The Gateway to Algorithmic and Automated Trading

Corvil: Rethinking Speed in Financial Markets, Part 4 - The Need for Machine-Time Data

First Published 8th September 2016

A big part of the human concern over high-speed trading is born from the fact that we can't easily or precisely see what the machines are doing - it's all a blur. But the root problem is not speed. It is time. This is the final instalment in a four-part blog series by Donal Byrne, CEO of Corvil which explores the new role of speed in modern market structurel.

Machine-time data is essential to operate electronically traded markets transparently and fairly.

While speed improves price certainty, it also makes it harder to see what is going on. The faster an object travels, the more difficult it is for an observer to make out the details of the object. I believe a big part of the human concern over high-speed trading is born from the fact that we can't easily or precisely see what the machines are doing. It's all a blur.

I believe the root problem is not speed. It is time.

Albert Einstein gives us a hint at the solution to our problem. He explains: "Time is what prevents everything from happening at once." Time is the essential mechanism that allows us to live our life in an orderly fashion. Time makes sense of the progression of existence and events that occur in irreversible succession from the past through the present to the future. Without time, we could not make sense of events that happen. We could not tell the past from the present from the future.

To make sense of time we must observe it with appropriate granularity and accuracy. Let's say you are at your local grocer and you are buying milk. You look at the "best before" date and see that it only has the year specified - that is, no day and no month. You look at another milk producer's product and discover the same thing except it is quoting a different year. Then you think, "What date is it anyway?" You look at your watch, and all it has is a single number - the year. You know, however, that milk will go sour within a week. The problem therefore is that if you buy the milk, you have no valid time data to inform your decision to purchase. You have no idea if you are buying fresh milk or sour milk.

The exact same situation arises in electronic trading when our ability to observe events is limited to a time granularity and accuracy that is much larger than the time it takes a machine to trade.

The SEC recently claimed a millisecond to be de minimis for events in US equities markets. In the machine world, a decision to trade can be made in 10 microseconds or less. We are now seeing sub-one microsecond algo decision times in FPGA implementations. Therefore, a millisecond time granularity in the machine world would be equivalent to us living our daily lives where time could only be measured with a granularity and accuracy of a single day. Everything we do in a day would be considered to happen at the same time. Needless to say, this would make life very confusing and problematic.

The problem is that we cannot tell time in today's markets.

We must be able to tell time in a machine world so we can observe and control accurately the actions of machines that we entrust to trade on our behalf. People often use the term "real time" to infer that we can see things as they actually happen. In a human world we typically equate real time to be approximately a second. If you get a response within a second, we generally consider this real-time. I call this "human real time." This explains why most humans wear a time management device on their person with a granularity and accuracy of one second. We call this a watch.

A machine world is different. Machines act much faster than humans. Their idea of real time is much closer to a microsecond. Roughly a million times faster. I refer to this as "machine real time" or "machine time" for short. We define machine time as the time within which a machine can act or make a decision. We therefore need a machine-time watch for a machine-time world.

There are signs that this understanding is happening. MiFID II in Europe has embraced this understanding in part with its recent rules mandating that all business clocks involved in high-speed trading must be synchronized to within 100 microseconds of Coordinated Universal Time (UTC), with a timestamp granularity of a microsecond or better. The only fly in the ointment is that 100 microseconds accuracy is not sufficient to provide the levels of visibility required to detect market abuse reliably. The details are referenced in ESMA RTS-25.

MiFID II also requires investment firms to maintain a record of all machine data involved in a high-speed trade transaction. This machine data needs to be synchronized to UTC and maintained for a period of five years. The value and quality of this data is highly dependent on the accuracy and granularity of the timestamp associated with each piece of data. For today's trading machines, we need timestamps with a minimum of a microsecond granularity and ideally a microsecond accuracy relative to UTC. Commercial technologies for UTC clock synchronization (e.g., GPS with PPS/PTP signal distribution) can deliver approximately 3 to 5 microseconds accuracy. This would be good enough.

We refer to microsecond time-synchronized data as "machine-time data." Machine-time data is the new type of data essential for orderly operation of any electronic trading business and becomes the main data source for assuring trade transaction transparency, execution quality and detection of potential market abuse. If regulators, venues and participants don't address the fundamental need for accurate machine-time data, we will continue to pursue ineffective agendas and ultimately fail to build trust and confidence in the operation of our electronic financial markets.

I will leave you with this final thought, as it relates to dealing with speed in financial markets:

"Most people spend more time and energy going around problems than in trying to solve them."
-Henry Ford (1863 to 1947)

Read part 1 here

Read part 2 here

Read part 3 here