Such a system has not been available in the past due to issues with speed of data capture; availability of realistic matching engine technology; and the complexity of sourcing and putting together all the required components.
Regulators globally have become increasingly concerned with algorithmic and HFT strategies since the ‘flash crash’ in 2008. However, firms are understandably reluctant to hand over their costly intellectual property, and regulators are not necessarily equipped with the resources to understand how these complex algorithms work. There are also acknowledged deficiencies in the simulators currently in use.
“A key issue in simulation is that it assumes zero market impact, wherein historic replay might trigger a strategy but then fails to take into account the impact of any HFT strategy order that is generated.” said Rob Hodgkinson. “That’s not realistic, as obviously a live market is the result of all the participants interacting with each other.”
The FD system has four key features that make it different from other approaches:
• A sophisticated simulation that takes market impact into account, generating each new order around the current bid/ask spread;
• A very fast ‘production’ matching engine that can simulate multiple markets at once;
• Injection of market events such as sector movements or a simulated ‘flash crash’ scenario to see how the algorithm will react;
• Realtime market data capture of all orders and trades to create a genetic footprint – the ‘Strategy DNA’ – of each algorithm’s behaviour.
Hodgkinson, who has been developing exchange matching systems since 1992 and has automated more than a dozen exchanges around the world, said a key new development in the FD system is a new kind of matching engine.
“A great problem that has prevented the production of this kind of sophisticated testing lab in the past is the cost of a matching engine, which is in the millions,” he said. “To test algorithms properly, you need to recreate a real environment – which may mean buying several matching engines to simulate several markets. This cost is prohibitive for even the world’s biggest firms and unthinkable for smaller one. You really need a ‘production’ matching engine that can do the job of a real matching engine but isn’t prohibitively expensive, and that’s what we have created.”
The Delta AlgoLab overcomes a number of hurdles that have been preventing such a system from being developed in the past.
“The total cost – even excluding the matching engine – of putting together such a system has been prohibitive to date,” Hodgkinson said. “It’s a complex job to integrate all the components as they haven’t all been available from a single source. Also, firms have traditionally seen market data capture as too slow to keep pace with matching engines.”
The Delta AlgoLab captures each algorithm’s behaviour across the simulation and stores it in a sophisticated database that can analyse the algorithm’s characteristics. It then provides a detailed report of what FD is calling the ‘Strategy DNA’ – a set of metrics providing a clear behavioural description of the algorithm.
“The beauty of this kind of reporting is that it can be provided to third parties such as regulators, in layman’s terms using agreed metrics,” Hodgkinson said. “This recognises the fact that while such strategies are usually developed by high-level mathematicians, they need to be understood by a broader audience concerned with market quality. At the same time, it protects the IP inherent in the strategy.”
FD is demonstrating the Delta AlgoLab to regulators and exchanges across Asia over the coming weeks. Today they also released a whitepaper called “High-frequency regulation – A new approach to monitoring performance on the capital markets racetrack” describing the technical components for the Delta AlgoLab.