Skip to content
Menu
Finabase Learn

Hedge funds and AI · Quants and high-frequency trading

Execution algorithms: TWAP, VWAP, POV and implementation shortfall

Why funds split big orders, how TWAP, VWAP, percent-of-volume and shortfall algorithms schedule them, and a computed day buying 400,000 shares with impact and drift.

On this page
  1. Why big orders are split
  2. The four common algorithms
  3. A computed day: four ways to buy 400,000 shares
  4. Measuring the result
  5. Where execution algorithms fail
  6. How Finabase does it
  7. Frequently asked questions
  8. Summary
  9. Check your understanding

A pension fund that decides to buy 400,000 shares of a stock cannot simply send one market order. If the stock trades 4 million shares on a typical day, that order is a tenth of the day's volume. Sent all at once it would sweep through every displayed offer and keep climbing, and everyone watching would see a big buyer arrive. So the fund, or its broker, cuts the order into thousands of small pieces and feeds them to the market over hours. The software that decides how many shares to send, when, where and at what price is an execution algorithm.

Four designs cover most of what brokers offer: TWAP, which trades evenly through time; VWAP, which trades in step with the day's expected volume; percent of volume (POV), which trades a fixed share of whatever actually trades; and implementation shortfall, which trades faster to limit the risk that the price runs away. This lesson explains each one and then runs all four on the same invented day so the trade-off between market impact and price drift is visible in dollars.

Why big orders are split

An order costs more than the commission. It pays half the spread when it crosses to the other side, and if it is large compared with what is available at each price, it pushes the price as it fills. Other traders who notice persistent buying raise their offers. That extra cost is market impact, and Liquidity and slippage: volume, depth, spread and market impact explains the rough rule that it grows with the square root of the order's share of volume.

Spreading the order out lowers the pressure of each piece. It also creates a second cost: while the fund waits, the price can move for reasons that have nothing to do with the order. If the stock drifts up during the day, the shares bought in the afternoon cost more. Every execution choice balances these two costs, impact and drift. The total, measured against the price when the fund decided to trade, is implementation shortfall.

The four common algorithms

SEC staff described the core broker algorithms in a 2020 report to Congress as VWAP, TWAP, minimum implementation shortfall and trading at a set percentage of market volume, noting that many brokers offer similar versions of each and let clients customize them.

TWAP: equal slices through time

A time-weighted average price algorithm divides the order evenly across the period. Buying 400,000 shares in thirteen half-hours means about 30,769 shares every half hour, broken into smaller child orders inside each one. It needs no volume forecast, which suits thin or around-the-clock markets. Its weakness is that it trades as much at lunch, when the market is quiet, as at the open, so its share of volume swings widely.

VWAP: follow the volume curve

US stocks trade most heavily near the open and the close and least around midday. A VWAP algorithm forecasts that curve, usually from past days, and schedules each slice in proportion to it: more at the edges of the day, less at lunch. The aim is an average price close to the day's volume-weighted average price, the benchmark many funds use to judge their traders.

POV: a fixed share of actual volume

A percent-of-volume algorithm, also called participation, watches the volume that actually trades and takes a set fraction of it, say 10%. If news doubles volume, it buys twice as fast; if the market goes quiet, it slows down. It adapts where VWAP cannot, but its finishing time is uncertain, and a POV order with no price limit is dangerous in a falling market. In the 2010 Flash Crash a mutual fund complex sold 75,000 E-mini futures contracts with an algorithm set to 9% of the previous minute's volume, with no regard to price or time. As volume exploded, it sold faster, finishing in about 20 minutes a sale that had once taken more than five hours (The 2010 Flash Crash: what happened on May 6 and what changed).

Implementation shortfall: trade faster to limit drift

An implementation-shortfall algorithm, also called an arrival-price algorithm, measures itself against the price when the order arrived. Because every minute of waiting adds drift risk, it usually trades harder early and slows as the order shrinks, accepting more impact in exchange for less exposure. How hard it front-loads depends on the stock's volatility and the client's urgency.

A computed day: four ways to buy 400,000 shares

The example below is invented. A stock trades at $40.00 when the order arrives at 9:30. The forecast is 4,000,000 shares for the day, U-shaped by half hour, from 560,000 in the first half hour to 200,000 at 12:30 and 1:00 and back to 560,000 in the last. Actual volume differs a little: the open is lighter than expected and news lifts 1:00 volume by 60%, so the day totals 4,130,000 shares. The market price, before the order's own pressure, drifts up from $40.00 to $40.32.

