What is artificial intelligence?
AI traditionally refers to human-like intelligence, such as learning, reasoning, and processing language, being carried out by machines through advanced modern technology. While many view AI as “machines taking over the world” this is hardly the case. In fact, humans are an intimate part of the AI process as they are needed to interpret output. Over the past two years, there has been an incredible uptick in data volume driving the need for more AI.
90% of today’s data has become available during this time. To put this statistic in a layman’s context, within a single minute, 3.5 million Google searches occur. AI is responsible for delivering search results back to the user against a massive backdrop of possible information. Foot traffic, credit card transactions, social media content and all other alternative data are collected and fed into machines, enabling AI to extract valuable insight like never before.
When it comes to FX, according to statistics released in April 2019 by the Bank for International Settlements (BIS), trading in foreign exchange (FX) markets grew to US$6.6 trillion each day, up 29% from US$5.1 trillion three years earlier (Figure 1). As trading continues to increase, more decision-making will rely on advanced analytics which are able to deal with voluminous, fragmented information. The nature of the FX market, which is an over-the-counter marketplace consisting of both cash and derivative instruments tests the boundaries of AI for both trading and analytics.
How has AI fostered better decision making for FX trading?
The bulk of FX trading currently takes place over multi-dealer platforms, or MDPs. Trades are sent out (often manually) to a handful of dealers using a request-for-quote (RFQ) protocol so that prices can be compared and the transaction takes place at the best possible level at that moment. While there is certainly value in collecting competitive prices – just as a consumer might do when it comes to choosing the best flatscreen – features and timing also matter.
AI lends itself well to the data-driven process which creates a feedback loop informing the trader of his or her best options prior to execution. For example, transaction cost analysis (TCA) solutions providers are using AI to develop models that can predict different regimes in the market during the day given changes to volatility and liquidity as well as other exogenous factors. These models provide on-the-fly intel informing the selection of optimal execution methods or algorithms which perform better in certain regimes according to attributes such as instrument type, currency pair, and size. In the flatscreen example, this process would be akin to understanding which retail stores had adequate stock and offered the best prices.
In terms of actual automated execution, advanced technology has influenced a sea change in everything from staffing to systems. For instance, traders today are a different breed compared to what they used to be. Gone are the days of the most persuasive personality winning the seat. Today’s traders are coders and coders are traders, particularly in data-driven markets like FX, equities, and futures. AI and its subcomponent (Figure 2) have fostered the development of trading algorithms on the buy-side and sell-side. In some cases, the most routine trades, like FX hedges, are fully automated from the time they are loaded into an order management system to post trade processing. Routing decisions for these trades are a consequence of AI-driven analysis selecting the best liquidity sources for such trades.
Another particularly interesting development is the use of natural language processing (NLP), which sits in the domain of AI. NLP has applications in everything from sentiment analysis to credit ratings, as well as the automation of RFQ using chatbots. The rise of alternative data over the past five years is the result of the advances in big data infrastructure and AI for news text extraction. The development has led to both advances in trading logic and opportunities for investors to look for untapped alpha in alternative data.
Machine learning (ML) is a cross disciplinary field that combines statistics and computer science and, like NLP, also sits within the domain of AI. While there is a lot of hope that more financial markets use-cases will continue to develop, these are still early days. ML has been adopted into the front office via quantitative strategies for asset allocation, and other points along a firm’s workflow like risk management. Trading data doesn’t lend itself as well to ML, however. This is because market data is filled with noise around signals. The dynamic nature makes it difficult to teach a machine to learn – even if there are some known factors which influence the market.
Finally, deep learning (DL) is part of a broader family of machine learning methods that mimic neural networks in the human brain. Whereas ML can be as simple as running a regression, many quantitative analysts use DL algos which are constructed with “connected layers” which means they can learn automatically without predefined knowledge being coded in.
Where do we go from here?
While AI is powerful, its adoption in the financial industry has been rather slow. However, all roads seem to be heading in the right direction to foster more AI development and an uptake in adoption. Given digital transformation initiatives happening at many banks, the sell-side is now opening innovation labs and collaborating with fintech companies of all sizes - even household name tech giants like Amazon, Google, and Microsoft. To unleash AI’s full potential in finance, the industry needs more experimental projects designed to tackle inefficiencies. Pushback comes from the fact that only a few of these will have near-term success, although the cumulative impact would be undoubtedly beneficial.
So what can we expect in the future? As human psychology catches up with technology in OTC markets like FX, the world is likely to see the shifting towards the use of more AI for the development of algos, predictive analytics – especially in more opaque FX instruments - and NLP in trading applications. Although human nature is a formidable challenge, it will be interesting to see where the regulatory community lands on this topic. Should concerns be raised around the transparency of trading algos and use of AI given its complexity and “black boxed” nature, the lingering fear of consequences from computer trading like 2010’s flash crash may force the industry to take a step back.