(MENAFN- ValueWalk)
'Machine learning' is all the in hedge fund strategies, as fund managers race to prove they can play the latest trend, and quantitative scientists . One traditional trend follower, Man Group, with a reported $43 billion in assets managed quantitatively, sees it as the 'next phase' in systematic trading. But there can be landmines or at least potholes going down the path to machine learning investment returns.
Get The Full Ray Dalio Series in PDF Get the entire 10-part series on Ray Dalio in PDF. Save it to your desktop, read it on your tablet, or email to your colleagues
/ Pixabay
Man Group Chief Investment Officer Sandy Rattray looks at machine learning as the future that is playing out today. The firm, which spawned some of the original quantitative traders such as David Harding, who left the firm to form Winton Capital, is forging into the machine learning, a technological approach that they claim is already generating returns.
Rattray told that in five years, he wouldn't be surprised if half of all quant trading was supported by machine learning.
"We will see it steadily increase in all areas and environments where we are less dictating what the model should do and more letting the model learn from the data on its own," he said. "Can they the earnings calls as well as a human can? No. But they can do a lot more. You get a lot more breadth out of the computers than you do out of the humans."
The behind machine learning is that the computer is programmed to recognize patterns and then make trading decisions on its own. While many quantitative investment firms are engaged in this practice, Rattray is the only known fund manager to publicly claim it is delivering positive performance.
While his is a story of success, there can easily be a significant amount of capital thrown to if managers are not prudent in their decisions.
In a September 2 presentation, Marcos Lopez de Prado at the Lawrence Berkeley National Laboratory Computational Research Division outlined 'The 7 Reasons Most Machine Learning Funds Fail.'
The first pitfall he cites is 'The Sisyphean Quants,' where quantitative minds are set off on their own without any interaction with other portfolio managers. He says the typical approach is to 'hire 50 PhDs, and demand from each of them to produce an investment strategy within 6 months' without cross-pollination of market ideas.
'This approach typically backfires,' he wrote, 'because each of these PhDs will frantically search for investment opportunities and eventually settle for: A false positive that looks great in an overfit backtest; or a standard factor model, which is an overcrowded strategy with low Sharpe ratio, but at least has academic support.'
He says every successful quantitative firm he is aware of applies a 'meta-strategy paradigm,' where a team of quants works together on sub-tasks 'while having a holistic view of the entire process.'
Other factors are balancing memory optimization, and including a sufficient sample size when studying data are also key points where quantitative scientists fall down. Wrong labeling of data, a weighting of non-IID samples, cross-validation leakage and backtest over-fitting are also issues.
Tammer Kamel, chief executive officer of Quandl, an provider for quantitative scientists and/or algorithmic traders, says that curve fitting is the biggest issue he sees in machine learning development.
Machine learning, like all forms of statistical inference, is not immune to the risk of 'curve fitting' which is what happens when one sees a pattern in something that is actually random,' he told ValueWalk. 'The laws of statistics are immutable and ML practitioners ignore them at their peril.'
He also noted that insufficient sampling is a problem, advising that quantitative scientists stick to prescribed methods.
His tip for success?
'The best machine learning and artificial intelligence any form cannot compensate for bad data,' he said. 'Practitioners should be zealous in ensuring they feed their algorithms reliable data and, ideally, data that has not been scoured over by all of Wall Street already. It is 'new' data combined with disciplined machine learning algorithms than that poses an existential threat to the status quo on Wall Street.'
Legal Disclaimer:
MENAFN provides the information “as is” without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the provider above.