The growing knowledge gap in quantitative investment strategies between technical experts and people without this background has provided fertile ground for intermediaries to step in and offer often distorted narratives about them, says quantitative finance expert Michael Kollo.

Speaking on Market Narratives, a podcast series hosted by Investment Magazine’s head of institutional content Alex Proimos, Kollo said as asset owners look to technology to scale and create efficiencies, most of the benefits have still been in back-end operations.

At the front end of forecasting markets and outcomes, Kollo said he was somewhat disappointed about the persistence of certain narratives in financial news, such as looking at quantitative tools as “black boxes”.

“‘Black box’ is a relative term, so what a black box is to you may not be a black box to me,” Kollo said. “When somebody says ‘it’s a black box’, it usually means they don’t fully understand the way information is being used by that model. The limitations, the strengths, the weaknesses of that model. What it can tell us about the world and what it can’t tell us about the world.”

Kollo has created models and led quantitative research teams at Blackrock, Fidelity and Axa Rosenberg in the UK, and established the quantitative team for industry superannuation fund HESTA. He hosts The Curious Quant podcast.

The amount of knowledge required for fund selectors doing due diligence on a factor or AI strategy is large and growing as data methodologies become more complex, Kollo said. Intermediaries are stepping in to explain these technical processes to non-technical people, but sometimes these intermediaries accidentally or nefariously imply false expectations about them.

To listen to the recorded interview with Michael Kollo on the Market Narratives podcast click above or find the series on Apple Podcasts, Google Podcasts or Spotify.

Kollo said there can also be a bias toward people who make broad generalisations, rather than toward people who take a fundamentally different approach of gathering evidence from a wide variety of sources and use careful methods of deduction to make reasonable conclusions – a phenomenon outlined in detail in the best-selling book Superforecasting: The Art and Science of Prediction by Philip E. Tetlock and Dan Gardner.

“If you have ever read it you get a sense that the people who rely on the latter way of thinking often are much harder to get sound-bytes out of, it’s much harder to put them in front of a TV camera because they appear highly uncertain, they are telling you about how uncertain the world is,” Kollo said. “They are telling you they have marginally better information than flipping a coin, and that’s not really attractive to hear.”

The asset management industry is laden with behavioural biases where highly regarded managers and asset owners may give a very different assessment on their probability of beating the market than what is borne out in statistics, he said.

Mathematics is the only common language for asset owners to describe their approach, Kollo continued, and there is room for a more systematic approach through the whole industry.

Proimos asked Kollo whether the widespread availability of data and processing power leads to “everyone flooding down the same path”.

Kollo said studies on the phenomena of crowding have shown it usually happens during a risk event, whether it be a risk-on or risk-off event, when many agents other than quantitative investors are flocking into one area of trade, for example buying certain defensive stocks.

“If you and I ran machine learning on the same data, the chances are, because of the huge array of hyperparameters and configurations and dials and knobs that we have to configure, we will not end up at same result, so therefore we will not end up at the same portfolio,” Kollo said. “So typically machine learning in its ideal phase does not lead to crowding, does not lead to this type of behaviour.

“But what does, however, is this idea that everyone has the same backtesting data, everybody uses the same methodology, everybody uses the same risk model… so, therefore, we all react the same way to whatever risk episode happens.”

Join the discussion