Forecasting elements that stand the test of time

00:08 Introduction
00:28 Rob, perhaps you could start by telling us a little more about your background?
01:27 What is the idea behind sustainability of real world forecasting techniques?
03:54 What other areas of research are you interested in?
05:26 What kind of challenges do you come across when you are trying to cater for so many different industries?
08:09 What are your thoughts on using alternatives to time series forecasts?
09:00 How often do you encounter problems that cannot be solved with time series forecasts?
11:29 From an academic point of view, why do you think there is not enough longevity in some of the pieces of software that people are creating?
20:21 How do you balance robustness and accuracy and the cost of the implementation?
25:26 Rob, what are you working on today that you think will be useful over the coming years?


Forecasting is an ancient practice that is constantly evolving. As such, many pieces of software fail to stand the test of time. For this episode of LokadTV, we have the pleasure of welcoming Rob Hyndman, whose implemented open-source software has been downloaded by millions of users. We discuss the sustainability of real-world forecasting techniques.

Rob is Professor of Statistics and Head of the Department of Econometrics and Business Statistics at Monash University, where he has been teaching for 26 years. From 2005 to 2018, he was also Editor in Chief of the International Journal of Forecasting, as well as a Director of the International Institute of Forecasters. Rob writes various research papers and has published three books on forecasting.

Most software naturally decays over time, often for various reasons. When it comes to scientific software it can often be considered as “throw away software”, as its sole purpose is to support a scientific paper.

While at Lokad, we focus on the supply chain side of things, Rob likes to forecast for multiple areas where large amounts of data are available: electricity consumption, mortality rates, or tourist numbers for example. Tourist numbers are obviously harder to predict in the current pandemic situation, so Rob is presently helping the Australian government to forecast Covid cases.

We talk more about alternatives to time series forecasting, as well as the problematics of feedback loops and the irreducible level of uncertainty that play a role in our forecasting techniques. In addition, we discuss the problems of purely academics inspired software. Academic software is often filled with many hidden flaws. Their methods may outperform on a benchmark, but when used in a real world implementation they are often numerically unstable, with ridiculously long compute times and are often fiendishly difficult to use. All this takes the focus away from the real issues at hand.

To conclude, we go into more depth about the delicate balance between robustness, accuracy and cost when it comes to forecasting technology. Rob also tells us about more of the software he’s working on today and how he plans to remain committed to the open-source world.