Arrival Price Execution Algorithms
An arrival-price algorithm tries to minimize cost relative to the market when an order becomes executable. It is appropriate when a client cares about implementation shortfall and accepts that the schedule should adapt to price, volume, and risk after release. This differs from a VWAP algorithm, whose primary objective is matching a volume-weighted benchmark over a stated interval.
For a buy, the algorithm balances the expected cost of trading now against the risk that prices rise while shares remain. A compact objective is
\[ \mina E[C{\text{impact}}(a)] + \gamma\,\operatorname{Risk}(X_{\rm remaining})
- C{\text{opportunity}}(X{\rm remaining}),
\]
where \(a\) is the aggressiveness decision. The arrival midpoint is a benchmark, not a prediction that price will remain stationary.
Establish the mandate before choosing behavior
“Beat arrival” is underspecified. A robust parent-order contract defines side, maximum duration, completion target, participation cap, venues, order-type restrictions, benchmark timestamp, and escalation authority. It also identifies whether the order contains alpha: a buy with positive expected alpha should normally be more urgent than a neutral rebalance.
| Parameter | Example | Why it changes policy |
|---|---|---|
| Completion deadline | 30 minutes | Sets terminal urgency |
| Participation cap | 15% of observed volume | Limits footprint |
| Risk aversion | High | Front-loads schedule |
| Alpha decay | 20 minutes | Penalizes waiting |
| Passive tolerance | Moderate | Allows queue exposure |
| Benchmark | Release midpoint | Defines success |
The benchmark clock must be sourced from a quote that was actionable at release, with venue and data-quality rules. Otherwise a favorable score can be an artifact of stale or nontradable pricing.
State-dependent urgency
The core controller recomputes desired completion from remaining quantity, forecast liquidity, time left, volatility, and current cost estimates. It should accelerate when falling behind a feasible trajectory, not merely because price moved against the order. Chasing every adverse tick converts benchmark risk into impact.
def desired_rate(remaining, expected_volume, seconds_left, cap,
risk_multiplier=1.0):
feasible = remaining / max(expected_volume, 1.0)
time_pressure = remaining / max(seconds_left, 1.0)
return min(cap, max(feasible * risk_multiplier, time_pressure))
In production, expected_volume must be a point-in-time forecast for the remaining interval, not final-day ADV. Rate units must be consistent: shares per second cannot be compared directly with a fraction-of-volume cap.
Aggressive versus passive allocation
An arrival algorithm needs a venue- and order-type allocator. Aggressive child orders provide completion certainty but pay spread and immediate impact. Passive orders can capture spread, yet carry nonfill and adverse-selection risk. Estimate their expected value using conditional fill probability and post-fill markout, not headline maker rebates.
\[ V{\rm passive}=pf(\text{rebate}+\text{spread capture}-\text{markout}) -(1-pf)C{\rm fallback}. \]
Book state matters: a thin, canceling offer and rising buy intensity may justify crossing for a buy; a stable deep queue can support a passive slice. The relevant signals are described in order-flow imbalance signals. Do not model displayed depth as certain fillable liquidity.
| Market state | Conservative response |
|---|---|
| Ahead of schedule, stable queue | Post or reduce rate |
| Behind schedule, normal depth | Increase participation gradually |
| Spread widens, volatility jumps | Re-estimate cost; reduce stale quotes |
| Toxic one-sided flow | Limit passive exposure; enforce urgency rules |
| Near deadline | Escalate according to explicit completion policy |
Protect against information leakage
Repeatedly signaling the same size, price, and cadence can reveal a parent order. Randomize child sizes and timing within bounds, use venue diversity where appropriate, and avoid cancel-replace churn that advertises interest. Randomization is not a substitute for a footprint model: a 20% participation buy remains detectable.
Monitor market response to the algorithm's own activity. If signed price response, spread, or rejection rate exceeds forecast, lower participation or pause according to predefined controls. A policy must never “solve” deteriorating liquidity by endlessly raising its own cap.
Evaluate with benchmark-consistent TCA
Compare realized arrival shortfall to a pre-trade distribution conditioned on order size, urgency, liquidity, and market state. Segment performance by participation, passive fraction, time-to-deadline, and volatility regime. A lower average shortfall may simply reflect lower urgency or easier names.
Measure opportunity cost of unfilled shares at the stated cutoff. An algorithm that improves average fill price by abandoning difficult orders is not better. The full attribution framework is in implementation shortfall; basic schedule comparisons are in execution algorithms.
Key takeaways
- Arrival-price execution manages decision-to-completion risk, rather than mechanically matching market volume.
- Define the benchmark, deadline, caps, and completion policy before tuning urgency.
- Adapt rates to remaining feasibility and state-dependent cost, not every unfavorable tick.
- Passive allocation requires calibrated fill and markout models; displayed depth is not guaranteed liquidity.
- Score the algorithm against forecast shortfall distributions, including residual opportunity cost.