AI’s slow creep into asset management

How effective are artificial intelligence big data algorithms without human oversight?

Fund managers are increasingly using artificial intelligence (AI) as a tool in their investment strategies, but human involvement remains essential to counterbalance some of the algorithms’ flaws.

The rise of big data and the exponential growth in computer power in the past few years has allowed fund managers to analyse more data than ever. However, the underlying strategies investors use remain more or less the same.

Big data has enhanced the investment process, but it has yet to revolutionise it, admits Bradley Betts, one of the founders of Blackrock’s Artificial Intelligence Lab, founded last year in California’s Palo Alto.

“We collect as much data as possible,” says Betts, adding that the text on his favourite T-shirt says, ‘I never met a bit of data I don’t like’. “It pretty much sums up our attitude.”

But with so much data to process, the challenge facing Blackrock is how to effectively monitor the algorithms’ work. “Algorithms are much better at analysing data than humans,” Betts says. “But there are things algorithms are not good at, identifying regime changes, for example, such as negative yields.”

And they can’t self-reflect. “Algorithms often find correlations that don’t make sense, but they are unable to identify them,” Betts says. “Humans, on the other hand, are generally good at saying, ‘Stop, I don’t know what’s going on’.”

It may be many years before a robot investing autonomously without human interference is a viable proposition. For the time being, according to Betts, the best results are a fusion of man and machine.

Smart beta 2.0

The use of artificial intelligence in asset management is essentially an advanced form of factor investing, according to Ram Active Investments, a Geneva-based investment boutique.

Ram AI was one of the first European asset managers to make AI an integral part of its investment process across all funds.

Emmanuel Hauptmann, a founding partner at Ram, says the group began building a deep-learning platform about three years ago.

“Instead of looking at traditional factors – such as value, low size, low volatility, high yield, quality and momentum – we can now exploit hundreds of different inefficiencies at once. We can analyse stocks using data sources that were previously not accessible,” he says.

Ram has, for example, added additional risk premia factors to momentum factor investing in order to better capitalise on a market trend.

“We use data from stock-lending databases to get information about the positioning of hedge funds and their short interests,” Hauptmann explains. “Companies that are shorted a lot have lower return expectations and will be more volatile going forward.”

The group’s deep-learning approach also takes into account other sources of information, including checking insider transactions in a stock to gauge shifts in sentiment among a company’s management.

According to Tjeerd van Cappelle, head of automated intelligence equity at NN Investment Partners, the biggest advantage AI offers fund managers is the ability to conduct detailed market simulations.

“Traditionally, investors conduct historical tests to identify risk premia. AI provides us with the opportunity to predict investor behaviour by conducting scenario analysis on the back of large amounts of data,” Van Cappelle says.

“You can compare this to the way a chess computer works. In the end being a successful investor comes down to better anticipating market behaviour than the average market participant.”

Does AI deliver alpha?

AI is beginning to open huge vistas of possibility for investors, but the million-euro question is, of course, does AI deliver alpha?

The answer to this all-important question remains shrouded in mystery. Blackrock’s Betts says his AI lab derives alpha, but he declines to give any details.

“Our systematic active equity unit is responsible for 30-80% of all our funds’ portfolios. But because we derive alpha, I can’t be too specific about it,” he says. Ram’s Hauptmann is equally reluctant to provide too many details. “We introduced our deep-learning strategy in emerging markets about a year ago and since then we have outperformed the benchmark,” he says.

Developed markets, he admits, have not gone as well. “Our bias to short highly valued stocks has cost us performance, as expensive companies have been outperforming cheap ones in the current rally,” he says.

Fund selector interest

An absence of provable track records and the complex nature of AI-based investment has meant few fund selectors so far have invested in such strategies.

The few that have invested are proceeding with caution.

“I want to be able to explain to clients why a certain strategy makes sense,” says Freddy van Mulligen, head of external manager selection at Achmea Investment Management.

“If a manager uses hundreds of different factors in their investment process, that becomes a challenge.

“Therefore, we are only testing the water. We just work with managers who use big data as a small input in their investment process, such as Axa Rosenberg. They use the price pattern of a stock and trends in the bid-and ask-spreads to determine stock risk.”

The AI revolution has begun but understanding the full implications for the industry remains years away.

For more insight on continental European investment, please visit


Leave a Reply