How to detect recent structural breaks

What is it about?

It is common to collect data on multiple related quantities over time. This paper looks at how to detect recent abrupt changes (also known as structural breaks) in these time-series, based on the idea that often the same underlying cause may lead to changes in some, but not all, of the time-series at the same time. By focussing just on detecting the most recent change in each sequence, we simplify the problem of searching over all possibilities of which series change at the same time: leading to a computationally fast and accurate method.

Featured Image

Why is it important?

Short-term forecasting is important in many applications, but standard forecasting methods can become unreliable if there is a structural change in the process being forecasted. The method presented in this paper provides an efficient method for detecting recent structural changes in a set of related time-series, so that forecasts can be made just using data after these changes. We show the benefits of the method on forecasting events in a telecommunication network.