Execution TCA KPI Stack (Arrival → IS → Markout)

Date: 2026-02-26 (KST)

TL;DR

Most desks over-focus on one benchmark (usually VWAP) and miss the full execution picture.

For live execution ops, use a layered KPI stack:

1. Decision/arrival cost (did we start badly?)
2. Execution cost (did our slicing/routing pay too much?)
3. Post-trade markout (did we get adversely selected?)
4. Stability/operations (did infra incidents create hidden slippage?)

If you track these separately, you can fix the right subsystem instead of “tuning everything.”

1) Why one benchmark is never enough

A single number cannot answer all of these at once:

• Was the order late vs signal? (timing problem)
• Was child-order placement poor? (microstructure problem)
• Did fills predictably move against us after fill? (adverse-selection problem)
• Did retries/reconnect/throttle incidents cause degraded execution? (ops problem)

Implementation Shortfall (IS) remains a core institutional benchmark, but operationally you need additional decomposition to make IS actionable.

2) KPI hierarchy (what each metric is for)

2.1 Decision-to-arrival metrics

• Decision delay (ms/s): order decision timestamp → first eligible send timestamp
• Arrival drift (bps): side-adjusted move from decision reference to arrival mid

Use when diagnosing: signal pipeline latency, risk gate bottlenecks, manual intervention lag.

2.2 Arrival-to-fill metrics

• Arrival slippage (bps): side-adjusted fill vs arrival benchmark
• VWAP gap (bps): fill vs interval VWAP during your participation window
• Participation rate (%): executed qty / market volume during execution window
• Fill efficiency (%): executed qty / intended qty within allowed horizon

Use when diagnosing: slicing schedule, urgency policy, queueing/routing quality.

2.3 Post-fill quality metrics

• Markout curve at multiple horizons (e.g., 1s, 5s, 30s, 5m)
• Realized spread / adverse selection decomposition (if quote data available)

Use when diagnosing: toxic flow exposure, poor quote placement, wrong aggressiveness.

2.4 Operational reliability overlays

• Throttle-hit rate
• Reject/retry rate
• WS reconnect bursts
• Clock-sync drift (ms)

Use when diagnosing: non-market causes of “mystery slippage.”

3) Canonical formulas (side-adjusted)

Define side sign:

• Buy: s = +1
• Sell: s = -1

Let P_ref be benchmark price and P_exec be execution price.

General side-adjusted bps cost:

cost_bps = s * (P_exec - P_ref) / P_ref * 10,000

Positive means worse execution cost for both buys and sells.

3.1 Order-level Arrival Slippage

arrival_slippage_bps = s * (P_vwap_exec - P_arrival_mid) / P_arrival_mid * 10,000

Where P_vwap_exec is your order’s fill-weighted average price.

3.2 Implementation Shortfall (practical order form)

For intended quantity Q*, executed quantity Qe, decision reference P_decision, end-of-horizon mark P_end:

IS_value = s * [
  (Qe * (P_exec_vwap - P_decision))
+ ((Q* - Qe) * (P_end - P_decision))
] + explicit_costs

Normalize to bps by dividing by Q* * P_decision and multiplying by 10,000.

This keeps partial-fill and opportunity cost visible instead of hidden.

3.3 Markout (fill-to-mid)

At horizon h:

markout_bps(h) = s * (Mid(t_fill + h) - P_fill) / P_fill * 10,000

Interpretation:

• Positive markout for buys = price moved up after buy fill (good)
• Negative markout for buys = adverse selection / over-aggressive fill (bad)

4) Recommended horizon set (starter)

Use fixed horizons and do not change frequently, or trend comparability breaks.

• Micro: 1s, 5s
• Short: 30s, 60s
• Tactical: 5m
• Session: close markout (optional)

For KRX-like sessioned markets, include auction/post-close specific markouts separately.

5) KPI-to-action playbook

6) Dashboard layout (ops-first)

Panel A — Cost overview

• IS bps (daily/rolling)
• Arrival slippage bps
• VWAP gap bps
• Explicit fees bps

Panel B — Post-trade quality

• Markout curves by horizon
• Markout by symbol bucket / venue / time-of-day

Panel C — Execution process

• Fill efficiency
• Participation distribution
• Child order cancel/replace intensity

Panel D — Reliability context

• Reject/retry/throttle counters
• WS reconnect events
• Clock drift and stale quote incidents

Rule: every cost panel needs an operational context panel beside it.

7) Minimum data contract

interface ExecutionFillEvent {
  orderId: string
  parentOrderId: string
  symbol: string
  side: 'BUY' | 'SELL'
  qty: number
  fillPrice: number
  fillTs: string
  arrivalMid: number
  decisionRef: number
  venue?: string
  explicitFee?: number
}

interface MarketSnapshot {
  symbol: string
  ts: string
  bid: number
  ask: number
  mid: number
  vwapSession?: number
  tradedVolumeWindow?: number
}

Without reliable timestamp alignment, all TCA results become noisy. Clock discipline is part of TCA quality.

8) Weekly review cadence (lean)

1. Monday: prior week IS + arrival decomposition
2. Wednesday: markout curve drift by strategy and symbol cluster
3. Friday: choose one parameter change only (urgency band, POV cap, or venue preference), then monitor for one week

Avoid multi-parameter changes in the same week; attribution becomes impossible.

9) Common anti-patterns

• Treating VWAP outperformance as universal success
• Ignoring unfilled residual opportunity cost
• Looking only at average markout (not distribution tail)
• Mixing benchmark definitions mid-quarter
• Running TCA without incident overlays (network, auth, throttle)

10) References

• Perold, A. F. (1988), The Implementation Shortfall: Paper vs. Reality.
• Databento Microstructure Guide — Markout overview: https://databento.com/microstructure/markout
• Wikipedia summary (quick refresher) — Implementation shortfall: https://en.wikipedia.org/wiki/Implementation_shortfall
• Busseti, Boyd, et al. — VWAP execution framing: https://web.stanford.edu/~boyd/papers/pdf/vwap_opt_exec.pdf

One-line takeaway

Don’t ask “did we beat VWAP?” first — ask where cost entered the pipeline (decision, execution, or post-fill), then fix that specific layer.