Digitisation and the role of technology have remained key industry themes moving into 2018, with smart automation, artificial intelligence (AI) and robotics changing the way firms in the derivatives market think about how their work is executed across the back, middle and front office. The march towards AI and machine learning has been difficult to ignore with conversations on the technology’s impact featuring heavily at industry events over the past year.

The International Data Corporation, headquartered in New York and providing market intelligence and advisory services, has predicted that spending on AI and machine learning will grow from $12 billion in 2017 to $57 billion by 2021. Acquisitive activity involving AI firms also shows investment in the field over the last five years has accelerated rapidly, with Q1 2017 peaking at 37 deals compared to Q1 2012’s two.

Artificial intelligence is not a new technological phenomenon. First coined by American computer scientist John McCarthy in 1956 – at the time defining it as “the ability to learn without being explicitly programmed” - AI has been explored by more familiar names like Alan Turing who, in his 1950 book ‘Computing Machinery and Intelligence’, touched on the ability of machines to simulate human beings and do intelligent things like play Chess.

In the 1990s, helped by the advent of the World Wide Web, machine learning became famous as data and statistics flooded the financial services. This period was described by some as the “golden era” of machine learning with computer science giving birth to probabilistic approaches in AI – perhaps exemplified most famously by Deep Blue, a computer developed by US technology firm IBM, beat the world champion at chess in 1996.

Hollywood too has fanned popular fascination with AI and machine learning. Central to the 1985 science fiction film The Terminator, Skynet was an artificial intelligence system which eventually became self-aware. Yet today, over 20 years on from Deep Blue, AI has today become almost a glorified catch-all for a number of quantitative analytics in various commercial industries – with no sign of Skynet yet.

The capital markets has been no stranger to technological innovation like AI and machine learning – technologies have been for decades touted as being able to perform better than humans, to improve speed, quality and competitiveness. Institutions across the buy- and sell-side are now using them to optimise scarce capital, back-test models, analyse risk across the trade lifecycle or study the impact of trading large positions.

Regulatory compliance, fraud detection, data quality assessment and trading surveillance are also areas benefitting from advancements in AI and machine learning. On the buy-side in particular, firms are using it to seek alpha – to ensure higher uncorrelated returns while optimising trade execution.

“AI has now become a ‘hot topic’ because algorithms have been developed quite significantly over the past years and hardware is now much cheaper,” explains Veronica Augustsson, chief executive officer at Swedish technology firm Cinnober. “Historically you needed quite extensive machinery in order to run those algorithms so it wasn’t worth the investment. Now we are in a very exciting moment because these algorithms have been developed further and hardware is cheap so a lot of players can use it.”

Data is king

The main advances in AI over the last thirty years have concerned machine learning – a subcategory of AI focused on the development and use of increasingly advanced search algorithms and the integration of statistical analysis in understanding the world around us.

While AI has a certain overlap with machine learning, there are some distinct differences between the two, with some features unique to each technology. The building blocks for both machine learning and AI are algorithms – a series of programmed instructions, written by a human and followed verbatim to direct how a piece of software will behave. However, machine learning programmes are able to write programmes themselves, even adapting their algorithms to perform better in the future.

Despite these nuances, both AI and machine learning rely on one key input – data. Enhanced computing power, combined with increased online storage space, has caused an explosion in the amount of data available to firms and financial regulators. Compliance requirements for firms, in sweeping post-crisis regulations like Dodd-Frank in the US and Mifid II in Europe, mean data is increasingly seen as a resource rather than simply a by-product of conducting business.

“AI gives us the ability to process large amounts of data quickly and accurately,” notes Michael O’Rourke, head of machine intelligence and data service technology at US exchange giant Nasdaq. “So, when data driven decisions, it can bring more data into view and can lead to better outcomes (for example when making trading and investment decisions).”

“Because of the nature of electronic markets, you have so much data right now that it can just be applied in any area where you’re going to create efficiency by being able to manipulate data,” adds Jay Biondo, product manager, surveillance at US firm Trading Technologies. “Any time you can use machine learning to, not only manipulate but, interpret and classify data to add insight for humans, is where the most value for AI lies.”

The aggregation of data, combined with the pooling of resources like data lakes – storage repositories that holds a vast amount of raw data in its native format until required, has opened up opportunities for AI to really prove its worth – with humans unable to match the speed at which data can be analysed.

“Our data lake is growing so fast – almost doubling on a yearly basis – so we need to use new differentiated techniques to refine all of it and make sense of it,” points out Matteo Andreetto, chief executive of Deutsche Bourse index business Stoxx. “At the point where that data not only doubles but grows faster, then how many people can we hire? The amount of information we can digest and process using this technology is significantly superior.”

