Back to Portfolio for the Future™

Big Data is Old Hat: Machine Learning is Hot

A year ago, in a report on Big Data and investment management, Citi Business Advisory Services predicted that “with the improved volume, velocity and variety of data inherent in the big data approach, the innovation seen in systematic trading models over the past decade could accelerate.”

One of the platforms highlighted in the Citi report was DataSift, a service that promises to “integrate social, blog and news data in a single place.” Or as Citi put it, DataSift aggregates “marquee data source partners, including Edgar Online, Wikipedia, and WordPress.”

Edgar, of course, is consistent with old-fashioned ideas of what hedge fund managers through various third parties should keep track of.  But Wikipedia presence on this short list might pull up short those who still think of it as a pastime for nerds who like to think of themselves as editors. And WordPress? The blog hosting service? Even more so.

The Citi did anticipate that the world wouldn’t always require humans to turn Big Data into trading decisions. The report said, “If current trends progress, there could even be a convergence of technical analysis and quantitative fundamental trading models that may result in a completely new type of automated portfolio management.”

The New Frontier

Now, on the first anniversary of that report, Eurekahedge has affixed to its monthly account of hedge fund performance and trends, a report on “artificial intelligence” as the new frontier for hedge funds.

Algorithmic trading part of the hedge fund mix for a long time now. Big Data? Well, it is hardly still in the crib itself. So what is new when Eurekahedge or other knowledgeable people and institutions speak of artificial intelligence as a trading strategy?

What is new is: machine learning, and over time machine self-programming. This isn’t some human writing a program telling a machine how to trade. This is a human writing a program telling a machine how to trade in the first instance, and how then to assess its results, and re-write its program to trade differently, and then re-assess the results, and so forth.

The last stage, perhaps, before a robot appears out of nowhere determined either to kill or to rescue John and Sarah Connor.  And then run for Governor.

AI/machine learning hedge funds have consistently posted better risk-adjusted returns than their peers, Eurekahedge tells us, with Sharpe ratios of 1.51 over the last two year period annualized, and 1.53 over the last three year period. The two-year annualized return in 10.56%, with vol at 6.31%. For purposes of comparison, the two-year annualized return of the Eurekahedge Trend Following Index is negative (-1.4%), with a vol of 8.07%. The two-year annualized return of the Eurekahedge Hedge Fund Index is 3.12%, with vol at 3.31% and Sharpe Ratio at 0.64.

Surprising the Overlords

Certain “idiosyncratic risk events” – the Brexit and Trump votes – have shown the continued limits of machine learning though, Eurekahedge says.  In the month of the Brexit win, the AI hedge funds were in the red. That was true neither of the Trend Following Index nor of the global Hedge Fund Index. Likewise, in June, the Brexit month, machine learning was in the red and uniquely so.

On the other hand, in the month dominated by the Greek referendum (June 2015) the global index and the trend following index were both in the red, the machine learning index was uniquely in the black.

The Eurekahedge report includes an interview with Yoshinori Nomura, director at Simplex Asset Management, a firm founded in Japan at the end of the last century. Nomura says that he “conducted a lot of data mining projects” as a business consultant, before joining SAM in 2008. He created a machine learning model and then set out to find a seed investor for it.

His program began with “a single time series of historical price which derives momentum index and mean reversion index.” More important, his AI model “can have daily automatic learning and adjusting function which allows the strategy to maintain its predictive capability even if the market environment changes.”

He acknowledges, though, that his program “had the same problem:” as certain others, in failing to anticipate the stock movements after Trump’s win.

So at least humans do still have some surprises in store for our Robot Overlords.