Machine Learning Still Needs A Little Human Touch

People aren't quite obsolete yet.
Publish date:
Updated on
By D J Shin (Own work) [CC BY-SA 3.0 or GFDL], via Wikimedia Commons

By D J Shin (Own work) [CC BY-SA 3.0 or GFDL], via Wikimedia Commons

The view in large swaths of the investment community is that machine learning is about to revolutionize trading and investments. The stark reality does not live up to the hype.

Machine learning – or the idea that machines can recognize patterns in data and capitalize on them – is a concept that seems tailor made to the data-intensive world of investing. Unfortunately, many practitioners are applying machine learning in ways that are not sustainable or just plain wrong.

Hedge funds have been hammered by eight years of underperformance versus markets, and increased competition in the industry has made the situation progressively worse each year. Against this backdrop it is not surprising that 58% of fund managers in a recent KPMG survey cited the transformative potential of machine learning. Indeed, industry leaders like Renaissance Technologies, AQR, Two Sigma, and Bridgewater have been using various forms of data crunching models for years to improve investment performance.

Yet what many fund managers both new and experienced fail to realize is that incorporating machine learning into an investment process is hard. Using machine learning for investment really is a process in itself, and managers will fail if they try to simply slap data algorithms onto of an investment process without thinking through how the new business works.

Indeed, no less an expert than David Siegel, co-founder of Two Sigma Investments, cites the difficulty in using machine learning in investments. At a Bloomberg conference last September, Siegel said, “I’m concerned that people may have unrealistic expectations of what is possible with the current state of the art… Machine learning systems can easily with high confidence make mistakes.”

One big problem with machine learning at present is that investors often latch onto any pattern in data and try to use that as the basis for an investment strategy. They also fail to build complete and comprehensive datasets properly, leading to problems in data transposition errors. And machines are very poor at recognizing let alone adapting to new situations and experiences (think Brexit).

These type of mistakes are easy to make – professionals working full-time with data in many fields make such errors all the time. In the investment world, the problem is that such issues can go unnoticed easily and cost investors significant sums from bad trades.

None of this means that investors should forgo machine learning – the experience of hedge funds over the last eight years shows the issues associated with that approach. Instead it means that the financial field needs to surround itself with people that understand the industry and understand risk.