Head of trading and market data, UK
14 July 2021
5 min read
In 2019 Gartner predicted that AI software would create $2.9 trillion of business value in 2021. Indeed, 80% of firms in this research are looking to, or having already, adopted this technology in the front office. But what are its limitations? Do current markets present a situation where AI and machine learning can help with trading strategies?
AI is only as good as the models being built, which rely, in part, on machine learning, which in turn is reliant on the underlying data consumed. The old adage of junk in, junk out still applies - but what constitutes junk? Is it incorrect data, too short a time series, or misinterpreted data?
One study mentioned that in early 2020, complaints about Amazon scented candles more than doubled. Looking deeper, it coincided with the rise of Covid-19 when many people were losing their sense of smell. Data has to be interpreted correctly and fully understood, if it’s not then it has no value rendering any AI model useless.
Let’s take the rise of retail as an example of how markets have changed in recent times, could models have given firms an edge?
At Iress we’ve seen a 200% increase in retail volume over our UK RSP network to our market makers. In the US, some reports show 25%+ of trade volume is retail.
Technology change facilitated the social media boom resulting in message boards, such as Wall Street Bets, attracting millions of participants acting more like one cohesive unit. Meme stocks have appeared and gone viral. Barriers to entry have been lowered with low / no commissions and fractional trading. Social influence played a part in a move against the finance machine.
The result? Sentiment based trading took over in many stocks now familiar to anyone who reads the news. We have adapted by introducing increased market maker protection mechanisms for our clients but could machine learning and AI have helped clients further?
Stocks evidently behaved “irrationally”, so it appears no amount of historic data could have predicted the sudden change. Alt data requests rose as firms tried to understand what was going on in chat rooms and bulletin boards. However, even the ‘meme language’ used on such platforms changes and evolves over time. With the basic language constantly changing, what hope do machine-based models have to understand what’s going on?
While it is hard, near impossible, to predict the future of markets, models can help consume vast amounts of machine-readable news, understand algo performance, create know your client models, trade ideas or enhance risk analytics.
‘Enhance’ is the key word here.
AI and Machine learning is not going to take over the desk. It’s a set of tools that hopefully provide traders and managers with more insight into their business, clients and markets. By embedding these models into the workflow process they have the potential to add value.
The traditional trading tools on the desk all need to work together to support machine learning and AI. System data must be sanitised and readily available on robust APIs, the models’ insights need to be surfaced within the user's workflow and used to enable automated decisions where confidence is high. Crucially, the OMS on the desk must be open and flexible enough to cater for this shift in technology and the skillset of the new generation of front office teams. So whilst AI and Machine Learning may just become part of the furniture there’s still a lot of work to be done on the house first.
This article was originally published in the TradeTech report - Data analytics and trading efficiency post COVID-19: How AI and Machine Learning are helping to overcome volatility for the buy-side and sell-side.
Discover more about the adoption of new technologies and the solutions your peers are putting in place to respond to current challenges.