Making sure the forecast fits

28 June 2018
Shanika Wickramasuriya, Lecturer, Department of Statistics
Department of Statistics Lecturer, Shanika Wickramasuriya

Knowing what the future holds is straightforward for a business that sells one kind of product in one region – say, dog food. But what if your business is more complicated – for example, you’re a clothing manufacturer supplying a range of men’s, women’s and children’s items across various seasons and regions?  

Shanika Wickramasuriya, who started with the Department of Statistics as a lecturer in February, can tell you how they do it. Her work in time-series analysis, forecasting and computational statistics helps companies look at their past data to predict their future needs.

In particular, Shanika works on hierarchical forecasts, which are multi-layered forecasts within an overall forecast. As an example, a fashion retailer might want forecasts for male and female clothing, broken down into forecasts for individual clothing items, and within this category, forecasts for individual sizes and colours. The retailer wants a hierarchy of forecasts with the overall forecast at the top.

But using time-series data to predict future needs isn’t as straightforward as it might appear – there could be millions of time series, and they won’t necessarily mesh together well. Helping to refine the process was the focus of Shanika’s PhD research at Monash University in Australia, which she completed last year.

At Monash, Shanika worked with her PhD supervisors Professor Rob Hyndman and Associate Professor George Athanasopoulos, who in 2011 developed an approach where each series in a hierarchy could be forecast and then reconciled to ensure they give the best overall forecast.

Shanika’s role as their doctoral student was to help develop a mathematical algorithm that optimally reconciled the individual forecasts and write an open-source R software package named hts that underpins it.

So far, their package has been downloaded more than 41,000 times. Among the companies benefiting from it are Walmart, Nestle, SAP, Huawei and Bank of New York Mellon, Tourism Australia – and Monash itself, which used it to forecast student enrolments.

After graduating, Shanika worked for Monash for six months before accepting her role as lecturer in the University of Auckland's Department of Statistics. The department’s high ranking was a major part of her decision, she says, as well as the fact that “the pioneers of R statistical software are from the same department”.

In Semester One, Shanika taught the course Professional Computing Skills for Statisticians (STATS 779), and in Semester two teaches Statistical Computing  (STATS 380) and Time Series (STATS 726).

Asked to choose a role model in statistics, Shanika doesn’t hesitate to select Professor Rob Hyndman, her primary PhD supervisor.

“I always use this phrase to describe him: The best teachers are those who show you where to look but don’t tell you what to see.”