (the threshold where impact exceeds spread cost); simulation cost models (a backtest without spread + impact terms is documented fiction at size — ms-slippage-friction); benchmark selection (VWAP-chasing when the decision price was hours earlier is self-deception the shortfall benchmark exposes).

Risk profile & failure modes

• Price-insensitive participation: POV without limit bands follows the market anywhere — the documented flash-crash configuration; every algo needs a price leash.
• Predictability leakage: static schedules (TWAP, known VWAP curves) can be detected and front-run — randomization exists because the pattern is the information.
• Benchmark gaming: an algo can beat VWAP while losing badly vs the decision price — benchmark choice drives behavior; shortfall keeps the score honest.
• Urgency mismatch: alpha with a short half-life executed patiently forfeits the edge; slow-decay theses executed urgently pay unnecessary impact — the Almgren-Chriss risk-aversion parameter IS the thesis’s decay rate restated.

Evidence & limits

Almgren-Chriss (2001) is the canonical framework; the square-root impact regularity is widely documented (labeled stylized fact); the flash-crash algo configuration is CFTC-SEC-report documented. Proprietary algo internals vary; the entry carries the public taxonomy and the parameterization discipline.

Falsifiable-thesis examples

Illustrations only, not signals:

• “Orders above 2% of ADV executed via schedule beat single-print execution by more than the spread cost on average (impact-management check)” — falsified by the paired execution log.
• “This strategy’s simulated edge survives a square-root impact model at intended size (capacity honesty)” — falsified by the re-run.

Cross-references

• The cost floor: ms-slippage-friction, ms-liquidity
• The benchmark: indicator-vwap; the primitives: ms-order-types
• The retail form: mgmt-scaling
• The cautionary record: episode-flash-crash-2010