Around the turn of the millennium after Reg NMS, decimalization and computers took over trading and propelled it into the 21st century, a revolutionary trading software program was created – the algorithm. It was revolutionary in that it took a parent order and according to certain base parameters, broker it down into child orders. And then it would send these child orders to multiple trading venues to either be filled immediately or sent to others in the hopes of being filled there. Traders through computer software could now find liquidity, size and ultimately and best execution through the click of a mouse.
The first algos were simple – they functioned based on either a volume-weighed average or time -weighted average approach to finding liquidity. And these algos functioned only when trading stocks – no options, futures and forex. While limited for a time, these basic algorithms were a boon to the industry and fast became a traders’ best friend.
Algorithmic strategies have become far more complex over the past 10 years,” began David Margulies, Head of Electronic Products Group at Weeden & Co. “Strategies that used to have very simple routing (i.e. divide the order equally among a list of venues) to a much more data driven approach to routing. Strategies now use a massive amount of historical and real-time data to make decisions on what venues to send orders to, what min fill size, what limit, etc.”
But that’s not all. Margulies added that order placement logic has also become infinitely faster in order to react to order imbalances, spread size, conditional invites, etc.
“In 2007 it was commonplace to measure latency in milliseconds (a thousandth of a second) as compared to microseconds (a millionth of a second) or even nanoseconds (one thousand-millionth of a second) in use today,” Margulies said. “In 2007, we launched the first non-internalized dark pool aggregator called Onepipe. At the time of the launch, Onepipe was considered cutting edge technology in how we accessed over 35 different pools. By today’s standards, Onepipe wouldn’t have been as successful using the logic and technology that was available in 2007. Over the years, we have made literally hundreds of changes to our strategies in order to constantly adapt to client needs, technology advancements, market structure changes and anti-gaming tactics.”
But then technologists said, “wait a sec, why can’t algorithms use other approaches to trading?” And thus more complex and bespoke algos were born.
You say you want a revolution?
And traders, not technologists, began to drive algorithm development. And the algorithms became more complex and required longer development and testing times.
So, is algo development still a long process given the trader’s more complex needs?
“This is an interesting question. The answer is both,” Margulies told Traders Magazine. “Development time is shorter due to deep investments made in rapid development tools and the rewriting algorithms to make them easier to adjust and customize. Many changes can be made overnight but it took a lot of work to get to that point.”
Back in 2007, almost every change to a strategy required a programmer to write new code, regression test, QA test, deploy, etc. Now many changes can be made with a simple parameter change, Margulies explained. This takes a lot of planning before the first line of code is ever written.
“Think about building a house. If you know you may want to add a bathroom in a new house down the road, it’s easy to have the rough-in plumbing installed during construction, as opposed to ripping up walls and floors to add a bathroom later on,” Margulies said. “The same concept applies to designing and building algorithmic strategies. We need to anticipate future changes and make it easy to make these changes when the time arises.”
Algo development never stops. Every day, traders like Margulies and his counterparts around Wall Street make changes to their strategies – some very small and some completely new in terms of features or logic. Just look at the market structure changes over the past 10 years – i.e. tick pilot, conditional venues, new order types – and it makes sense. It also takes a lot of time, effort and expense just to stay current. At Weeden, the firm employs a “Best Ex Committee” where they spend a great deal of time discussing venues, routing logic, performance metrics, etc.”
Peter Maragos, CEO of Dash Financial Technologies agrees, noting that traders too are more complex – just like the algos they now use.
“Buy side traders are extremely sophisticated today,” Maragos began. “Many of them are looking to really drive the routing and posting behavior of their trading strategies, which means they need a degree of control well beyond simple aggressiveness settings. Most of our clients want to work with us to help them customize every aspect of the algo in the same way a tailor would design a bespoke suit.”
And the trading desk, not just the trader began to change. The trading floor, once only the bastion of traders and salespeople, started make room for algorithm developers who would work hand-in-hand with the trader to make his algos better, stronger, faster.
“They (traders) can only do that with a full view into how the algo performs, however, which is why full transparency into the order’s routing behavior is so critical,” Maragos said. “Ultimately, we see all buy side adopting a hands-on approach like this, which helps them drive value for their firm by maximizing alpha capture.”
To read the full article, click here.