How Financial Services can leverage AI for direct access trading and digital innovation

Patrick Lastennet, Director | Business Development - Enterprise

14 July 2020

The underlying market principles that drive the UK’s thriving financial services community, remain largely unchanged from previous years. Largely, but not entirely.

For the most part, when we talk about trading, for instance, we’re still looking at the same type of asset classes such as equities, fixed income and foreign exchange. The environment in which these firms operate is driven by markets that are now highly automated and rely heavily on algorithms to determine the best assets to trade and when to trade to yield the best returns. In order to fuel these algorithms, huge amounts of data need to be gathered from various sources, analysed quickly, and reacted upon.

In recent years, latency was a critical factor for financial services organisations. Anything they could do to minimise latency between a decision being made and a trade actioned was something they could use to win customers, increase yields, and outflank competitors.

Latency is still a critical factor, but others are emerging. CIOs in financial services businesses are taking notice and pushing ahead with digital transformation initiatives. As the race for latency has equalised and consolidated, markets continue to be dominated by a few big players, who are very efficient. Businesses occupying more niche aspects of the market need to adapt and develop strategies which consider different things other than speed.


Beyond Latency: New Frontiers

One new frontier being explored is increasingly intelligent, machine learning-enhanced business models. And one of the key parts of the market to benefit is derivatives pricing.

Derivatives are financial contracts used for a variety of purposes, whose prices are derived from some underlying asset or security. Futures contracts and options are two very common types of derivative, though more exotic varieties can be based on factors such as weather or carbon emissions.


Monte Carlo Simulation

In order to settle on a price for these contracts, large amounts of data are taken in to consideration, and to do this requires lots of computing power. Often those responsible for pricing derivatives will run several different calculations in parallel, in something called a Monte Carlo simulation to arrive at a competitive price. Running these complex computations requires a highly efficient data centre capable of providing the high density of GPUs needed to crunch all the numbers.


Natural Language Processing

Another, major area where financial institutions are using data centres and machine learning to advance the way they work is natural language processing. If you’ve ever watched older films or TV series involving financial services, you’ll have probably seen trades and orders being made over the telephone. While it’s true that significant amounts of the market have now moved to more digital means of execution, there is still a good deal of market making conducted by phone.

Increasingly organisations that participate in these parts of the market are employing AI to capture and convert the voice data into text, so that it can be better used in future applications, such as automating trade execution from voice commands, and to better aid regulators who closely monitor traders’ activity for any impropriety.

Hedge funds and quantitative traders are also applying machine learning to their operations, increasingly looking to alternative data to try to identify correlations and causations in market movements to inform their trading and investment strategies, and deliver the best returns to their customers.

These firms are increasingly pooling large data sets on everything from private jet charters to virus outbreaks and applying complex machine and deep learning models to better understand how the market behaves as a result, and how to exploit that to deliver value to.


The Hardware Challenge

However, running these models is something of a technology challenge. The infrastructure requirements to build and train these models is significant. So much so that highly specialised equipment is generally needed.

To build these models in an efficient way, hyperscale datacentres equipped with high performance GPUs and capable of handling very high-power densities are required. Locations that benefit from geothermal energy sources, such as those in Iceland are very popular for this type of workload.

Once these machine learning models have been built and trained, they can be run in a much more convenient environment for the financial services organisations who will reap the benefits – much closer to key liquidity venues with low latency access to key exchanges and market participants. This is where Interxion comes in to its own. Review our AI-as-Service Free Trial Service here.


How can Interxion help?

Strategically-located in central London, Interxion’s data centre campus is critical to Europe's trading infrastructure. Sitting between the Square Mile, the centre of global finance, and Silicon Roundabout in London’s Tech City, the EMEA region’s largest technology cluster – it the obvious choice of data centre partner for the country’s community of financial services and fintech businesses.

Close proximity to the London Stock Exchange, in addition to ready access to microwave connectivity to Frankfurt, make it a leader in terms of European trading venue access. It is also just a walk or Underground away from most central London offices. Indeed, the Interxion London financial services hub is one of the most established data centre communities in Europe. Some 200+ Financial Services organisations take advantage of this site for proximity, performance, reliability and access to a large and diverse customer base.

This ready-formed community of potential suppliers, partners and customers all in one place, sees a number of other trading firms from around the world interested in partnering with Interxion to ensure best execution on their global trading strategies, including those who run these intelligent trading algorithms.


Powering DX Innovation

The kind of remote model building described above has been happening for a few years now. What’s becoming more common however, is for enhanced processing capability nearer to trading venues. For example, Interxion partners with NVIDIA to deploy specialised AI chipsets and its DGX appliance in the London campus, making it easier to deploy these sophisticated market modelling and intelligence algorithms where they are most effective.

This latest generation of AI appliance optimises the number and power of GPUs being used – moving the competition in the market away from microsecond speeds, and towards the compute.

For communities of interest like the financial services ecosystem in Interxion’s London campus, the ability to cross-connect within the data centre itself, is valuable not only from a connectivity perspective.  The partners who would like to connect into that community are often providing data, testing and improving the models of Quant firms. The new frontier is to be able to connect in a much higher level of processing to inform trading decisions. Wider deployment of things like NVIDIA appliances in colocation data centres will enhance traders’ ability to crunch data and deliver results of that analysis on the timely basis needed to identify investment opportunities and make greater returns than their competitors.

Interxion is a DGX data centre ready partner – NVIDIA recognise that Interxion can handle the GPU loads of the kind of deep learning applications people are increasingly interested in running. Interxion is onboarded with the latest appliances to reduce any obstacles to adoption and are now beginning to see more customers exploring the use of the platform for their business. Find out more about our AI-as-a-Service Free Trail here.

Whether Financial Services CIOs want to buy their own HPC platform and install in their own racks, or work with a systems integrator to access that compute as a service, Interxion can be a valuable partner. The on-ramps to public cloud from the big players, AWS, Azure, and GCP, mean it is convenient to use Interxion as an inbound source of best execution.


Changing Dynamics

It’s fair to say, however, that the road that led us here is not the same one we see before us today. Up until the recent COVID-19 pandemic hit, you could say the markets were very efficient. Relatively few people over-performed based on speed; generally speaking, you had an efficient market, very low interest rates and low volatility. It was to boost returns in this environment, that trading firms were focussing on diverse strategies and leveraging machine learning, rather than outright speed.

Our current situation is quite different and volatile. From a trading perspective this offers more opportunities. We are now anticipating a severe period of disruption and it is impossible to predict exactly what that will mean. Whether financial services firms charge ahead with their digital transformation programs and incorporate greater levels of machine learning now, or postpone investment and wait until the worst of this crisis is over, remains to be seen. But, there is no doubt that when the markets do settle and an efficient, low volatility world returns, those with greater machine learning models will be the ones outperforming the competition.

Whatever your role in the market, Interxion’s London Data Centre Campus provides an ideal location to run your workloads and benefit from the most connected data centre in central London, with unrivalled access to the world’s markets.  It could be a good time to find out more.