The TradTech Liquidity event was very good this year. It provides a good outlook on the state of the industry as well as perfect networking opportunities. Personally, I enjoyed the panel on How Smart Are Smart Order Routers? the most.

The final question in this panel was if there is any revolutionary improvement that can happen in Smart Order Routing area in a foreseeable future. Michael Seigne from Goldman Sachs commented that for every leading SOR system in the market improvements happen on an incremental basis. To keep up with the competition, one needs to upgrade quite often and this in turn leads to constant improvements of the system.

On the other hand, some potential for a revolution exists in making systems less static and more dynamic (Quod Financial has a very good word for that - Adaptive). Discussions in the exhibition area confirm that the complexity of advanced smart order routers grows and at the same time changing markets demand from the key players to make frequent changes to multiple instances of their engines present in production.

To remain on the cutting edge, companies must have robust quality assurance procedures for their platforms. Automated regression is necessary to keep up with aggressive release schedules. The tests should cover not just functional correctness of liquidity seeking algorithms and their technical performance (latency, throughput and stability), but also their efficiency from the business point of view.

What makes a SOR or algorithmic trading system efficient? It is price, speed of execution, market impact and the trading cost delivered by it.

The test solution should consist of a large library of non-deterministic scenarios and a comprehensive set of metrics obtained each time these scenarios are executed. Prior to deploying a new build on a set of production instances, it would be possible to run the tests many times in the test environments and understand whether a new build/configuration has introduced any significant changes from the business efficiency point of view.

Obviously, simulation tests are not sufficient. They can be used only when the production system is carefully monitored and the observation results are reflected in the testing. This is especially important for dynamic smart order routing systems.