Transaction Cost Analysis and Market Impact

A trade has no cost of its own

Ask what a trade cost and there is no answer until you name a benchmark, because a cost is a **difference**. Compare your average fill with the mid when you sent the order (**arrival price**) and you measure the whole job. Compare it with the day's **VWAP** and you measure whether you traded better or worse than the average print while you were working. Compare it with the **close** and you measure the value of having waited until the auction. The three numbers can disagree in sign on the same trade, and every one of them can be computed correctly. What is being measured is not one cost but four, and they behave differently. **Commissions and fees** are explicit, and since 2019 they are usually zero for retail equities while institutional rates are a fraction of a cent per share — the smallest part, and the only part that appears on a statement. The **spread** is the cost of immediacy: you cross it whenever you demand liquidity (M3). **Market impact** is the price move your own order caused, and it is the only cost that is caused by *you* rather than by the market — a small order has none and a large one is almost entirely this. And **delay** (or timing risk) is what the market does to you while you wait: it is not paid to anyone, which is why it never appears in a report and why a desk that trades slowly for a month can lose far more than the spread ever cost. For a retail order the ordering is usually spread, then fees, then impact. For an institution buying a month of volume the ordering inverts completely: impact and delay dominate, and the spread is rounding. That is why "how much did it cost?" has an answer that starts with "how big was it?". One round trip, four costs, retail size — Commission, 200 shares: $0.00 — explicit and now usually zero · Spread, 2.0¢ on a $72 stock: ≈ $4.00 · 2.8 bps · Impact, 200 of 4,000,000 shares a day: ≈ $0 — you are 0.005% of the day · Delay, held a day at 1.8% daily vol: ± $260 · 1 sigma ← The column on the right is the point: the largest line in a retail trade's cost is often the one nobody charges — the price you did not get because you were early or late.

Impact is a shape, not a line

The naive model says trading twice as much costs twice as much. It does not. Fitted across decades of real fills, impact grows with roughly the **square root** of participation — the share of the day's volume you take. Taking 1% of a day's volume might cost 4 bps; 4% costs about 8 bps, not 16; 16% costs about 16 bps, not 64. The shape matters because it makes size *sublinear* in cost: the first slice of a large order is nearly free and the last slice is very expensive. The mechanism is that liquidity is not a warehouse. What you see resting on the book is a commitment that other people are free to withdraw the moment your order shows up, which is why the depth you walked at M4 was resting liquidity, not promised liquidity. Your order signals that someone knows something, market makers widen or step away, and the price moves partly for real and partly to make you go away. The part that persists is **permanent** and is a genuine price change; the part that fades back after you stop is **temporary**, and it is a fee you paid the market for immediacy. Two strategies follow. If impact is the problem, **slow down**: slice the order, work it in proportion to volume (**POV**), or follow a time schedule (**TWAP**) or a volume schedule (**VWAP**). If risk is the problem, **speed up**: take liquidity, cross the spread, or send the whole order to an auction where everyone is forced to one price. A good desk does not pick one philosophy; it allocates the order between them in proportion to how fast the edge decays relative to volatility. And it separates the decision to own the stock from the decision about the moment, which is the reason execution is a desk and not a person. The same 500,000 shares, two speeds — All at once — 12.5% of the day: 40 × √0.125 = 14.1 bps · ≈ $50,900 · Over five days — 2.5% a day: 40 × √0.025 = 6.3 bps · ≈ $22,800 · The saving from patience: ≈ $28,000 · The risk taken for it — 1σ over five days: ≈ $720,000 ← The risk of waiting is not a metaphor. On an average exposure of 250,000 shares at $72 with 1.8% daily volatility, one standard deviation of adverse drift over five days is roughly $720,000 against $28,000 of saved impact — twenty-five times larger. Patience is a bet, and it is usually the bigger bet.

Choosing how to work it — and who is watching

An execution algorithm is a rule for balancing the three costs. **VWAP** schedules the order to finish near the day's average price and is judged against that benchmark; it is a good choice when you are too small to matter and your goal is to avoid doing worse than the crowd. **TWAP** spreads evenly over time and makes no claim about volume. **POV** follows the tape and participates at a fixed share of volume, which is the natural choice when impact is the binding cost and the schedule should bend to liquidity rather than to the clock. **Implementation shortfall** algorithms weight the order towards the *front*, because they treat the deviation from the arrival price as the thing to minimise — and they trade more urgently as the price moves against the order and less urgently as it moves in favour. Then the venue choice returns, now with a cost attached. The lit exchange is where you pay the spread but can see the depth; a **dark pool** or **ATS** lets you rest at the midpoint without displaying, which is cheaper when your information is short-lived and useless when it is not; a **block** in the dark can move hundreds of thousands of shares in one print with no impact at all, if a counterparty happens to be there. And the **closing auction** is where the largest, cheapest trades in the market happen, because everyone settles at a single computed price and nobody can front-run it (M11). For an index fund whose mandate is the closing price, the auction is not a strategy; it is the target. Finally, someone is auditing all of this. **Best execution** is a duty, not a result (M9), and the disclosures that make it checkable have grown: Rule 606 requires brokers to say where they routed customer orders and how they were paid, and the expanded Rule 605 statistics describe the quality they delivered. Once you know what a fill costs against a benchmark, that disclosure stops being paperwork and becomes the only evidence you have about the venue your broker chose for you. A practical test for any execution decision: name the benchmark first, then the plan. "Get it done today" and "beat the close" are different mandates that produce opposite schedules on the same order.

Sources

• Implementation shortfall, VWAP and participation: the vocabulary of trade costKissell — The Science of Algorithmic Trading and Portfolio Management
• Temporary and permanent impact, and optimal execution over timeAlmgren & Chriss — Optimal execution of portfolio transactions
• Price impact grows with the square root of order sizeTóth et al. — Anomalous price impact and the critical nature of liquidity in financial markets
• Best execution, order routing and the Rule 606 disclosureSEC — Rule 606 order routing reports and best-execution guidance
• The auction as the cheapest place to be largeNYSE / Nasdaq — closing auction mechanics and volume statistics