New UNCTAD website monitors global trade and GDP in real time

04 July 2022

The site improves the utility and accessibility of UNCTAD’s nowcasts for the international community.

© Shutterstock/Avigator Fortuner

UNCTAD launched on 1 July a new interactive website that provides real-time estimates on global trade and GDP, showing the impacts of the multiple global economic shocks.

The website provides data-based estimates, known as nowcasts, which shed light on the continuing economic pressures of the war in Ukraine and the persistent strong inflationary concerns. They offer timely information to help guide policy responses.

UNCTAD has long produced nowcasts on the global quarterly growth of merchandise exports. These have provided a timely picture of the development of global trade since the start of the COVID-19 pandemic.

But their dissemination via end-of-quarter bulletins was limiting.

Latest data available

To generate the real-time estimates, the latest data available and data revisions are fed into models on a weekly basis to update and revise the nowcasts and provide insight on current economic and trade conditions well before final figures are published with several months’ delay.

The estimates are drawn from models for UNCTAD’s three trade series covering global merchandise exports expressed in values and volumes and services exports. And for the first time – the estimates include global annual GDP growth.

Besides the estimates, the website displays the nowcasts’ development over time, showing how data releases have shaped them as the economic situation has evolved.

The models behind the nowcasts are known as long short-term memory artificial neural networks (LSTM), which outperform UNCTAD’s previous dynamic factor model approach.

The methodology and information on the accompanying open-source, multi-programming language library are available in an UNCTAD research paper.

A second UNCTAD research paper details the methodology’s performance during the pandemic, while a third research paper compares its performance to other common nowcasting and machine learning methodologies.