Achieving Best Execution is no longer just a compliance objective, it is critical to trade execution businesses in maintaining their competitive edge. Achieving the best price is a complex undertaking as market liquidity is fragmented across multiple venues, and any predictable order flow can lead to front running by ultra-low latency players driving up prices to affect traders’ competitiveness and ultimately profitability.
Furthermore, the Markets in Financial Instruments Directive (MiFID-II's) post-trade transparency rules raised the standards for Best Execution to even higher levels, so developing innovative algorithms for strategic order execution has never been more important.
Smart Order Routers (SORs) that use Artificial Intelligence (AI) can address order routing and liquidity discovery challenges to achieve best performance and ultimately increase profitability. However, integrating AI into order router systems brings complex technological challenges in four key areas, and significant R&D is needed to achieve this goal. While this represents a significant investment, in the UK R&D tax credits for such projects are worth between 10% and 33% of related expenditure: qualifying expenditure includes staff, contractors and software licences costs attributable to the R&D.
Liquidity discovery for order routing
The ability of a trading system to discover liquidity at trading venues is crucial to generating optimum real-time order flows. The order queue of the exchange, the order flow statistics, the liquidity at venues, and even the client's urgency determine the optimal response. It is also vital to split orders into child orders to prevent information leakage. Moreover, orders need to be routed to the best trading venue, either via a market order for taking liquidity or via a limit order for creating liquidity. Managing such complexity requires the development of advanced algorithms.
First-generation SORs had static, rules-based designs that determined optimal routing based on a periodic snapshot of market conditions – making them slow to react to the market. To enable a dynamic response, quantitative analysts created SORs which split a large order into child orders and sequenced them across venues in multiple sweeps of passive or aggressive rounds.
Despite these advancements, any predictable behaviour makes SOR trades vulnerable to information leaks. For example, if "sniffing algorithms" detect an imminent buy order flow, they will bid up the price, increasing the trade’s implicit cost. This is why we believe that the development of innovative algorithms with tactical intelligence will give SORs an edge in this zero-sum game.
Detecting information leakage
AI-driven SORs can react almost instantly to information leakage by detecting changes in implicit costs of realised trades. On detecting information leakage at a venue, it can pull trades from it and reconfigure order flows. Furthermore, AI-powered SORs using Deep Learning can learn from trades of correlated securities and predict implicit costs before placing orders. We believe the next wave of innovations in SORs will harness Reinforcement Learning to make SORs that are truly intelligent and autonomous.
The latency challenge
Not all decisions made by AI systems can be predicted because of their black-box design. While this is a competitive advantage, any latency they introduce into trading platforms impacts the ability to react quickly to adverse order fills or price movements. Reaction times of just a few microseconds can damage the competitive edge of platforms which rely on ultra-low ‘tick to trade’ intervals. A significant amount of R&D investment will be directed towards AI-based SORs that do not adversely affect the latency of trading systems.
The future of SORs
Field-programmable gate array (FPGA) systems can improve latency by integrating AI at a hardware level. But they lack flexibility and are not easy to reprogram. A low latency version of AI such as Real-Time AI, used in online ad-bidding systems is flexible and fast. In combination with FPGA systems, Real-Time AI can deliver ultra-low latency solutions with strong AI capabilities. Using such hybrid solutions may very well become the gold standard for achieving Best Execution.
While ongoing R&D projects are key to maintaining a strong business model for any financial services business, they inevitably carry cost and risks. Fortunately, the vast majority of the activities undertaken on AI and SOR projects will qualify for R&D tax credits – limiting the financial risk. For help and advice on managing your R&D relief claims please contact Carrie Rutland or Nicoletta Papademetris.