Each slice pays the half-hour's market price plus impact, modeled as 0.5 times a 1.5% daily volatility times the square root of the slice's share of that half hour's volume. The schedules:

  • TWAP: 30,769 shares every half hour.
  • VWAP: the forecast curve, 56,000 at the open and the close, 20,000 at 12:30 and 1:00.
  • POV 10%: a tenth of actual volume each half hour; it reaches 400,000 shares in the last half hour, buying the last 48,600 shares there instead of the 61,600 that a tenth of its volume would allow.
  • Front-loaded, an implementation-shortfall style: 60% in the first two hours (72,000, 60,000, 56,000 and 52,000 shares), the rest spread thinly after.
Buying 400,000 shares in one day: shares bought each half hour TWAP VWAP POV 10% 31 56 50 9:30 31 36 34 10:00 31 30 30 10:30 31 26 26 11:00 31 24 23 11:30 31 22 20 12:00 31 20 20 12:30 31 20 32 1:00 31 22 26 1:30 31 24 24 2:00 31 28 28 2:30 31 36 38 3:00 31 56 49 3:30 Thousands of shares. Invented stock: forecast day volume 4,000,000, U-shaped; a news burst lifts 1:00 volume 60%.
Invented stock: forecast volume 4,000,000 shares, U-shaped; a news burst lifts 1:00 volume 60%. Thousands of shares. Computed in the Finabase figure script.

The results, measured against the $40.00 arrival price:

Schedule Average price Cost vs arrival Market impact Price drift Highest share of volume
TWAP $40.2715 $108,613 (67.9 bp) $39,074 $69,538 15.5%
VWAP $40.2655 $106,184 (66.4 bp) $37,744 $68,440 11.1%
POV 10% $40.2684 $107,363 (67.1 bp) $37,593 $69,770 10.0%
Front-loaded $40.2272 $90,876 (56.8 bp) $42,356 $48,520 20.0%
Cost against the $40.00 arrival price: market impact and price drift Market impact (our own pressure) Price drift while we waited $39.1k $69.5k TWAP $37.7k $68.4k VWAP $37.6k $69.8k POV 10% of volume $42.4k $48.5k Front-loaded Impact = 0.5 × 1.5% daily volatility × √(slice ÷ half-hour volume). Invented prices rise from $40.00 to $40.32.
Cost of buying 400,000 shares against the $40.00 arrival price on an invented rising day. Impact model and prices are illustrative. Computed in the Finabase figure script.

Three things stand out. First, VWAP and POV spread the order more evenly across the market's own activity than TWAP, so their impact is about $1,300 to $1,500 lower. Second, on this rising day drift is the larger cost for every steady schedule, and the front-loaded one, which pays the most impact, saves $15,000 to $18,000 overall by buying before the price climbs. Third, the ranking depends on the day.

Run the same schedules on the mirror-image day, when the price falls from $40.00 to $39.68, and the order comes in below the arrival price: the steady schedules average about $39.92, around 19 to 20 basis points under $40.00, while the front-loaded schedule averages $39.9841, only 4 basis points under. Its early buying, which helped on the rising day, now costs it about $25,000 compared with the others. An implementation-shortfall algorithm is a bet that avoiding drift risk is worth the extra impact, not a way to avoid cost.

Measuring the result

Funds judge executions against benchmarks, and each one rewards a different behavior.

  • VWAP asks whether the fund paid about what the market paid. It is easy to understand, but a large order is part of the day's volume, so its own buying pushes the benchmark up. In this example the day's VWAP before the order's own pressure is $40.1790, and every schedule pays 12 to 23 basis points above it. Beating VWAP can also be gamed by trading slowly on a trending day.
  • Arrival price (implementation shortfall) asks what the decision really cost, including drift and any shares never bought. It is harder to beat and noisier day to day.
  • The close matters to index funds valued at the closing price, which can trade in the closing auction instead.

Brokers that route orders in bulk must conduct regular and rigorous reviews of execution quality, at least quarterly under FINRA Rule 5310, and comparisons against these benchmarks are one way they do it (Order routing and best execution: NBBO, Rule 611, PFOF and 606).

Where execution algorithms fail

  • Predictability. Regular slices are easy to detect. Other traders who spot a steady buyer can raise their offers ahead of it, so most algorithms randomize size and timing and hide in Dark pools: how off-exchange trading works and how much of it there is when they can.
  • No price limit. A volume-following algorithm with no limit will keep selling into a crash, as the Flash Crash showed.
  • Wrong forecasts. A VWAP schedule built on a normal day's curve misfires on an index rebalance or an earnings day, when volume is far from normal.
  • The model is not the market. Real impact is larger in stressed markets and smaller in calm ones, and it partly persists after the order is done. The square-root model in this example is a rough rule with wide error bars.
  • Fees and venue choice. Where the child orders go, and whether they post or take liquidity, changes the cost too; Maker-taker fees and rebates: how US stock exchanges charge for trades covers that layer.

