Why machine learning powered investment algorithms trump traditional methods

11 December 2017: Since news broke last week Tuesday of alleged accounting irregularities involving South Africa-based German listed retail giant Steinhoff, its share price has plummeted from R46.25 on Friday to between R6 and R7 today amidst a flurry of investor fear and panic selling. According to Stuart Reid, chief engineer at NMRQL Research (pronounced ‘Numerical’), the warning signs were there but ignored by many fund managers due to their inherent cognitive biases, highlighting the benefits of machine-learning powered investing.

“Most stock crashes are preceded by warning signals – but a fund manager’s inherent cognitive biases prevent them from seeing the wood for the trees,” says Reid.

This is precisely why the algorithms behind the NMRQL SCI Balanced Fund, South Africa’s first machine learning powered unit trust, administered by the Sanlam Collective Investments platform, were not invested in Steinhoff. “The majority of our algorithms stopped picking Steinhoff in 2016,” says Reid.

While he emphasises that the algorithms did not, and could not, know about fraud allegations, Reid explains that the algorithms assess a combination of structured data that includes market data, public company information, currencies, indices, economic data, risk metrics, and more. “The algorithms analyse financial statement line items alongside market data and aggregate them to make a prediction. Using this information, the algorithms actively try to predict what Steinhoff and other JSE listed shares will do – in the case of Steinhoff it predicted the stock would go down.”

However, according to Reid, the real question should be why fund managers failed to act despite the warning signs. “We are wired to want to minimize cognitive dissonance – this is the discomfort we feel when we disagree with the ‘status quo’. For example, many would feel uncomfortable not holding Steinhoff as it was the seventh biggest company in the FTSE/JSE Shareholders Weighted Index (SWIX) and present in many other large manager’s portfolio,” he says. Reid believes that this behaviour can result in suboptimal decisions which hurt investors.

“There were many warning signs about Steinhoff which fund managers did not heed as they appear to have been more concerned with upholding the status quo than maximizing shareholder value. This behaviour leads to groupthink and concentration risk,” says Reid. “And as the saying goes: if everybody’s thinking alike, somebody isn’t thinking.”

Reid explains that the benefit of machine learning is that it is free from cognitive biases and makes decisions objectively without any regard for what everybody else is doing.

He also points out that NMRQL Research are working on ways to further understand the algorithms picks so that they can pick up warning signs early on. He believes that all fund managers, including those who avoided the Steinhoff debacle, should use this opportunity to interrogate their investment processes from first principles

0 replies

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *