Capstone Lab: The Order Desk Challenge

How a desk writes an execution plan

An execution plan is written in a fixed order, and starting in the middle is the most common way to get it wrong. First, the **mandate and the benchmark**: what is the client trying to achieve, and what will they measure it against? Arrival price for a decision, VWAP for a passive order, the close for an index fund. Naming the benchmark first is not a formality — it is the choice that decides which costs are acceptable, and it has to be made **before** the trade, because a benchmark chosen afterwards is a story rather than a measurement. Second, the **constraint inventory**: size against average volume, the spread, any scheduled event (earnings, an index rebalance, a corporate action with an ex-date), any borrow or margin limit, and any halt or reopening. Third, the **plan per ticket**: order type and time-in-force, venue, and schedule. Fourth, the **cost estimate** in dollars and basis points, because a number you predicted is the only way to find out whether the day went badly or you simply did not understand it. Fifth, the **review**: fill against the benchmark, and a note on what you would change. A worked example shows how tightly the constraint drives the answer. *Ticket 2, the index deletion: sell 45,000 shares of an $18.40 stock, 6¢ spread, average daily volume 200,000 shares, so the order is 22.5% of a day; the client is a fund that must hold zero of this name by the close, and its benchmark is VWAP.* The constraint that matters is that the exit is mandatory and the stock cannot absorb it without impact — nearly a quarter of a day in one name. The plan: a limit order resting near the bid, worked through the day with a participation cap of about 10% of volume, a mild willingness to cross when the book thins, and the residual swept in the closing auction, where the client’s benchmark actually lives and where the cross can absorb size without walking the book. What the plan refuses to do is send a 45,000-share market order, because the modelled impact of 19 bps on a 22.5% participation is a cost the desk chose rather than one the market imposed. The plan, in five lines — Mandate and benchmark: Must be flat by the close · measured against VWAP · Constraints: 22.5% of average volume · 6¢ spread · no event risk · Order type and venue: Passive limit, POV around 10%, residual in the closing auction · Predicted cost: ≈ 35 bps, $2,900 — with the spread the largest component ← · Review: Fill against VWAP; the question is whether patience was cheaper than crossing

The five tickets, and the decision each one forces

**1. The index addition — 120,000 shares at $38.20, 2¢ spread, 3% of a 4-million-share day.** This is the easy ticket and the plan should be short. The stock is liquid, the spread is small, and there is a known event (the rebalance date) that means other desks are trading the same side. The decision is *when*, not *how*: work it with a schedule, and if the fund is benchmarked to the close, use the auction and stop thinking about it. Predicted cost around 9.5 bps, mostly impact from the size rather than the spread. **2. The index deletion — 45,000 shares at $18.40, 6¢ spread, 22.5% of a day.** The spread is 16.3 bps of a 35.3 bp total, so the lever is patience. Work it passively, cap participation, and sweep the remainder into the close where the cross can absorb what the continuous book cannot. This is the ticket where a good desk and a careless one diverge by the most money per dollar traded. **3. The small cap reporting tomorrow — 8,000 shares at $7.90, 4¢ spread.** The worst rate in the blotter (about 37 bps) and only about $235 of execution cost — which is the point: the real exposure is the earnings print, where a normal overnight gap dwarfs every execution decision available to you. Either the position is right to hold through the event or it is not, and that is a **position** decision to escalate to the client, not an execution problem to solve with order types. A desk that spends the morning optimising this ticket has optimised the wrong thing. **4. The large cap into the close — 60,000 shares at $210.00, 3¢ spread, 0.5% of a day.** The largest dollar cost of the day ($4,460) and the smallest rate (about 3.5 bps). At 0.5% of volume, impact is nearly irrelevant, so the decision is about the benchmark: if the client is measured against the close, the auction is both the target and the cheapest venue; if not, working it passively costs almost nothing and carries little risk. Do not spend attention here. **5. The halted stock — 5,000 shares at $52.00, reopening this afternoon.** Nothing about the order changes what will happen: the reopening auction sets one price from everyone who wants it (M12), and the only decisions are whether to participate in the cross, what to do with any residual order the halt left behind, and whether the client understands that a stop or a market order into a reopen is the worst possible instrument. The plan is documentation and communication, and the cost estimate is honest uncertainty rather than a number. Notice the pattern across all five: the ticket with the biggest dollar cost is rarely the one with the most decision content, and the ticket with the worst rate can still be too small to matter. Attention goes where the plan changes the outcome.

The memo, the review, and what is actually graded

A **best-execution memo** is the document that makes the plan auditable, and it has a settled shape: the mandate and the benchmark for each ticket, the constraints recorded before trading, the order type and venue chosen with the reason, the predicted cost, and afterwards the fill measured against the benchmark it committed to. The last point is the discipline that separates a desk from an order-sender: the benchmark is chosen **before** the trade, so the review cannot be fitted to the outcome. A review that says "we got a great price" without naming the benchmark has said nothing; a review that says "fill of $72.86 against an arrival of $72.50 is 49.7 bps of shortfall, and we beat VWAP because the market fell while we worked" is an analysis someone can learn from. The review is also where the **next** decision gets better. Three questions do most of the work: was the cost what you predicted, and if not, which component — spread, impact or delay — was responsible? Did the schedule match the constraint, in hindsight: was the ticket a patience problem that you worked urgently, or an urgent one you worked too slowly? And did any assumption fail — a halt, a gap, a liquidity withdrawal — in a way the plan should have anticipated? Those answers are what make the fifth ticket next week cheaper. Finally, be honest about the limits, because a capstone is a place to demonstrate judgment rather than certainty. Some of these outcomes are not controllable: the reopening auction price, an overnight gap on the earnings name, whether other desks crowded into the same rebalance. The standard is not a good fill — it is a plan whose reasoning survives contact with the tape, revised by evidence rather than by results. That is the same standard the whole subject has been building toward: mechanism first, then the number attached to it, then the judgment of which number should move. The most common capstone failure is not a bad order type; it is a plan written after the fact. Commit to the benchmark and the predicted cost in writing before the day starts, or the review measures nothing.

Sources

• Implementation shortfall, participation and the schedule that follows the costKissell — The Science of Algorithmic Trading and Portfolio Management
• Best execution as a duty, and the memo that documents itSEC — best execution guidance and Rule 605/606 disclosures
• The closing auction as the cheapest venue for sizeNYSE / Nasdaq — closing auction mechanics and volume statistics
• Trading into a reopening after a haltSEC — Limit Up-Limit Down Plan and reopening auction procedure
• Pre-registering the benchmark, and why the choice is a claim about the jobCFA Institute — trade evaluation and benchmark selection