The Gateway to Algorithmic and Automated Trading

Machine learning boosted by big data in finance

First Published 14th January 2015

Though far from mainstream, machine learning continues to gain momentum in financial services even as experts warn of the consequences of artificial intelligence technologies.

Steve Wilcockson, MathWorks

Steve Wilcockson, MathWorks

"...we need to think about machine learning and AI very carefully, work together closely on it, understand the perils and pitfalls, opportunities and ultimately come up with some best practices."

Massive IT disruption is providing the financial industry with new avenues for business and managing risk, according to a survey by MathWorks.

Underpinning that disruption is a blend of 'big data' technologies intersecting with modelling and analytics, both simple and sophisticated.

Steve Wilcockson, industry manager for Financial Services at MathWorks, said that when the survey was first conducted in 2012, the term big data had the feel of hype around it. Now, its importance in shaping the industry's future seems to be taken for granted.

"There has been a lot of work done in terms of data aggregation at the mechanistic level - data exploration, consistency, interpolation - to make sure we are working with even data," he said. "There are also developments around some of the areas of text, or unstructured data."

Moreover, firms are paying attention to the implementation of new data environments, such as the shift from SQL to NoSQL databases. There's still a long way to go however.

"Many of the larger institutions have got a big data project on the go. Some of them have defined what that project looks like, others are looking for projects," he said. Some 31% of survey respondents reported that their institution has implemented a project to react to a specific big data challenge.

Trade execution is one such project being pursued by asset managers, Wilcockson noted, as the industry starts to learn the benefits of understanding and modelling the nuances of execution in the shift to on-exchange dealing. Considering the complexity of market structure these days, a tier one asset manager wants to know that a large order for single stock is filled the way intended.

With a little help from robot friends

The survey shows that some 12% of respondents said they applied machine learning technologies. However, there is clearly interest from across the industry to take a closer look - 40% of respondents said they want to understand more about the different machine learning methods.

Wilcockson said that when he talks to people from the banking sector, he sees interest in developing fraud monitoring systems, or improving trading strategies albeit tempered with caution. Two major hang-ups have to do with overfitting and blackbox modelling.

"One of the issues with machine learning is there are lots of different models, with lots of different nuances which can have multiple variability when applied to different data sets," he added.

A "middle way" to automation, said Wilcockson, is emerging as fund managers use it to validate human trading decisions, where the automated is working in sync with the human.

The survey comes at a time when artificial intelligence is making headlines. Renowned physicist Stephen Hawking and CEO of Tesla Motors Elon Musk have both sounded warnings about unfettered advances in artificial intelligence. Musk went so far as to write that it could be "potentially more dangerous than nukes".

The concerns have resulted in an open letter from the Future of Life Institute, which calls for AI experts to make a pact to develop artificial intelligence technologies responsibly. Aside from Musk and Hawking, signees include academics from top schools as well as business leaders from Microsoft, IBM and Google.

If a dystopian future of man-machine warfare seems a bit Hollywood blockbuster, it probably is. But the open letter, which highlights characteristics of responsible computer science, does resonate with MathWorks' survey results, Wilcockson said.

"What you are seeing in the survey is kind of the practical insight into what the open letter recommends, which is that we need to think about machine learning and AI very carefully, work together closely on it, understand the perils and pitfalls, opportunities and ultimately come up with some best practices," Wilcockson said.

Still, it is early days, and only the beginning of a cycle that is also a second wave of interest in AI applications to financial markets. Neural networks for example were a hot topic in markets not too long ago, but ultimately fell out of favour.

But this time is different, and the difference is big data.

A year from now, Wilcockson expects the financial sector, particularly the institutional side, to be testing some of these capabilities. "They will have got their hands dirty," he said, adding that beyond 2016, a more systematic roll-out could be in the cards.

He emphasises that there are many angles to the big data space - looking at how models are calibrated, the spread of machine learning, more sophisticated statistical methods, or investigating data series.

"We will see a trend towards optimisation - how we navigate our parameter space to hone in onto the best (data) points. There is a big element of the quant world, of the computer science world, that is focusing in on this area alongside the AI and I think the two areas along with the data side are all intersected," said Wilcockson.

The survey of 78 professionals was conducted at the MATLAB Computational Finance Conference in London in June 2014. Respondents included buy and sell side, as well as consultants and central bank representatives.