How liquidity is aggregated

One instruction, several sources 

A broker that does not manufacture its own price obtains one, and in practice obtains one from several places at once. Banks, non bank market makers and other wholesale firms each stream a two sided quote for the instruments they deal in: their own bid, their own ask, their own maximum size at each. They are quoting the same instrument, so the numbers sit close together. They are quoting from different inventory and different views of the risk they already carry, so the numbers are rarely identical, and which firm shows the best of them changes many times a second.

A firm that streams a price and stands ready to deal on it is a liquidity provider, and the set a broker has relationships with is usually described as a panel. The panel is not visible from the outside: its size, its membership and the terms on which each member quotes are commercial arrangements, and they differ from firm to firm.

The problem that creates is mechanical rather than commercial. Several quotes arrive continuously, disagree with each other, update at different rates, and one instruction has to be worked against all of them. The software that resolves it is the aggregator.

What an aggregator does with competing quotes 

An aggregator maintains one ordered list. Every bid it receives is placed in descending order, highest first, and every ask in ascending order, lowest first. Each entry carries three things rather than two: the price, the size the quoting firm will deal in at that price, and the identity of the firm that sent it. The result is a single book assembled out of several firms' quotes, and the entry at the top of each side is the best price available anywhere on the panel.

The two sides of that top row do not have to come from the same firm, and most of the time they do not. The best bid can arrive from one provider while the best ask arrives from another, so the pair is a composite no single firm quoted on both sides. That is the ordinary case rather than an exotic one, and it is why an aggregated quote can be tighter than the quote of every provider feeding it.

The composite pair sits inside every quote it was assembled from. Nothing was discounted and nothing negotiated. It is the best of several numbers on each side, taken independently of the other side, and the sides are independent precisely because they are two different firms' willingness to deal.

The top of the book, and the rest of it 

The entry at the top of each side is called the top of book, and it is one entry: one price, and the amount that can be dealt at it. It is the number on a quote panel, in a watchlist and as the last price on a chart, and what almost every conversation about a price is about. It describes a transaction small enough to fit inside it, and says nothing whatever about a larger one.

Underneath it sits the rest of the sorted list: the next price on that side and the size available there, then the next, for as far as the panel is quoting. That structure is the depth of the market, and it determines whether a quantity can be completed near the top of the book or has to reach into worse prices to finish. Two instruments can show an identical top of book while one has substantial size resting behind it and the other almost none.

Aggregation acts on both, and differently. It improves the top of book by selecting the best price on each side from a wider set of quotes. It improves depth by summing the sizes of every provider quoting at or near a level, so the amount available at a price becomes the total across the panel rather than one firm's appetite. The first effect is visible on any quote panel. The second is invisible unless the platform displays the book, and what the platforms YAL runs, MetaTrader 5, show of that structure is the subject of the next lesson.

Why aggregation can improve a price 

Three of the mechanisms at work are arithmetic, and they are worth holding apart because they also fail separately.

• Selection. The best bid among several is at least as good as any one of them, and the same holds on the ask side. The composite can therefore only equal or better the best single quote, for as long as every entry in the list is one the aggregator can actually deal on.
• Independence of the two sides. Because the bid and the ask are chosen separately, the distance between them compresses whenever different firms are best on different sides. A panel whose members lean in different directions produces a tighter composite than one whose members all lean the same way, without any member changing its behaviour.
• Accumulated size. Depth at a level is the sum of what every provider is showing there, so the quantity that can be completed without reaching a worse price grows with the panel rather than being capped by one firm's appetite.

A fourth effect is described rather than measured: a provider quoting into a competitive aggregate has a reason to quote tightly, because a quote that is never the best is a quote that never trades.

Why it can fail to 

Aggregation selects. It does not create. Everything it can offer has to exist already as somebody's live quote, and there are several ways for that to stop being true at the moment it would matter most.

• Correlated withdrawal. Providers are not independent of one another in the way the selection arithmetic quietly assumes. They price off the same information and manage the same kind of risk, so a release or a shock makes them widen or step back together, and a panel of many firms behaves like one firm for the seconds in which the difference would have counted.
• Rejection. A streamed quote is an indication that a firm is willing to deal, not a commitment that it will. Some providers operate a last look, a short window in which an order matched against their quote can be declined after the match. A declined order comes back unfilled and has to be worked again against a book that has moved on.
• Staleness. The composite is assembled from quotes that arrived at different times over different connections, so in a fast market the top of the book can be a picture of what several firms were willing to do a moment ago rather than what any of them will do now.
• Duplication. Where two providers hedge with each other, or source from the same place, their quotes can represent the same underlying risk twice, and displayed depth is then larger than the size that would genuinely be there if all of it were dealt on at once.

When an order is larger than the size at the best price 

The top of the book holds for a stated quantity, and the excess of an instruction larger than it is not refused. The order is filled in parts: as much as is available at the best price, then the remainder against the next price on that side, and so on until the whole amount is done or the list runs out of resting interest. Each part can come from a different firm, so one fill can be assembled from several counterparties, none of which quoted for the whole size.

What is recorded is one average price, weighted by how much was completed at each level. Because every level after the first is worse than the first, that weighted average is worse than the top of the book by construction. Nothing moved, no provider widened, and the quote that was displayed was accurate for the size it was quoted in.

The two halves of that block are the same calculation with the direction reversed, and the distance comes out identical because the assumed depth was identical on both sides. Real depth is not symmetrical, so a real book produces different distances on its two sides at the same size, and neither direction is favoured by the mechanism itself. This is also where a wider panel makes its most concrete difference: the same instruction against a single provider's book reaches further down that one firm's ladder to find the same quantity. Aggregation does not remove the effect, since a large enough order exhausts any book. It changes how far down the list the order travels before it is complete.

What sits behind the combination step 

Two things about how a broker runs its aggregation change what the composite book looks like. The first is who is on the panel, since selection can only choose among firms that are quoting. YAL describes the pool its aggregation draws on as deep institutional liquidity. The second is distance in the physical sense: quotes travel over connections, and a book assembled from quotes that took longer to arrive describes an earlier moment. YAL's aggregation runs at Equinix LD4 in London, with under 1 ms to its liquidity sources.

Where practitioners disagree 

Whether a larger panel produces a better aggregate is genuinely contested. One position holds that every additional provider adds selection on both sides and size at every level, so the composite can only improve as the panel grows. The other holds that past a certain number the additions quote the same underlying risk as existing members, so displayed depth inflates without real size behind it, while spreading each order across more firms raises the share of it that gets declined. Both descriptions are accurate about different panels, which is why the argument has never resolved into a number.

The second is about last look itself. Firms that operate it argue that a quote streamed to many recipients over connections with real latency has to be revocable, or the fastest recipient deals against a stale price at the provider's expense every time the market moves. Critics answer that a revocable quote is an option granted for nothing, held by the firm that also decides when to exercise it, whose cost lands on the orders arriving at the least convenient moments. Regulators in several jurisdictions have addressed disclosure of the practice rather than prohibiting it.