Machine learning in the South Korean markets: an Asian case
The use of machine learning in finance has ballooned in recent years, showing rapid expansion across all sectors globally. Asian firms are sometimes cautious about embracing new technology. Nevertheless, once that hurdle is overcome and they understand the advantages available, they can be even swifter than their peers in Europe or the US to adopt the very latest technology available. Consequently, now that it is clear that artificial intelligence (AI) is vital to success in finance, the markets of Asia are welcoming it wholeheartedly, with market participants on all sides increasing their stakes in AI development.
Asia’s financial institutions are particularly looking to machine learning, that subset of AI that focuses on teaching computer systems to ‘learn’ independently from data, as a way to automate processes and increase efficiency. Of the countries in Asia that are diving into the field, South Korea has made some high-profile strides in recent years.
According to a 2017 survey by Korea’s Institute for Information & Communications Technology Promotion, the country ranks number three when it comes to holding AI-related patents after the US and Japan, using data from January 2005 to September 2017. And top brokerage firm Mirae Asset Daewoo recently announced a partnership with Internet portal Naver to invest KRW 500 billion in AI and other digital projects.
Bouncing back from a difficult start
South Korea’s enthusiasm for AI, however, has historically been tempered by some challenges. The biggest initial hurdle was language. Much of the open source material shared on blogs and other platforms has been in English, and focuses on how to process English text. It’s taken some time to adapt the open source natural-language-processing algorithms to be of use in the Korean language.
Also, regulators in the region tend to be more conservative. Securities firms are often requested to explain certain trades. The processes behind automated trading algorithms are harder to explain than those of manual trades, especially in very volatile markets. Consequently, automated trading is taking longer to take off in the region.
Information inequality also reduces the incentive for adoption. Much of the value of AI is its ability to crunch massive amounts of data and come to conclusions that others might otherwise miss. But that ability is valuable only when there is information equality. If everyone gets more or less the same information at the same time, then the value added is the ability to do something different with that information. But in many Asian markets, there is less information equality than in the US or Europe, meaning that stocks regularly move before the news comes out to all.
News is also a less prominent feature in decision-making, and this also has the effect of tempering the value of machine learning. While news plays a prominent role in decision-making for Western investors, it’s less integrated in the investing process for those in Asia. Survey data from Bloomberg’s recent Machine Learning Decoded event in Korea found that 67% of conference attendees believe news sentiment is just one of many factors that drive stock prices. And news, while important, is not yet integrated into the workflow in an automated fashion – only 3% of participants said they have direct access to a machine-readable newsfeed and apply a natural-language process to it. The largest group, at 42%, said news is not incorporated into the workflow at all, while 30% say they check top headlines manually. So while the nation shares some of these issues with other Asian countries – Japan shares restricted adoption due to regulation and limited access to coding in its native language, for example – the extent of the issues is greater in Korea and will likely require a different approach to adoption than the rest of Asia.
Making progress in the back office
Perhaps as a result of these many hurdles, Korean institutions seem to be currently spending most of their time automating their back office rather than focussing on the decision-making aspects of automation. Firms are realising that in order to be competitive they must automate, especially in processes surrounding real-time pricing. About 26% of attendees at the above Bloomberg event stated that trading automation is critical, while another 20% said trading automation is critical but that they weren’t completely comfortable with the concept. That means that about half of the audience saw some value in trading automation, whether actualised or not.
In fact, 34% of conference attendees polled said they are using machine learning solely for ad hoc research purposes. Another 11% said they are using it to generate signals or factors, and 15% said they are using it systematically in production. The numbers point to a landscape where institutions are experimenting with the possibilities to see what works.
The current use of machine learning in Korea is primarily aimed at creating efficiencies in pricing and execution. With pricing, machine learning can help to hedge market risk – for instance by adjusting foreign exchange exposure whenever a central bank announces sudden tapering or easing in anticipation of volatility. Researchers are also looking at machine learning to gauge market and news conditions to help traders choose which trading algorithm to use at any given time automatically with little supervision.
These are still early days for AI in Asian financial ins titutions, but progress is now being made more quickly. More partnerships and projects are sure to follow soon as small successes point to greater competitive advantages.