Are you getting your fill?
Are you getting your fill?
by Andy Ross and Ralph Bird
One key aspect of electronic trading often taken for granted is the behaviour of the matching engine in the order book; specifically, the allocation process. When orders match – who gets the fill, in what order and why? It’s a crucial aspect of trading – and yet another area in which CurveGlobal is looking to inject new thinking, a clear benefit of more competitive markets.
Traders intuitively understand that different allocation methodologies favour different market behaviours. For example, FIFO (first in, first out) or price/time allocation will reward the first to post an order at a particular price. In markets showing higher than average price volatility, this would seem like the fairest way to go. But is this approach contributing to the connectivity arms race, in which participants buy their way to the front of the queue?
In markets with low price volatility, you will see orders stacking up at the best price levels, all vying for a share of any incoming matching orders. A FIFO allocation would mean large orders at the front of the book effectively blocking later orders from receiving any fill volume. It was for this reason that exchanges started to use pro-rata allocation, effectively distributing traded volume among participants based on the relative sizes of their orders.
So is pro-rata any fairer? Well, you can always rely on traders to find a way to gain an advantage; in pure pro-rata markets some started to “over-offer” so that they would gain better fills. This was highlighted in a study by Karel Janecak and Martin Kabrhel (Matching Algorithms of International Exchanges, see link below) which also pointed out how some market participants tried splitting orders hoping to take advantage of rounding to get better fills. They even suggest that pure pro-rata allocation could favour High Frequency Trading (HFT) participants because there is a minimal extra cost for them to cancel and re-enter orders.
It’s also worth pointing out that while the order depth appears large, there are restrictions on individual order size – so participants can’t “sweep to fill” and lift the entire book in one go. – this also reduces the risks associated with “over-offering”.
In response to this behaviour, exchanges tried a number of measures hoping to get the best of both approaches, mixing FIFO and pro-rata, effectively dividing incoming orders between the two algorithms. They also tried giving priority to orders that had improved the market, filling these first before sharing out any remaining volume.
STIR-ing it up
Short Term Interest Rate Futures (STIRs) markets use pro-rata allocation. STIRs exhibit low volatility compared to their respective exchange-configured minimum price increments and bid and ask prices commonly hold steady for substantial periods of time. As a result, some variant of pro-rata matching is desirable since it can lead to execution without necessitating traders to cross the spread to get ahead in the queue.
In 2007, LIFFE, introduced a time element to the allocation algorithm used in the STIRs markets to add stability and reliability to the order book. They did it to counteract what they saw as the practice of people putting “larger and larger orders in to include their fill even if they don’t need that amount.” The idea was to mathematically add weighting for each order's size dependent on its position in the queue -- before allocating fills in a pro-rata fashion using the new weighted values. This would ensure that orders at the front of the queue would getter better fills than later orders and reduce the potential to gain an advantage by “over offering”. LIFFE pointed out that the new algorithm would add stability and reliability to the order book because “there will be less volume switching in and out quickly”.
Is this the best of both worlds?
When CME introduced their latest STIR Future, the Quarterly IMM SONIA Future, they opted instead for pro-rata allocations with top order priority. This algorithm fills the order that had been awarded priority first and then allocates any remaining volume among the other orders at the matching price according their relative sizes in a pure pro-rata fashion. ICE launched their three-month SONIA Future contract with their gradual time-based pro rata tuned to give the time element the smallest possible weighting without implementing pure pro rata.
The ICE gradual time-based pro-rata algorithm gives weighting to orders depending on their position in the queue using a configured time exponent. The higher the value of the time exponent, the more the allocation favours orders at the front of the queue, regardless of size. They set the time element to 2 (pure pro-rata has a setting of 1) and from 3 December they tuned the three-month Sterling Future matching algorithm to have the same small-time component. Both these allocation algorithms give priority to the first order at the best price but the top order priority in both the CME and ICE algorithms can be taken away by any order that betters the price, no matter how briefly.
Only CurveGlobal (for its three-month SONIA Future offering) has implemented a progressive time weighted allocation algorithm with a significant time exponent together with a top priority that, once awarded to an order, stays with the order until it trades or gets cancelled.
Implied orders
Matching algorithms also differ in the way that they handle implied orders. The first thing the ICE matching algorithm does is divide allocation between aggregate outright volume and aggregate implied volume dependant on their relative sizes, but this disregards their relative time priority. Moreover, they match implied volume based on the “longest chain” or greatest number of potential matches. So implied volume from a butterfly order is going to have priority over implied volume from a calendar spread order, regardless of when the respective butterfly and spread orders were entered.
You could argue that the longest chain enhances liquidity. It certainly appears that the exchange generates more trades as a result and possibly therefore more revenue. But if you happen to be the person sat at the front of the queue on an outright hedge the longest chain methodology may not always be seen as the fairest. CurveGlobal’s matching algorithm gives priority to outright orders and when allocating volume to implied orders does it on a first-come first-served basis.
CurveGlobal’s mission is to be a meaningful competitor in rates income futures markets. This includes promoting innovation around matching rules rather than copying what’s out there. In fact we’d argue that competition should drive innovation around matching rules. So what do we do differently?
· We don’t do a "longest chain". Those joining the market in outrights should not be disadvantaged to those with complex trades.
· We have a persistent best price setter flag. That means the person that sets the best price is front of the queue at that price (so you don’t lose it in volatile markets).
· CurveGlobal offer a time based pro-rata algorithm – tuned to give better volume to those orders that have been waiting longer, i.e. at the front of the queue – not just biggest size.
· We look to fill outright volume before implied volume and we provide price time priority for implied orders where relevant.
As we’ve outlined, given that ICE have changed their allocation parameters, this could impact the market and how you trade. There is now a significant difference in the 3 month Sterling Futures markets. If you’re trading and want to add large size at the back of the queue, the ICE change is clearly a benefit for you. If you’re a passive trader who wants to sit on the offer at your desired size, your chance of getting filled may just have decreased even further.
Give CurveGlobal a try – you might be pleasantly surprised.
REFERENCES / BACKGROUND:
Paper analysing trade matching algorithms, with particular focus on Time Pro-Rata algorithm introduced by Euronext.LIFFE in 2007:
Janecek, K & Kabrhel, M (2007) Matching Algorithms of International Exchanges
Available at: https://pdfs.semanticscholar.org/6d92/0528fc5a3a25cb7a627b93ae3e7d5789bde8.pdf
Risk.net article where LIFFE announced the time pro-rata allocation methodology:
ICE Notice detailing the new algorithm configuration for 3month Sterling Futures:
Global Head of Prime Brokerage
6 年Ralph Bird?