The predictable ape: Generating alpha from human behaviour

In the funds it runs for Curate Investments, Robeco uses rigorously tested quantitative methods to extract returns from predictable patterns of behaviour.

Robeco’s quantitative investment team has a central view on markets and market behaviour: it is possible to identify patterns in the way prices move because, ultimately, those movements are a predictable result of human emotion.

“There are financial patterns in markets everywhere because human behaviour has not changed in thousands of years,” says Robeco’s head of quant client portfolio management, Jan de Koning. “Ninety percent of the decisions we make as humans are heavily influenced by emotions. And those emotion-based decisions are also driving markets.”

The firm believes it is possible to capture these patterns through the use of data and quantitative techniques. This is what it does in managing the Curate Global Sustainable Equity and Curate Global Emerging Markets Equity Funds.

“We started doing quantitative research over 30 years ago,” de Koning says. “We are a fundamental manager by origin, and we started to investigate whether we could improve investment decision making using data.”

For nearly 10 years, Robeco used this research to enhance its bottom-up process. But in 2004 the firm launched its first quant-driven portfolio, and it is now the largest quant manager in Europe.

“We think enhancing indexing is a smarter approach to passive investing,” de Koning says. “Like passive investing, it gives you a liquid, diversified portfolio at low cost, but instead of just replicating an index we use the knowledge available on market behaviour to identify patterns and signals.

“We take all the constituents of an index, then assess how each of them are positioned for future alpha. We then build a portfolio by tilting towards those that are positioned for alpha and away from those that aren’t. If you do that every month, those decisions compound.”

Market signals

The signals Robeco looks for fall into four categories: value, quality, momentum and short-term signals, with the first two being the most important.

“Value is about whether a stock is trading below intrinsic value,” de Koning explains. “We have a number of signals to determine that.

“We also love quality – strong sustainable earnings power. This is mainly about how good or conservative management is, and sustainability. For example, we look at how happy employees are. Very happy employees lead to very high sales ratios.

“We augment these models with momentum as a catalyst to unlock intrinsic value. And we look at short term signals coming from data sets that are relatively new.”

In total, Robeco uses over 50 variables in selecting stocks. Each of which has been thoroughly researched and tested.

“The beauty of being part of a fundamental house is that we don’t just have people torturing data all day long,” de Koning says. “Our research starts with conversations we have with our fundamental colleagues. We listen to how they look at certain sectors and what they look for. We then come up with ideas and test them on big data sets.”

Significantly, around 70% of the ideas that Robeco sets out to test, are ultimately discarded.

“It’s far from every idea will be successful,” de Koning (pictured below) explains. “In fact, we have a big graveyard.

“Even if the rationale for a signal is good, sometimes it just doesn’t stick. Other ideas we might find only work very well in certain markets, with low coverage in other markets. Those we might put in the fridge to look at again later.”

He adds that Robeco can spend 2,000 hours testing and refining a research idea. An example is how they looked at the question of whether you should buy the dip.

“If a stock that has been trading up suddenly falls, academic research and our own research confirms that 60% to 70% of the time it’s a good idea to buy the dip, because usually the long-term trend continues,” de Koning notes. “But we had a researcher who asked whether we shouldn’t approach this a bit more intelligently.”

The question put forward, essentially, was whether you should buy every dip.

“We said if there’s good news while the stock is trending down, it should bounce back pretty well. If there’s bad news, then we should stay away. But what if there’s no news? That’s what we wanted the data to reveal.”

To test this, however, required developing a dedicated tool.

“Journalists don’t label news good or bad,” de Koning explains. “You have to do that yourself. That’s where data analysis comes in. We developed our own large language model to reads the news and give it a score.

“And, interestingly enough, what we found is that if there’s good news, then buy the dip, and if there’s bad news don’t. But what we didn’t expect was that if there is no news, then you should really buy it.

“That’s because if there’s no news and a stock trends down, there may be a forced seller. Someone could be liquidating a position, and so once they are done selling, the natural buying pressure will come back again. We see this a lot in the mid- and small-cap space.”

This also illustrates how Robeco closely assesses every idea and whether it makes economic sense.

“It’s not a fully automated approach,” de Koning adds. “We believe in using human judgement with machines, because at the end of the day you can have a lot of data and insights, but human beings need to judge whether a certain pattern should be integrated or not.”