The combination of humans and machines is better than either in isolation, and the hybrid approach is crucial for success in investment management research, says Nick Thomas-Peter, senior vice-president at Winton Group in London.

Speaking at the Portfolio Construction Forum in Sydney, Thomas-Peter used the example of the US Postal Service to illustrate how human skill and judgement are still required to navigate intrinsically human issues.

As the USPS has expanded via technology, machines have often got it wrong because the input – the address on the envelope – was written by humans. Thomas-Peter quoted a postal worker he spoke to, who said: “It used to be that we would get letters that were somewhat legible…Now we get letter and handwriting that is more awful than you can imagine.”

Thomas-Peter said: “While the introduction of machines has allowed the [USPS] to reach a scale, speed and cost that could never have been achieved with humans alone, they weren’t able to do the whole job.”

The conversation is often pitched as “humans v machines” but for investment managers looking to use technology to analyse markets, the solution is much more nuanced, he said.

“It should be about humans enabled by machines,” he asserted. “We need the combination of the two.”

Thomas-Peter told how Winton Group looked at what would happen if people “handed over the keys to the machines”. Using the German DAX equity index as a proxy, the group tried to build a “predictor machine” based on more than 11,000 available predictors. Staff then “let the machines decide”.

Thomas-Peter called the results “slightly disappointing”, even though the machine sometimes got it right.

The issue was that in data mining, people can often be misled when things go right. “There is a meaningful opportunity that some things will work [by accident], which will give the appearance of an accurate prediction,” he said.

There were also issues with the human biases that crept into testing, he reckoned.

“The predictors that I’m likely to pick are going to be the ones I’ve heard about or read about in the news,” Thomas-Peter explained. “The media are very good at coming up with stories about why putting something alongside something else shows a causal relationship.”

It’s not enough to just “sprinkle in, in some naïve way, a little human intellect”, he said. “What you have to do is have a process that guards against the sorts of biases that humans bring.”

Thomas-Peter said a “best-practice guide to performing research” includes three components: idea generation, idea testing, and implementation. The testing is where our biases need to be addressed, he explained.

He said testing must be done with “the highest integrity” and that testers should not seek to push a “null result over the line of significance so that it becomes positive. That’s an extremely pernicious activity that can degrade an institution’s ability to break new ground on ideas.”

If the investment community doesn’t get the testing phase of machine technology right, there can be major issues down the line.

“What we know is that markets are changing in response to the environment and the psychology of the people operating them, which makes things more difficult when you’re using machines,” he said. “There’s a sort of buyer beware [involved]; you have to combine humans with machines in an intelligent way. And you require both experience and a rigorous, methodical approach. It’s only by doing this that we as an investment community will be able to capture and profit from the opportunities that the future holds.”

Tahn Sharpe is a Sydney-based financial services journalist with a background in financial planning. He writes on advice, superannuation, investment, banking and insurance issues, is a certified SMSF Adviser and holds an Advanced Diploma of Financial Planning.
Leave a comment