The accelerated automatization of entire industries leaves no technological and procedural stone unturned, and trading is no different. The sheer number of fintech companies addressing various needs is the evidence. This is where the traditional emphasis on highly regulated middlemen, controlled risks, and tried-and-true legacy processes clashes, having to deal with growing amounts of data and the necessity to scale, embrace new products, and conquer new markets while reducing costs. 

These rapid changes present a unique set of challenges to automated trading systems designers. Sure, the environment of constantly shifting requirements is very familiar to us, but how do we deal with this “business as usual” and still build robust, adaptive, and cost-effective systems? And how can we make our offering stand out among a vast variety of competing products?

Mind the edge conditions

Data-driven software often relies on models that best describe “average” scenarios, additionally constrained by various simplifying assumptions. Real-world cases however rarely fall into some generic category, and electronic trading with its asynchronous and unstable communications, uncooperating market participants, and high levels of data noise is all about edge cases. 

Technology can add a lot of value here by specifically dealing with unusual, inconvenient, and marginal scenarios where models fail. This is where the system’s unique characteristics and hidden virtues can often lie. And this is something that would also be most noticeable to users: observing the system perform well on a difficult case probably impresses more than a routine demonstration of a middle-of-the-pack volume-weighted average price (VWAP). Good performance on one-off cases, low-liquidity instruments, very small or very large orders, and extended hours trading can be a great differentiating factor. 

Automation and customization

Similarly, technologists benefit from understanding each client’s preferences and distinct trading patterns. Electronic trading is not just about best execution and quantitative performance but also about automating things that used to be manual. Analyzing existing processes for ways to formalize and automate them creates an opportunity to add value through highly customized products. How many different VWAP strategies do we need? But offering clients a safe and quick way to automate what they already do can be a very attractive proposition. Engineers should have that in mind and design easily customizable systems with a reasonably short release cycle from the get-go.

“Engineers should be open to new and freely available technologies, readily engage various stakeholders including customers, and be willing to experiment.”

Experiment— build, try, discard, build again

Electronic trading is a world of uncertainty characterized by shifting requirements and fluctuating market conditions. A promising and exciting way of dealing with this is to embrace exploration. A trading system can be designed to introduce experimental code and run A/B-like tests on the split order flow in a transparent manner communicated to users. The split can be based on a random factor, specific instrument properties, or both. For example, run the new code only on every tenth order, or only on an instrument from a class where the performance is worse than average. Combinations are endless. Such “fun” interactive exercises can bring valuable insights into the entire ecosystem behavior and at the same time potentially discover clients’ true needs.

Be lightweight and agile

Running experiments in production requires a lot of native flexibility. That leads us to a lightweight scalable design where, ideally, entire chunks of the system can be discarded and replaced by new ones quickly built from scratch. If the experimental logic demonstrates good performance, it can later be “tidied up” and integrated properly. And why not build out a bespoke trading stack on a single piece of hardware dedicated to a client? Easy to deploy, redeploy, modify quickly in any possible way and tear down when no longer needed. If we can do that, we can even do it on a cloud, if that makes sense for the overall business model and economics. 

Our Cantor Fitzgerald Precision Algos trading platform is designed on these principles which makes it well suited for white labeling since we can quickly replicate an entire algo stack for a client on independent hardware and then replace whatever is needed with highly customized logic without fear of affecting others. 

Embrace open source, and not just because of cost

Open standards and open-source software bring advantages that go well beyond cost savings and immediate convenience. 

Firstly, open and free technology gets improved and enhanced continuously by countless contributors. Not all the changes are safe and sound, but the huge success of the open-source movement speaks for itself. It is now a fundamental part of the enterprise technology world, from operating systems to advanced machine learning libraries. Standardizing on widely shared tools and upgrading periodically means getting new features, performance improvements, and fixes constantly and essentially “for free”.

Secondly, since open technology is available to all, the community of engineers of all levels with some exposure to it is much wider. That, in turn, makes hiring someone with relevant experience a more manageable task. Compare that with trying to find someone sufficiently familiar with a closed, niche, and potentially outdated product maintained and supported by a single company that may go out of business in the long run. This is in addition to the cost of the technology itself, a factor entirely outside of the user's control. Proprietary systems, in other words, while possibly of high quality and with a superior support model, pose a separate category of risk.

Automated trading systems function in a world of uncertainties and rapid changes. This is where risks but also opportunities lie. Engineers should be open to new and freely available technologies, readily engage various stakeholders including customers, and be willing to experiment. It is the approach that we in the Cantor Precision Algos team fully embrace.