Post Trade Execution Quality Scorecard

When to Use

Use this skill to build a house post-trade scorecard that ranks brokers, algo wheels and execution venues from your own executed-order records. It computes arrival-price slippage, VWAP slippage, effective spread and the effective-over-quoted ratio ($E/Q$), fill rate, a Perold implementation shortfall including opportunity cost, and a notional-weighted composite grade ($A$–$F$) per venue and overall.

Two of these statistics — effective spread and $E/Q$ — are defined the way SEC Rule 605 defines them, so your numbers are comparable in kind to a market centre's published Rule 605 report. That comparability is the point: it lets you check a venue's own claims against what you actually received.

The obligation this serves is ongoing execution-quality monitoring, not report filing: MiFID II Art. 27(7) and Art. 66 of Commission Delegated Regulation (EU) 2017/565 require an investment firm to monitor the effectiveness of its order execution arrangements and review its execution policy at least annually. Both remain in force. See references/standards.md for the jurisdictional detail.

When NOT to Use

To produce a filable SEC Rule 605 report. This engine cannot. Rule 605 is a reporting obligation on market centres, on broker-dealers introducing or carrying 100,000 or more customer accounts, and on single-dealer platforms. A filed report needs monthly aggregation, notional size categories, fractional/odd-lot/round-lot classification, price and size improvement, realized spreads at five post-execution horizons, sub-100-microsecond speed buckets, and a CSV+PDF summary. None of that is here.
To produce RTS 27 or RTS 28 reports. Both obligations have been deleted from MiFID II — do not build them. See references/standards.md.
On a batch where most orders lack an end_price. Without a mark for the unfilled residual the opportunity-cost term is unknowable and implementation shortfall is reported as None. A scorecard built only on price-based statistics systematically flatters a broker that missed half the order.
To compare a maker-rebate venue against a taker-fee venue. Every metric here is gross of commissions, fees, taxes and borrow. On these numbers alone the comparison is meaningless.
To grade a book of large worked parent orders on the default weights. The $E/Q$ penalty is calibrated for marketable orders; on worked parents it saturates every score to zero. Recalibrate eqr_penalty_per_unit first — see Common Pitfalls.
To attribute cost to timing versus sizing. The record carries one arrival price, one average fill price and one interval VWAP per parent order — points, not paths. Use execution-slippage-attribution-timing-vs-sizing.
To measure adverse selection or reversion. Realized spread needs post-trade marks this record does not carry; see adverse-selection-measurement-for-passive-orders.

Prerequisites

Executed parent-order records: order_id, venue, symbol, side ('BUY'/'SELL'), parent_qty, executed_qty, avg_fill_price, arrival_price, market_vwap, arrival_midquote, arrival_quoted_spread.
arrival_midquote and arrival_quoted_spread stamped from the consolidated quote at the time of order receipt — Rule 605's reference point — not at the time of execution.
Optional end_price: the price marking the unfilled residual, normally the last price of the trading horizon. Required for implementation shortfall; there is no safe default.
Scorecard config: benchmark_target_is_bps (default $10.0$ bps) plus the penalty weights, all house-calibrated (see references/standards.md).

Workflow

Validate the whole batch before computing anything:
- Prices must be finite and $> 0$; parent_qty $> 0$; 0 \le executed_qty $\le$ parent_qty; side in {BUY, SELL}.
- Decision point — an unrecognised side must raise, never default. Falling through to SELL inverts the sign of every cost metric: a broker that paid $50$ bps is reported as having saved $50$.
- Decision point — a locked or crossed book has no $E/Q$ denominator. A zero or negative arrival_quoted_spread must exclude the order, not be floored to a small positive number — flooring converts an unmeasurable ratio into an enormous fabricated one.
- Validation runs over the entire batch first, so a malformed record can never contribute a partial result to an aggregate.
Compute per-order metrics with $\text{SideSign} = +1$ for BUY, $-1$ for SELL, so positive always means cost: $$\text{ArrivalSlippage}{\text{bps}} = \text{SideSign} \cdot \frac{\text{AvgFill} - \text{Arrival}}{\text{Arrival}} \cdot 10^4$$ $$\text{Slippage}{\text{VWAP,bps}} = \text{SideSign} \cdot \frac{\text{AvgFill} - \text{VWAP}}{\text{VWAP}} \cdot 10^4$$ $$\text{EffSpread} = 2 \cdot \text{SideSign} \cdot (\text{AvgFill} - \text{ArrivalMid}), \qquad E/Q = \frac{\text{EffSpread}}{\text{ArrivalQuotedSpread}}$$
- Decision point — a wholly unfilled order has no fill price. Skip every price-based metric for it. Feeding a placeholder avg_fill_price of $0.0$ into the slippage formula produces a fictional $-10{,}000$ bps saving that then drags the whole aggregate down.
Compute implementation shortfall per Perold (1988) — execution cost on the shares that filled plus opportunity cost on the shares that did not: $$f = \frac{\text{ExecutedQty}}{\text{ParentQty}}, \qquad IS_{\text{bps}} = \text{ArrivalSlippage}_{\text{bps}} \cdot f + \text{SideSign} \cdot \frac{\text{End} - \text{Arrival}}{\text{Arrival}} \cdot 10^4 \cdot (1 - f)$$
- Decision point — no end_price, no IS. Report None, not the filled-share cost. A broker that filled $500$ of $1{,}000$ shares at $5$ bps while the stock ran $200$ bps away delivered $102.5$ bps of shortfall, and the filled-share number shows $5$.
Aggregate notional-weighted, never as a mean over orders:
- Price metrics weight by executed notional; shortfall and score weight by parent notional; overall fill rate is $\sum\text{ExecutedQty} / \sum\text{ParentQty}$.
- Decision point — $E/Q$ has two forms and they are not interchangeable. Rule 605 publishes a ratio of share-weighted averages, $\overline{\text{EffSpread}} / \overline{\text{QuotedSpread}}$ (as a percentage). The mean of per-order ratios is a different number: on two equal-size orders with the same $0.10$ effective spread against $0.10$ and $0.02$ quoted spreads, the ratio-of-averages is $1.67$ and the mean of ratios is $3.00$. Compare like with like when benchmarking against a filing.
Grade per venue and overall, worst venue first. A venue below min_venue_notional_for_grade is reported but graded NR — a letter grade from two odd lots is noise wearing the costume of a measurement, and desks route on letter grades.

