Strategy in software offers flexibility, strategy in hardware offers performance. Hardware acceleration through FPGAs offers both.
In automated trading, latency is crucial. The sooner you are on the market, the better chance you have of executing the desired order. Predictably, this provokes a response from hardware and software designers. What we are witnessing today in the realm of automated trading, is a shockwave of innovations when it comes to trading platforms. A day trader can choose from a wide range of platforms available, but not every platform can offer an ideal functionality. Software-based trading platforms cannot match hardware-based platforms’ performance and latency. Hardware-based trading platforms are not as flexible as software-based platforms and are much more difficult to operate.
In an effort to combine the performance of hardware and flexibility and user-friendliness of software, FPGA-based trading platforms have emerged. FPGAs allow your logic to be executed in hardware, which allows for much higher performance and lower latency. In a race to create the perfect FPGA-based trading platform, developers now have pretty much a clear idea of what this platform should look like.
Since offloading the tick-to-trade processing to hardware is a great way to reduce latency, the issue here is how to run as much of the latency sensitive processing in FPGA and only the rest in CPU. Some manufacturers (i.e. Netcope Technologies) already offer a platform that makes it possible to run the whole tick-to-trade processing, including the strategy, in FPGA. This results in deterministic latency below 1us, depending on the complexity of the trading algorithm.
Running the whole strategy in FPGA, however, carries with it a substantial problem for the person responsible for the strategy implementation. Should the FPGA manufacturers implement the strategy themselves, the trader has to share his trading secrets with them. Needless to say that no trader wants to reveal decades of experience embedded in a trading strategy to a third party. And once the strategy is embedded in the platform, any update or change would require assistance of a skilled hardware designer. What is required is a platform that would take full advantage of FPGA’s programmability. Even better if the platform allows you to write your logic as easy as software, for instance in C++. Only one commercially available solution provides this functionality - Tradecope.
In summary, when buying an FPGA-based trading platform, a day trader should look for a platform that enables running the whole strategy in hardware and simplifies the task of programming the FPGA. Running the whole strategy in hardware will help them to achieve very low, deterministic latency and programmability will enable easy implementation and independent operation of the trading strategy.