Yet, this explosion of data doesn’t necessarily guarantee optimum results for firms using AI technologies, as Yvonne Zhang, chief executive officer and co-founder of Singaporean-based think tank the Aquifer Institute, points out. “The quality of data that gets put into the analytics impacts how good the employed methodology is.”

AI & DLT

With data pools growing exponentially year-on-year and cloud computing becoming even more powerful, opportunities to develop AI appear to be almost endless. Yet this optimism brings with it both legal and practical challenges, which also presents opportunity for other nascent technologies like distributed ledger technology (DLT).

“I think data will always be a problem,” explains Nigel Solkhon, chief executive officer of trade association ISITC Europe. “Until industry participants all use the same data flows or the same databases, there will be a need to data translation enrichment.”

Data enrichment refers to the process of enhancing, refining and generally improving raw data, making it more valuable as an asset for a firm. The process allows users to get more out of the data on hand, making it more accessible and being more proactive with its use.

“If you look at the typical order flow, there are different data sets that are used at various points,” Solkhon adds. “Now I think we’re getting closer to higher quality data if you look at the way AI can be applied to things like DLT which allows for cleaner data to be transferred upstream and downstream.”

However, while the opportunity to unite DLT - a digital asset database that can be shared across multiples jurisdictions or institutions – and AI is high, according to O’Rourke, it is “still early days”.

“We’re just seeing DLT commercialisation becoming mainstream and it’s too soon to know the impact the two technologies might have on one another,” he said.

“We’re at the stage now where there’s going to be hundreds of DLT interfaces that people have to use,” adds Solkhon. “Our job as an industry is to seek a common approach like an API (Application Programming Interface) and apply it to that space. I truly believe data is one of the key things we have to resolve for the future.”

However, this is not to say that the legal ramifications of the technologies are shared by both. While AI and machine learning aid decision making in risk and compliance, they do not replace it. They definitely do not trigger outcomes either, according to Zhang.

“DLT and the implementation of smart contracts with the sharing of potentially private information and the removal of centralised decision-making to a more decentralised systems is at odds with a lot of traditional common law and civil legal systems,” she explains.

Having said that, the consequences of using AI across the value chain will always pose risks for firms – many of the same risks associated with incorporating any relatively nascent technology. “If Apple is using AI in its phones to train Siri, it’s not the end of the world if something goes wrong with the algorithms in the software and someone calls their grandmother instead of their grandfather,” notes Augustsson.

“But the consequences of a malfunction can be much worse in other industries like the capital markets. So it’s great that Apple and similar companies are using it where the consequences are less critical. People like us are still learning from their findings and algorithms.”

In the middle and back office there are also more legal ramifications in terms of integrating AI, according to Biondo. “I think humans still have to be heavily involved in the middle and back office. The AI can create efficiencies by providing valuable information, but the human expert should still draw their own conclusions – the AI shouldn’t necessarily draw legal conclusions,” he said.

“There’s also more work to do than just trade surveillance or anti-money laundering (AML) surveillance – these are just facets of what a compliance department should be doing. There’s so much more as far as developing effective policies and procedures and other areas of compliance that need to be managed by people.”

A deep learning future

Looking ahead, some believe the future of AI in the derivatives space will be dominated by deep learning – a method of machine learning inspired by the workings of the human brain which has existed since the 1960s but again has benefitted from the explosion in data and advancements in computational power.

“These could be machines or algorithms learning from other machines and algorithms,” Biondo explains, “…where they learn against each other in incremental steps and improve and adapt without the supervision of the data scientist.”

This is because the computer, and the algorithms contained, gather knowledge from experience without the need for a human operator like a data scientist to specify the knowledge needed by the computer.

“I think the front office is way ahead in terms of using AI – they already using deep learning techniques and deep learning algorithms,” said Biondo. “Where you will see a catch-up in the middle and back office will now likely concern the testing of algos. We’re seeing increasing requirements for algo testing in Mifid II, SEC 15c35 and also the Commodity Futures Trading Commission’s (CFTC) Reg AT if that gets passed.

“There’s going to have to be a process whereby the algos that are being created by the front office are tested and approved by the middle and back office. I believe you’ll see potentially the CFTC and/ or the Securities and Exchange Commission start holding people within the organisation, and even potentially third party vendors, responsible for testing the algorithms more thoroughly and understanding the implications of what an algorithm will do before putting it into production.”