How Finabase does it

Finabase does not execute orders, so it has no execution algorithms. What it does show is the benchmark: the chart's VWAP (HLC/3) study. On the 1-day chart, built from 1-minute bars, it is the classic session VWAP that institutional traders compare their fills with. On the 5-day chart it accumulates from the first of the five days, and on daily charts it is anchored at the start of Finabase's two-year daily history, so it is not a trading-day benchmark there. The paper account at /trading offers market and limit orders filled on the next matching trade at live prices, which is enough to practise splitting an order by hand but does not model impact.

Frequently asked questions

What is the difference between TWAP and VWAP?

TWAP trades the same amount in each period, ignoring volume. VWAP trades in proportion to the expected volume in each period, so it buys more near the open and close and less at midday, which usually lowers market impact in US stocks.

What is a POV algorithm?

A percent-of-volume algorithm buys or sells a fixed share of the volume that actually trades, such as 10%. It speeds up when the market is busy and slows when it is quiet, so its finishing time is not fixed.

Is beating VWAP a sign of good execution?

Not on its own. A large order raises the VWAP it is measured against, and on a trending day a trader can beat VWAP simply by trading slowly while the cost against the decision price grows. Implementation shortfall measures that cost directly.

Why do institutions not just use limit orders?

They do, inside the algorithm. Each child order is often a limit order, posted or sent to take a price. The algorithm decides how many to send, when and where; a single large limit order would show the market the fund's whole intention.

Summary

Execution algorithms exist because a large order moves the price it trades at. TWAP spreads an order evenly through time, VWAP follows the expected volume curve, POV takes a fixed share of actual volume, and implementation-shortfall algorithms trade faster to limit the risk that the price drifts away. In this lesson's invented day, the steady schedules each cost about $106,000 to $109,000 to buy 400,000 shares against a $40.00 arrival price, and the front-loaded schedule about $91,000; on the mirror-image falling day the front-loaded schedule did worst. The right choice depends on the order's size, the stock's volatility and the fund's urgency, and the benchmark it is judged against.

Check your understanding

4 questions on this lesson. Each answer comes with the reason, and nothing is scored.

  1. A fund must buy 260,000 shares over 13 half-hour periods. The stock usually trades most at the open and close. Which schedule buys the same 20,000 shares in every half hour?

    1. TWAP
    2. VWAP
    3. Percent of volume
    Show the answer

    A. TWAP TWAP divides the order evenly through time: 260,000 ÷ 13 = 20,000 a period. VWAP would buy more near the open and close, where volume is heavier, and POV would follow whatever volume actually trades.

  2. A POV algorithm is set to 10% of volume. In one half hour 300,000 shares trade, and in the next, after news, 600,000. Roughly how much does it buy in each?

    1. 30,000 and then 60,000 shares
    2. 30,000 in both, because the target is fixed
    3. Nothing in the second, because volume spiked
    Show the answer

    A. 30,000 and then 60,000 shares A percent-of-volume algorithm takes a fixed share of actual volume, so it doubles its pace when volume doubles. That adaptability is also its danger: with no price limit it speeds up exactly when a market is moving fast.

  3. An implementation-shortfall algorithm buys 60% of an order in the first two hours. Which statement is true?

    1. It pays more market impact to reduce the risk of the price drifting away; on a day the price falls, it can do worse than VWAP
    2. It always costs less than VWAP because it finishes sooner
    3. It avoids market impact by trading when the market is busiest
    Show the answer

    A. It pays more market impact to reduce the risk of the price drifting away; on a day the price falls, it can do worse than VWAP Trading faster raises each slice’s share of volume and so its impact, in exchange for less exposure to drift. In the lesson’s example it saved $15,000 to $18,000 on a rising day but cost about $25,000 more than the steady schedules on the falling one.

  4. Why can a large order look good against the day’s VWAP and still be a costly trade?

    1. Its own buying pushes VWAP up, and VWAP ignores how far the price moved from the decision price
    2. VWAP is calculated only from the opening auction
    3. VWAP includes commissions, which hide the market impact
    Show the answer

    A. Its own buying pushes VWAP up, and VWAP ignores how far the price moved from the decision price A big order is part of the volume that defines VWAP, so it partly sets its own benchmark, and on a trending day trading slowly can match VWAP while the cost against the arrival price grows. Implementation shortfall measures from the decision price instead.

Sources and further reading

  1. SEC staff — Staff Report on Algorithmic Trading in U.S. Capital Markets (August 5, 2020), section on brokers’ execution algorithms (VWAP, TWAP, implementation shortfall, percent of volume)
  2. CFTC and SEC staffs — Findings Regarding the Market Events of May 6, 2010 (September 30, 2010): the 9%-of-volume sell algorithm
  3. FINRA Rule 5310 — Best Execution and Interpositioning, Supplementary Material .09 (regular and rigorous review)
  4. Finabase worked example — invented volume curve, prices and impact model computed in the Finabase figure script (scripts/learn/figures/professional-investing-quant-b.ts)