The next generation of smart order routers

Vincent Burzynski, chief product officer for SunGard’s global trading business, recently answered Dealing With Technology’s questions on the future of smart order routers. Read on…

How would you describe the current generation of smart order routing?

First-generation smart routers, particularly in Europe, have focused on maximizing the efficiency of their trading across ‘lit’ public-book venues – the established exchanges and major multilateral trading facilities (MTFs). Identification of trading opportunities is based primarily on real-time analysis of the prices streamed from these venues, including considerations of size and market depth. Routing strategies, designed as they must be under MiFID to obtain the best possible trade prices from a broker’s connected venues, then take into account other considerations such as trading platform latency and the fee costs of trading at each venue. These strategies may include significant amounts of ‘passive’ trading, such as placing orders on the markets as a price maker in anticipation of being hit by a taker on the opposite side – the MTFs’ maker-taker pricing models particularly encourage this tactic. Optimizing the results for both broker and client in such cases is complex, and consultation is often necessary.

In concept the algorithmic approach is usually quite simple, but it has already yielded two important benefits for user firms: Prices available on the MTFs can in many cases beat those on the ‘home’ exchanges. Aggressive users of smart routing regularly report average price improvements for trades routed to the MTFs in the order of 8 – 10 basis points, and are often sending 30-50% of their trades this way as the major markets have fragmented more heavily.

The maker-taker pricing models of the MTFs, and also the pricing of their associated clearers, are so advantageous that, even with strict adherence to best execution policies, brokers can obtain very significant fee cost savings.

Another significant characteristic of current SORs is ‘non-integration’. Although the smart router can be considered as ‘just another algo’ contributing to the increasing fragmentation of institutional orders into many much smaller transactions, it is at present normally quite separate from the mainstream algo suites of brokers. Simply put, one can say that the algo suite determines when a trade should be done, and the smart router then decides where it will be done.

What do you see as the characteristics of the next generation of smart order routing? When and where do you see it being available?

The next generation is already appearing. We won’t see a disconnect: additional parameters and strategies are in most cases added to existing installations. The main issues being addressed include the following:

• SOR strategies are now moving beyond pure real-time analysis. As sufficient large-volume liquidity history becomes available for a given security across several venues, statistical analysis is being used to identify when and what kind of orders to send to a specific venue. This can help both to reduce the intensity of real-time calculation and also in developing under- and overweight strategies for specific venues at different times.
• Smart routers are increasingly routing to Dark Pools. The statistical analysis mentioned above is vital here, as in the absence of real-time prices it is the only valid basis on which to decide how much of an order to send to the pool and for how long, usually before progressing the sweep to another pool.
• At the same time, algos (while still typically separate from the SOR) are taking real account of multi-listed execution. For instance: where 30%+ of liquidity is away from the reference exchange, percentage volume algos should incorporate volumes on the MTFs.
• Client demand is also driving improved reporting from smart routers, including detailed transaction cost analysis.

We expect to see these factors become increasingly prevalent during 2010.

Are we seeing current smart order routers scale to meet the introduction of new trading venues and order types?

Scaling is an important issue for an application where real-time performance is vital, especially when more calculation intensity (statistical analysis) is being demanded at the same time as also new venues are being added – more data to watch, more complex decisions to be made. The optimization and performance tuning of router algorithms is therefore getting a lot of attention at present, and of course hardware power is also being thrown at the problem.

Arguably this problem is approaching a peak as the number of public-book trading venues for most stocks still continues to increase, whereas expectations are that we will subsequently see some degree of consolidation among the MTFs (though experience in the US perhaps indicates otherwise). The need to sweep more dark pools will also keep the pressure on.

The issues of handling heterogeneous order types, tick sizes and so on already appear to be becoming less problematic as new venues are added, because the MTFs have adopted reasonably uniform approaches on these topics.

Is there an average lag time between the introduction of venues and order types before they’re incorporated into smart order routers? Is that time period shrinking?

There is by definition a time lag between the introduction of venues and their availability for trading via a smart router. Clearly, the specifics of the new venue – order types supported, etc. – have to be handled, and testing has to be carried out. Routing algorithms may need to be adapted, particularly If the venue offers new instrument coverage or onward routing, or there are peculiarities in the commercial offer.

As mentioned earlier, the fact that MTFs tend to follow fairly standardized approaches in many areas, meaning that new introductions are becoming progressively easier to handle, so time lags are shrinking. Lags can also be reduced where specifications are available well ahead of start dates, and also where introduction of instruments is phased, as is usually the case.

Current initiatives should contribute further to the trend of market simplification: for example, the common-symbology initiative being driven by several MTFs removes the need for smart routers to manage some “trans-coding” features.

How can smart order routing co-exist with low-latency execution? How much latency are smart order routers adding a trade execution?

The low-latency objective is of course the reason why the scaling and performance issues discussed above are so important: Smart routers have to be as fast as they are smart: otherwise their decisions, based on real-time prices, will be useless. Most routers therefore aim at sub-millisecond latency, and this can be achieved where algorithms are not too real-time intensive. Clearly there is a trade-off between speed and complexity, with tuned algorithms and faster hardware deployed to maximize latency as far as possible.