AustralianSuper first dipped its toe into the machine learning pond two years ago, trying to codify a strategy that echoed the internal team’s investment philosophy. It takes a long-term view with an emphasis on quality-style companies.
But where it differs from the fundamental approach taken by its analysts is that it looks for situations where history rhymes. Can we find companies in today’s markets that have characteristics which are similar to previous success stories?
“We start off with a premise that we are looking for stocks that are high quality, that are large caps, typically they are the number one or two in their markets, they have good returns on capital and they’ve got a business moat, a highly defensible business model,” Innes McKeand, Head of Equities at AustralianSuper says in an interview with [i3] Insights.
“So [we start with] common sense characteristics like this and then we translate that into a factor space. We are looking for 25 different dimensions, broadly captured by those headline issues and we are trying to find the nearest statistical neighbour in the current market place, using prior examples.
“It comes up with a very concentrated portfolio of 30 to 35 names,” he says.
But McKeand doesn’t let it just run by itself. After the algorithm has finished its job, the team sifts through the results and looks at whether there are any companies that simply don’t make sense, in other words, they look for false positives.
I know the kind of businesses it should find, whereas a complete black box is very difficult to penetrate
“We are always kicking the tyres; we are not leaving it as a black box,” McKeand says. “The machine generates what it generates and then we look through it and typically when we rebalance, we knock out two or three names, because they don’t fit with the common sense criteria that we set off with.
“I know the kind of businesses it should find, whereas a complete black box is very difficult to penetrate. You don’t quite know what you are getting. I quite like this model, because it delivers something that you can understand. There is smart science behind it, but it is grounded in common sense,” he says.
The strategy started small, but has been performing so well that is now almost a core strategy. McKeand is hesitant to say exactly how much they run in the strategy, but is comfortable disclosing that it is somewhere between $5 – 10 billion.
“It was the best performing portfolio in the equity construct last year,” he says. “That won’t always be the case. It was because the market circumstances aligned with the strengths of the model, but those particular characteristics that we were looking for paid off big time last year,” he says.
The strategy doesn’t adapt to different market circumstances, instead it is about having a more sophisticated way to find the type of companies the team is comfortable with as long-term investments.
“It is using structured data. Essentially, it uses historical examples of stocks with the right characteristics,” McKeand says. “It just simply says: ‘In the market right now, can I find something that looks a lot like previous examples?’
“For example, we use the HHI (Herfindahl-Hirschman Index), which is a measure of how concentrated the market is that you are looking into.
“ A classic example here is Nestle. Is Nestle the number one or two in the chocolate markets? You bet it is. That will show through an HHI index, but is not necessarily something that you can get through Bloomberg, nor is it web scraping, or some other form of new data.
“So we use already structured data; we are not using data lakes looking for little nuggets,” he says.
We are running two strategies; I don’t want to run that many more
More recently, AustralianSuper has started another quantitative strategy, which is more of a traditional factor model, targeting value-style companies.
“That has been running since the start of this calendar year, so it is still early days. We are looking at things to tune up that model,” McKeand says.
He doesn’t expect to add any more quantitative strategies to the lineup. “We are running two strategies; I don’t want to run that many more,” he says.
Yet, the need for people with deep knowledge of quantitative techniques is likely to remain high, McKeand argues.
“The need for quantitative skills is definitely going to increase for us from here. Quant skills can even be used in fundamental teams to help improve efficiency and insights, and help to make the job easier,” he says.
“So we will definitely have more resources and pay more attention to building up our quant team and processes.
“And that involves things like being able to take more data, clean it, process it efficiently, build efficient portfolios, all those steps in a quant process we will be putting more resources towards, because we know we are going to put more money in those strategies.
“But will we run many more of them? I think we want to keep everything as simple as possible,” he says.