A friend who runs a small book asked me why his exits in a mid-cap alt kept getting worse with every clip. He was selling a seven figure position in equal slices, one every five minutes, and by the fourth slice the bids were stepping back before his order even arrived. He was running a TWAP, which is the most honest and most predictable execution algorithm there is, and the market had read him by slice three. Nothing exotic happened, just equal size and equal spacing on a book thin enough that anyone watching the tape could see the rhythm and get out of the way.
Every execution algorithm is an answer to the same trade-off. Trade fast and you pay market impact, because you eat through the near side of the book and push price against yourself. Trade slow and you take timing risk, because the market can move away while you wait. TWAP, VWAP, and POV are the three workhorse schedules for splitting a large order, and each one picks a different point on that curve. Each one also leaks money in a specific, predictable way, and knowing where the leak is matters more than knowing what the acronym stands for.
TWAP is honest and everyone can read it
TWAP is time weighted average price. Divide the order into equal pieces, send one every fixed interval until done. If you want to buy over four hours you send a clip every few minutes, same size, same spacing. What it optimizes for is time diversification and certainty. You know exactly when you finish, and your average price lands close to the simple average of prices over the window.
The leak is predictability. A fixed clip at a fixed interval is a signature, and signatures get traded against. On a liquid book with lots of competing flow you can hide in the noise, but on a thin book the pattern shows up fast. Once someone spots it, they know a large order is working, they know the direction, and they can make a decent guess at how much is left. Market makers widen against you, faster traders step in front of your next clip, and every fill comes back a little worse than the last. Crypto amplifies this, because on-chain TWAPs are visible to anyone who bothers to look, and exchange-native TWAPs on thin pairs are easy to spot from the trade tape alone.
The second leak is that TWAP is deliberately blind to liquidity. It sends the same clip into the dead hours as into the busy ones. In equities that means paying up through the midday lull. In crypto, which trades around the clock, it means your early-morning clips execute into a far thinner book than your afternoon ones. Any TWAP worth using randomizes clip size and timing, works limit orders inside the spread instead of crossing every time, and stretches out when the book is thin. The pure textbook version is closer to a donation.
VWAP tracks a benchmark you might be moving
VWAP slices the order in proportion to expected volume instead of time. Markets do not trade evenly. Equity volume is U-shaped, heavy at the open and close and quiet through the middle. Crypto has its own rhythm around the US and Asian sessions, flatter but real. A VWAP algorithm takes a historical volume curve and schedules your clips to follow it, so you trade more when everyone else does and hide in the crowd. What it optimizes for is tracking the day's volume weighted average price, which is the benchmark most institutional execution gets graded against.
It leaks in two places. The first is that the volume curve is a forecast built from history, and on the days that matter most, news days, listing days, liquidation cascades, the realized curve looks nothing like the historical one, so your schedule is wrong precisely when execution is hardest. The second is subtler and worth sitting with. If your order is a meaningful share of the day's volume, you drag the benchmark along with you. You can beat VWAP by a few basis points and still have gotten a terrible absolute price, because your own buying pushed the average up and the benchmark politely followed. VWAP flatters large orders in exactly the situations where they did the most damage.
VWAP makes sense when your order is a modest fraction of expected volume, when you are actually graded against the benchmark, and when the day looks normal. It makes much less sense in a market where volume profiles are unstable, which describes most smaller crypto pairs most of the time.
POV adapts, and adapting means chasing
POV, percentage of volume, drops the fixed schedule and pegs itself to the market. You pick a participation rate, say ten percent, and the algorithm trades so your fills stay near that share of whatever volume actually prints. Quiet tape, you go quiet. Volume shows up, you speed up. What it optimizes for is adaptiveness. It never forces size into a dead book, and it takes liquidity when liquidity appears, which is roughly what a careful human would do by hand.
The leak is baked into the correlation between volume and price movement. Volume spikes when price moves, so a POV buyer accelerates exactly when price is running higher and slows down when price drifts back toward levels they would love to own. You systematically do more size at worse prices whenever the market trends against you, which is the polite way of saying POV chases. In a strong trend a POV order can finish with an average fill meaningfully worse than a dumb TWAP over the same window, because the TWAP at least kept buying through the quiet dips.
POV also gives up the completion guarantee. If volume never shows, the order is still sitting there when the session ends, and now you have an unfinished trade plus whatever risk comes with carrying it. And at higher participation rates the adaptiveness turns into a mirror. Above roughly twenty percent you are a large share of the tape, the volume the algorithm reacts to is substantially your own, and the loop can feed on itself. Most desks cap participation somewhere in the five to fifteen percent range and treat anything above that as an impact trade that needs a different plan.
Picking one comes down to size, urgency, and liquidity
I ask three questions before any large order. How big is it relative to what the asset typically trades in a day? How urgent is it, meaning what does it actually cost me if this takes all day or does not finish? And how lumpy is the liquidity, meaning does volume arrive on a reliable curve or in unpredictable bursts? The answers map to a rough guide.
- Under roughly one percent of typical daily volume, skip the algorithm entirely. Work a limit order near the touch and be patient. The impact you are trying to manage barely exists at that size.
- A few percent of daily volume with no urgency: VWAP over the full session if the asset has a stable volume curve, TWAP with randomized clip sizes and jittered timing if it does not. Most mid-cap crypto belongs in the second bucket.
- Five to fifteen percent of daily volume: POV at a modest participation rate, with a hard end time and a plan for the remainder, either finish aggressively at the close or carry it and accept the risk.
- Genuinely urgent, meaning you believe price is about to move: front-load. Take the impact early instead of pretending a schedule protects you. Schedules optimize average price, and urgency means average price is no longer the objective.
- Thin or around-the-clock assets: stretch the horizon, randomize heavily, and skip the known dead hours entirely rather than trusting any schedule to handle them gracefully.
All three leak somewhere. TWAP leaks through predictability, VWAP leaks through a stale forecast and a benchmark your own flow can push around, and POV leaks through chasing. Picking an algorithm is really picking which leak you can afford on this order, in this asset, on this day.
The habit that improves this over time costs nothing. Record the mid price at the moment you decide to trade, then compare it against your final average fill, on every large order, in a spreadsheet if that is all you have. That gap is your implementation shortfall, and after twenty or thirty orders the pattern gets embarrassing and specific. Mine was POV orders in trending markets, which is why I mostly stopped using them for anything I genuinely wanted to own.