Full procedure: see references/workflows.md. Standards reference: see references/standards.md. Printable pre-flight checklist: see assets/checklist.md.

Common Pitfalls

Calling the filled-share cost "implementation shortfall". It is the implicit cost component only. Perold's IS also charges the opportunity cost of shares never traded; omitting it is the single most common way a scorecard rewards a broker for quietly not working the order.
Averaging unweighted across orders. A $1$-share fill and a $1{,}000{,}000$-share fill carry equal weight in a plain mean, so one excellent odd lot can outrank a whole badly-executed programme. Weight by notional.
Averaging per-order ratios where the standard averages the numerator and denominator separately. Small quoted spreads blow up per-order $E/Q$ and dominate a plain mean.
Flooring a denominator instead of rejecting the record. max(0.0001, price) does not make a zero price safe; it turns it into a ~$10^9$ bps figure that enters the aggregate looking like a measurement.
A fill-rate denominator of max(1.0, parent_qty). For fractional shares, crypto or sub-unit FX lots this silently understates the fill rate — a fully filled $0.5$-unit order reports $50%$.
Letting an unfilled order's placeholder fill price into the price metrics. Exclude it from price statistics; count it in fill rate and opportunity cost.
Stamping the midquote at execution time rather than order receipt. Rule 605 measures effective spread against the quote when the order arrived. Using the execution-time quote measures something else and makes a slow, drifting execution look tight.
Relying solely on VWAP. VWAP is gameable: trade slowly into high-volume windows and the VWAP number flatters while implementation shortfall blows out.
Ignoring fill rates. Low slippage on partial fills plus a high cancel rate is not good execution.
Uncorrected side sign. For a SELL a higher fill price is a saving; the sign must invert on slippage, effective spread and opportunity cost alike.
Carrying $E/Q$ over from marketable orders to worked parent orders. Rule 605 computes $E/Q$ for individual marketable orders against the receipt-time quote, where $pprox 1.0$ is normal. A parent order worked over minutes walks the book, so $E/Q$ of $5$–$15$ is routine in a tight-spread name — at the default eqr_penalty_per_unit of $20.0$ that pins every such order to a score of $0$ and the scorecard stops discriminating between brokers. Recalibrate the weight, or set it to $0.0$ and rank on slippage and fill rate.
Presenting the composite grade as a regulatory measure. The weights and the $A$–$F$ boundaries are house convention. No regulator defines a grade, and no regulator mandates a $10$ bps IS target or a $95%$ fill rate.

Verification

Instantiate PostTradeExecutionQualityScorecard(). Input a BUY of $1{,}000$ shares fully filled @ $$100.05$ vs $$100.00$ arrival, $$100.10$ VWAP, $$100.00$ midquote, $$0.10$ quoted spread $\implies$ arrival slippage $= +5.00$ bps, VWAP slippage $= -5.00$ bps, effective spread $= $0.10$, $E/Q = 1.00$, fill rate $= 100%$, grade $A$.
Mirror check for sign: SELL $1{,}000$ @ $$99.95$ against the same $$100.00$ arrival/midquote $\implies$ arrival slippage $= +5.00$ bps and effective spread $= $0.10$ (both costs, not savings).
Opportunity cost: BUY $1{,}000$, only $500$ filled @ $$100.05$, end_price $= $102.00$ $\implies$ opportunity cost $= +100.00$ bps and $IS = +102.50$ bps, against a filled-share cost of only $5.00$ bps. Omit end_price and $IS$ must be None, not $5.00$.
Weighting: a $1$-share $0$ bps fill plus a $100{,}000$-share $100$ bps fill $\implies$ notional-weighted slippage $\approx 100.0$ bps, unweighted $50.0$ bps.
Negative checks: side='SHORT', arrival_price=0, arrival_quoted_spread=0, parent_qty=0, executed_qty > parent_qty, and a NaN price must each raise.
Run python -m unittest discover -s skills/post-trade-execution-quality-scorecard/scripts.

Post Trade Execution Quality Scorecard

When to Use

When NOT to Use

Prerequisites

Workflow

Common Pitfalls

Verification

Hands off to 8

Handed off from 9

When to Use

When NOT to Use

Prerequisites

Workflow

Common Pitfalls

Verification

Related Skills

Hands off to 8

Handed off from 9