New Zealand Super is applying data science to improve productivity, investment return and risk mitigation at the fund. We speak with the fund’s Head of Data Analytics, Kathryn Kerner, about the implementation.
Institutional investors are acutely aware of the importance of data in today’s investment landscape. But how funds leverage the available data to them can make a big difference in their ability to compete with other asset owners.
Stephen Gilmore, Chief Investment Officer of New Zealand Superannuation Fund (NZ Super), realised his fund needed to build a more comprehensive data function that sat across the organisation and could help its staff make better investment decisions, achieve higher productivity and better mitigate risk.
To build this function, NZ Super hired Kathryn Kerner as Head of Data Analytics two years ago.
Kerner joined the fund from the Federal Reserve in the United States, where she helped the various quantitative teams achieve better productivity by automating and centralising the numerous models in use.
“Before the 2007-08 financial crisis, stress-testing and financial stability groups didn’t really exist at the Fed. They were innovations from the Great Financial Crisis,” she says.
“How these tests were implemented was through quant teams all over the US. So a quant team in San Francisco might have been the point people for securities. And then a quant team in New York were the point people for trading.
“But it was a system of models and that needed to be automated and centralised. I was part of that team [that did that]. We made it so that it could be more efficient and more push button-like.”
“We're thinking about disaggregated, flexible, automated solutions in the areas of backtesting, scenario analysis, portfolio management, assumption testing and P&L prediction
At NZ Super, Kerner has been applying her knowledge to not only make the fund more efficient, but also improve investment decision-making.
“We’re thinking about disaggregated, flexible, automated solutions in the areas of backtesting, scenario analysis, portfolio management, assumption testing and P&L prediction,” she says in an interview with [i3] Insights.
Kerner works closely with the strategic tilting team, led by Alex Bacchus. Strategic tilting is a contrarian strategy based on the belief in mean reversion of asset prices and risk premia. It is key to the success of the fund and since inception has added NZ$4.6 billion in investment returns.
Heavily quantitative, the strategic tilting team is a natural fit in discussing the application of data science to investment problems.
“My team’s competitive advantage is operationalising things, breaking really complex problems into building blocks of code so they can be reused,” Kerner says.
“Tilting’s competitive advantage is obviously taking risks and making money, so those two together are really interesting because we tend to approach problems differently.”
But despite the different approaches, the two teams usually end up with similar conclusions.
“We think about problems differently and we can debate it, but usually we come up with the same answer because in the end it’s all maths,” Kerner says.
The collaboration has resulted in the development of a series of models that together are referred to as the strategy testing tool. The purpose of the tool is to enable better investment decisions by improving the understanding of risk and return outcomes from current and potential investment choices.
“The strategy testing tool is about testing strategy outcomes and sensitivities through a lot of different scenarios, through time series and through historical, manipulated or simulated data,” Kerner says.
“Scenarios can be created by the user. You can alter time series and models and model parameters and tilting assumptions. So it’s really flexible. And then the end user can kind of specify what they want to run and then compare results between runs.”
The tool is being jointly developed by the two teams. Starting with the analysis of foreign exchange investment choices, it will be extended to cover more asset classes, strategy choices and other functionality.
“We are very into trying things out, experimentation, iteration, and then delivering in small pieces because we want to deliver quickly and we don’t want to not talk to people for two years before anything happens,” Kerner says.
Unstructured Data
Although work on the strategic testing tool continues, Kerner is also interested in exploring the ability to improve investment decision-making by analysing unstructured data that comes to the fund, including annual reports, research reports, call transcripts and news articles.
“This is actually about the unstructured data that we receive in our documents and we’re trying to do experiments with that to see if it will make us money, improve productivity and reduce risk,” she says.
“We often start with a question. How consistent are we in making decisions across investments? What is our view on opportunity attractiveness? And can we predict it into the future? That is also linked to access-point strategy.
“We’re going to use unstructured data to inform that and we have some predictive models that relate to that. It’s in its early stages, but there is a lot of interest in that at the fund right now.”
Building these models requires thousands of lines of code and Kerner points out it is important to establish best practice guidelines for how to write clear code and standardise it to avoid any legacy issues in the future.
“You can learn how to code, but not code in a readable way or in a best practice way. And so it’s really important that people follow those kinds of guidelines. This is not about disabling; it’s about enabling and actually giving people clarity that empowers them,” she says.
We often start with a question. How consistent are we in making decisions across investments? What is our view on opportunity attractiveness? And can we predict it into the future? We're going to use unstructured data to inform that and we have some predictive models that relate to that
Asked if she is horrified by the often elaborate prose ChatGPT spits out in response to requests for code, she says it can actually be quite helpful if used in the right way.
“I actually love [ChatGPT] because I think it’s a productivity booster. But then you can’t just use it blindly. You have to check it,” she says.
“You have to actually test your code and your model methodology because there are a lot of tools in a quant tool belt. Solving complex problems doesn’t always require a complex model.
“And at the end of the day, you can’t just build models to build models. It has to result in an actionable, optimised investment decision.”
Kathryn Kerner spoke at the [i3] Investment Strategy Forum, which was held on 9 &10 May 2024 at the RACV Resort in Torquay, Victoria.
__________
[i3] Insights is the official educational bulletin of the Investment Innovation Institute [i3]. It covers major trends and innovations in institutional investing, providing independent and thought-provoking content about pension funds, insurance companies and sovereign wealth funds across the globe.