Markets In Flux Have Stymied Quants. Can Machine Learning Hedge Funds Prevail?

The problem with quant investing may have been that the algorithms were unprepared to deal with situations they had not been programmed to expect. Machine learning investing strategies may improve performance by learning from experience rather than strictly historical data.

The question is whether too many funds using similar models to chase the same indicators will reduce financial returns. Again. JL



Justina Lee and Ksenia Galouchko report in Bloomberg:

Whipsawing markets and the over-crowding of quant strategies have battered performance. Machine-learning takes quant investing to the next level because the robots are programmed to adapt and improve their performance based on the data they sample over time, without needing explicit human instructions. Machine-learning opens the prospect of uncovering new signals in large masses of complex data to boost returns. “A machine is going to be faster than human beings when it comes to detecting new alphas."(But) most funds based on machine learning fail, (because)“pervasive misuse” of techniques lead to false positives, losses and failures.
Maybe machines can figure out this crazy stock market.

At least that’s what quantitative traders who have struggled to beat the market for years will be hoping as a band of their peers roll out computer-driven strategies that learn from their own mistakes.
Lynx Asset Management, for one, is planning a new fund in October that executes strategies thought up by a machine—an approach that helped the $5 billion Swedish hedge fund beat most of its trend-following rivals in 2018.
It won’t take long for funds managed entirely by robots to be everywhere. Two to three new ones will start trading each month this year, reckons Alex Allen, who runs a fund for EFG Asset Management that invests in machine-learning strategies.

“It’s the next evolution in the investing arms race,” said London-based Allen, who already invests in eight machine-learning funds, and plans to add two more soon.
Machine-learning takes quant investing to the next level because the robots are programmed to adapt and improve their performance based on the data they sample over time, without needing explicit human instructions.
That tech-savvy, data-driven quants are dabbling in the field is hardly new — it’s been touted as the next big thing for years, and the tools have been getting cheaper and easier to use all the while.

What sets the latest flurry of activity apart is the backdrop. Whipsawing markets and the over-crowding of many quant strategies have battered their performance and started to undermine investor enthusiasm for this once red-hot corner of the investing world.

Proponents say machine-learning has the potential to give quants the winning edge they’re missing, in part because it opens up the prospect of uncovering new signals in large masses of complex data to boost returns.
Yves-Laurent Kom Samo is so much a believer that he left
his career at Goldman Sachs Group Inc. and JPMorgan Chase & Co. in 2013 to get a doctorate in machine learning at Oxford University. Now, he’s getting his own business, Pit.AI in San Jose, California, off the ground with a machine-learning hedge fund KXY Singularity, which started trading in March.

“A machine is going to be considerably faster than human beings when it comes to detecting new alphas and a machine is going to work at a scale that no human being can,” Kom Samo said.
Lynx Asset Management AB will unveil its new offering, the Lynx Constellation fund, on Oct. 1 and it will essentially be a long-short strategy investing in futures and forwards across various asset classes. The actual strategies used could involve everything from trend-following and contrarian trading to relative-value.
That’s part of the challenge in judging the effectiveness of machine learning for quantitative trading. Quants use advanced mathematical models to determine investments in what is already a huge and varied field of finance, and the application of these techniques promises even more complexity. Machine learning itself covers a wide spectrum of methods, from traditional statistical analysis to mimicking the way neurons in the brain provide layers of learning.
“In terms of the scope of data that we can analyze, it’s much greater than before,” said Boyan Filev, the London-based co-head of quantitative equities at Aberdeen Standard Investments. “We’re analyzing billions of data points every month.”
A Eurekahedge index tracking those using artificial intelligence and machine learning has outperformed the past five years. Still, because the index covers funds with a wide range of strategies, it’s hard to draw broad conclusions.
Quant researchers at Societe Generale SA have developed a U.S. equity index based on the bank’s machine-learning model. The long-short gauge beats the HFRX Equity Hedge Index handily on a one- and five-year basis. But so far in 2019 it has fallen behind, so it's not quite conclusive.

Aberdeen Standard’s $10 million Artificial Intelligence Global Equity fund, which analyzes stocks for so-called factors such as momentum or value, has gained 9.1% this year, compared with about 11% for the MSCI All Country World Index.
To Acadian Asset Management, a quant firm with $96 billion, machine learning is handy for analyzing complex data sets, but isn’t mature enough to take over the human reins just yet. It’s also adding unwelcome complexity to an already intimidating trading landscape.

Investors are “forced to try and make this distinction between the ever proliferating set of strategies,’’ said Seth Weingram, the Boston-based senior vice president and director of client advisory at Acadian. “You run into a bit of a Wild West-sort of environment happening out there with people making claims about machine learning and it’s not entirely clear how to make this judgment.”
Plus, many new techniques come with the same old pitfalls, like the risk that computers will discover patterns that don’t actually work in the real world. Only this time, the model may be even harder to deconstruct to an irate client.
“It’s very important that the algorithms use information-rich data and not all possible data out there because in financial markets, there’s so much noise,” said Martin Kallstrom, a partner at Lynx in Stockholm.
In a 2017 paper on why most funds based on machine learning fail, AQR Capital Management LLC’s departing head of machine learning Marcos Lopez de Prado said the “pervasive misuse” of such techniques by quants will continue to lead to false positives, losses and failures.
Unbowed, believers like Kom Samo, the Oxford grad, are stepping up. His backers include Howard Morgan, a venture capitalist and co-founder of Renaissance Technologies LLC, and Michael Seibel, chief executive officer of Y Combinator. His fund uses both long and short strategies and trades futures across asset classes.

“The role of the quant in the future is not going to be to find investment ideas, but to design mathematical processes to empower machines to find investment ideas,” Kom Samo said.