Real-Time AI-Powered GDP Tracking: Lessons from Eight Years of GDPLive – Christoph Schumacher

Sprecher

Christoph Schumacher (Massey University, NZ)

Abstract

Official GDP statistics are released with a lag of up to six months at the national level and considerably longer at the regional level, leaving policymakers and businesses to make decisions on the basis of outdated information. In 2018 we introduced GDPLive, a live GDP nowcasting system for New Zealand that combines daily consumption, freight and macroeconomic data with an ensemble of machine learning models to estimate economic activity at the national, regional and industry-sector level. This paper updates that original working paper with evidence from eight years of continuous, public operation. Over this period GDPLive has been extended with a companion inflation and Official Cash Rate tracker, has absorbed several new data partners, and has had its predictions tested against a full business cycle, including the COVID-19 shock and the tightening cycle that followed. We also situate GDPLive within a considerably larger academic literature on machine-learning nowcasting than existed in 2018, and against a new institutional entrant into New Zealand's real-time GDP space: the Reserve Bank of New Zealand's own weekly nowcast, Kiwi-GDP, launched in 2025. We report on where GDPLive's models have generalised well and where they have not, and set out a revised research agenda for the next phase of the project. 

Autoren

Christoph Schumacher (Massey University, NZ) – Teo Susnjak (Massey University, NZ) 

Raum

Medienraum (A-211)