AI to deliver realtime GDP data
As we wait for Gross Domestic Product data to be released on Thursday, Massey University has launched a high tech, real time GDP tracker which could revolutionise the way we keep track of the nation’s economic progress.
The GDPLive project aims to create a real time model of GDP using up-to-date data like PayMark’s electronic card spending figures.
This week’s GDP data will be for the third quarter to September 30, meaning some of the data it is based on will be almost six months out of date.
This project was a bid to use the latest technology and the growing volume of digital data that organisations collect to better model GDP in real time, says Christoph Schumacher, professor of economics and innovation at the Massey Business School.
It uses complex selflearning (artificial intelligence) to constantly improve its modelling.
“If we know how much money exchanged hands yesterday using cards, then that is a pretty good indication of economic activity because GDP measures market based transactions of how many products are sold,” says Schumacher.
The project has a partnership with Port Connect to provide data from Ports of Auckland and Tauranga to monitor movement of goods.
KiwiRail offered freight data and, via the Interislander Ferry, real time tourism data.
The project also used publicly-available data from government sources like Stetson, The Reserve Bank, Immigration NZ, NZ Transport Agency and the Ministry of Business Innovation and Employment as well as traffic data from the Ministry of Transport, all the macroeconomic indicators from Stats NZ and the Reserve Bank.
The project was developed by the Knowledge Exchange Hub, a research hub at Massey University. It is headed by Schumacher and Dr Teo Susnjak, a computer scientist with Massey University’s Institute of Natural and Mathematical Sciences.
Schumacher says the key was to build a programme that could process the up-to-date data and continually benchmark it against historic data.
“So it keeps learning in little steps and when we expose it to the new data, with what it has learned from the past it then makes a prediction.”
The project had been underway for three years and had proved highly accurate and inside expected margin of error, Schumacher says.