Thursday, May 24

What the buy-side needs to know about the data explosion

LinkedIn Google+ Pinterest Tumblr +

The convergence of big data and machine learning is nothing short of a revolution and the buy-side is piling into the way that alternative data is fueling it, said Armando Gonzalez, CEO of data provider RavenPack at a conference last week.

Conference attendance data alone helped prove Gonzalez’s point: the audience was composed of 85% buy-side, of which some 36% were quants and 16% fundamental.

What the buy-side wants, and can now get thanks to technological developments in machine learning, are speed and accuracy in analyzing large amounts of unstructured content.

This can take a wide range of forms: website scrapes, language analysis, credit card purchases and satellite data, for examples.

Dan Furstenberg, head of data strategy at Jefferies, an investment bank, said that the last 24 to 36 months have brought an “explosion” of data usage, and the growth of “quantamental”.

“There are very few discretionary managers right now that are not using data in some form whether its traditional price volume or alternative data as well,” he said.

 

Read MarketBrains’ interview with asset manager Mediolanum on going quantamental

 

When people think of where it is that data scientists work, typically it’s at high-frequency shops or equity product traders, but mutual funds and even fixed income and activist funds are getting in on the use of alternative funding data, added Furstenberg.

“The point is, it’s not relegated to any one bucket. It’s much more broad,” he said.

AI straight talk about marketing hype

AI can be considered shorthand for a few things, and how it’s understood and implemented makes the difference whether that’s “artificially inflated” or “accelerated innovations”, said Manoj Saxena, chairman at Cognitive Scale and former GM of IBM Watson, among other achievements.

The larger opportunity comes from augmented intelligence, rather than automation when it comes to creating new revenues, experiences and business models.

AI and blockchain, he added, is a “marriage made in heaven”: “If AI is a self-driving car, blockchain is the guardrails.”

Along with the attention on machine intelligence in capital markets have come a lot of promises: some are ambitious, some are speculative and some are just plain wrong, said Roland Fejfar, executive director at Morgan Stanley in the fintech group.

“We are still at the very early stages of applying this technology in the capital markets context and other sectors have been more advanced, and just applying the learnings and techniques of other sectors into the capital market sector doesn’t always work,” he noted.

Neither does the application of academic research, which often has to be anchored with financial industry knowledge, said panellists discussing whether machine learning is the new alpha generator.

Academia and practitioners are on the same side, right?

Although machine learning is evolving quite rapidly, as those techniques evolve, it’s difficult to convince fund managers to take them on, said Mark Salmon, a professor of finance and financial econometrics who teaches asset management at Cambridge, and is also a visiting professor in the computing science department at Imperial College.

“The techniques have to be used with great care, but there is a clear conservatism which I think is justified within the fund management industry,” he said.

“You need a track record and I don’t think machine learning has that yet, and you need interpretability, particularly from the sales side, so the managers are wary of adopting techniques which they can’t themselves explain to the people they are trying to sell strategies to,” he said. “We need to establish credibility.”

The two big problems of machine learning are well known: overfitting and a confusion between causality and correlation.

“Machine learning gives you a boost of efficiency on the research side that you didn’t have before, we can do things 100x faster than we could before,” said Morgan Slade, CEO of CloudQuant, a US-based investment manager that uses crowd researchers to develop investment strategies.

“But there’s a problem, it’s very easy to go to the dark side,” he said, referring to overfitting.

“Convincing people that you have those safeguards in place is a key part of adopting this.”

Andrej Rusakov, co-founder of Data Capital Management, a systematic hedge fund, said that “academia have a lot of clean, very robust datasets to experiment with”, but they also have “a bit of a lack of understanding of practical problems that practitioners face” in terms of execution, transaction costs and order fill rates, as examples.

On the other hand, practitioners feel not invested enough, or not at all, in the latest cutting-edge machine learning technologies, Rusakov added.

“These technologies are not proven yet: why take the risk? Why dive deep into something we don’t know will pan out, and something you don’t really understand the logic and mechanics and mathematics behind,” he said.

What the future looks like for AI in trading

Data Capital Management is based in New York and specialized in machine learning and “novel” data to generate forecasts of short- to medium-term movements in securities traded in the US.

Speaking to MarketBrains, Rusakov said that DCM’s vision is simple: the more data you have about capital markets and securiites, the better decisions get made.

“There is a tectonic shift that is happening in the industry: the amount of asset price relevant data is growing exponentially and humans are not able to cope with the amount of that data being generated,” he said.

Not bringing data in, and not taking it into your decision-making process is a “fatal mistake”, he added.

“There is no other way in the longer term than to build systems which allow you to bring all the datasets in and then process them in real time and make systematic decisions in real time without human bias,” he said.

“That’s really where I think the big shift in the way people manage money is happening.”

Even while the panellists were speaking, the Wall Street Journal sent a headline that PIMCO’s next bond king could be a robot in Texas.

What does that kind of future vision look like for Rusakov?

His 10-year prediction is that all the lower latency, intraday, and day trading will be dominated by machines and humans will not play a role.

Panel from L to R: Roland Fejfar, Morgan Stanley; Andrej Rusakov, DCM; Morgan Slade, CloudQuant; Mark Salmon, Cambridge University

And people who are packaging beta rather than delivering true alpha will also be gone because of the advance of the robo-advisors and the computers that are delivering the same beta exposure but at a much lower rate.

“If a hedge fund is really taking beta and not alpha, then you are going to be gone, because you are not going to be charge 2/20 if someone is going to be charging 20bps for the same job,” he said.

As a clarification, he explained that alpha meant excess returns to the market, and beta as he is using it refers to the equities markets, and also different style factors such as growth and value, for examples.

“Investors who are delivering true alpha, doing some specialized work for example that are trading on special events that are critical in terms of understanding fine print in the documents, those guys will stay,” said Rusakov.

So too will people who project tectonic shifts in industries and sub-industries, he added.

“Those tend to pan out over multiple years rather than a couple of days, and machines will not be able to compete with them in the foreseeable future, but everybody who is just trying to draw conclusions from a lot of data for short periods of time will be gone.”

Share.

About Author

Leave A Reply