TWAP order: the algorithm slices volume by time, not by liquidity

A TWAP order (time-weighted average price) is an instruction to the trading system: take a large order, break it into a stream of small child orders and release them in equal portions at equal intervals until the end of a set window. The aim is to get an average execution price close to the security's time-weighted average price over that window, in other words, to keep your own volume from moving the market against you. An important point straight away: the exchange does not know this order type. What arrives in the order book are ordinary limit orders or market orders, while the schedule of their appearance is kept by an algorithm on the side of the broker or the trading terminal.

What happens inside the window

The trader defines the parent order: the instrument, the direction, the total volume and the boundaries of the window, from the moment it starts to the moment it ends. The algorithm divides the volume into equal shares by the number of intervals and sends its portion in each interval. The logic is deliberately dumb: the schedule depends neither on how the price is moving nor on how much volume is sitting in the order book right now. Time is the only reference point.

This gives TWAP its main property: it does not need a forecast. To distribute the volume, the algorithm does not have to guess when liquidity will arrive today. That makes it robust where the trading profile is unstable and past days are a poor description of the current one.

How TWAP differs from VWAP and the iceberg

VWAP distributes volume in proportion to the expected liquidity profile: it buys more at the times when, historically, more is traded. As long as the day resembles a typical one, VWAP cuts costs more precisely. As soon as the day is atypical (a piece of news, an auction, a sharp spike), the profile lies, and VWAP steers execution to where there is in fact no volume.

TWAP is protected from this error, but it pays for the protection in a different coin: in the hours when the order book is thin, it is still obliged to place its portion, and that portion costs more.

It should not be confused with the iceberg, although the task looks similar. An iceberg order hides the size of a single order already resting in the book; this is a property of the order itself at the exchange level. TWAP hides nothing: each of its child orders is of ordinary size and fully visible, and the only thing concealed is that another will follow it. The difference lies in what is being masked: the iceberg hides volume, TWAP hides urgency. In practice they are combined, since TWAP child orders can be placed as icebergs.

Where TWAP works and where it gets in the way

The sensible scenario is building or unwinding a position that is large relative to the security's usual turnover. If your volume is comparable to what passes through the order book in a trading day, a single market order sent all at once will eat through several price levels, and you will be the one watching the pullback afterwards. Look at the current liquidity of the security in the Russian stock profiles: comparing your volume with average turnover is precisely the criterion for whether an algorithm is needed at all. For {{instrument:SBER}} the answer is almost always "not needed"; for securities with infrequent trades it is almost always "needed".

In bonds the picture is different. Liquidity there arrives in clumps: trades are rare, but large. In such an environment an even schedule simply stands idle and then fills the remainder at a worse price. In OFZ and other issues from the debt market list, the negotiated trading mode and working resting orders often give a better result than any clock-driven algorithm.

The parameters that decide everything

The length of the window. A short window brings the result close to a market order, with all of its costs. A long one dissolves the volume but keeps you in the market for longer, and during the window the price may move away of its own accord. TWAP does not manage this risk; it manages only execution costs.

The type of child order. Market orders guarantee that the portion is filled and give up the spread. Limit orders save the spread and leave a remainder unfilled; by the end of the window the algorithm either fills it aggressively or cancels it. Read up on exactly what you pay with in each case: market and limit orders.

Behaviour at the boundary. What to do with the unfilled remainder: extend the window, fill it at market, or pull it. This is the step that is most often left at its default setting, and then the result comes as a surprise.

What TWAP does not do

It does not improve the price and does not protect against market moves. It solves one narrow task: keeping your own volume from becoming the cause of a bad price. It is not a substitute for a stop order as a way of limiting losses, nor a way to do better than average, because by construction TWAP aims for the average.

There is also a calendar point: the algorithm's window should not be laid over earnings releases and corporate events or over the nearest record dates — {{dividend_calendar|limit=5}}. On those days the time-weighted average price stops being a meaningful benchmark, because the very thing being valued changes inside the window.

A caveat on completeness: the platform has no data on which brokers on the Russian market offer TWAP to retail clients, or with which randomisation settings. This has to be checked with your own broker directly.