Template-type: ReDIF-Paper 1.0
Author-Name: Laura Felber
Author-Name-First: Laura
Author-Name-Last: Felber
Author-Person: pfe607
Author-Name: Dr. Simon Beyeler
Author-Name-First: Simon
Author-Name-Last: Beyeler
Author-Person: pbe1152
Title: Nowcasting economic activity using transaction payments data
Abstract: In this paper, we assess the value of high-frequency payments data for nowcasting economic activity. Focusing on Switzerland, we predict real GDP based on an unprecedented 'complete' set of transaction payments data: a combination of real-time gross settlement payment system data as well as debit and credit card data. Following a strongly data-driven machine learning approach, we find payments data to bear an accurate and timely signal about economic activity. When we assess the performance of the models by the initially published GDP numbers (pseudo real-time evaluation), we find a state-dependent value of the data: the payment models slightly outperform the benchmark models in times of crisis but are clearly inferior in 'normal' times. However, when we assess the performance of the models by revised and more final GDP numbers, we find payments data to be unconditionally valuable: the payment models outperform the benchmark models by up to 11% in times of crisis and by up to 12% in 'normal' times. We thus conclude that models based on payments data should become an integral part of policymakers' decision-making.

Length: 48 pages
Creation-Date: 2023
Contact-Email: forschung@snb.ch
File-URL: https://www.snb.ch/en/publications/research/working-papers/2023/working_paper_2023_01
File-Format: text/html
Number: 2023-01
Classification-JEL: C52, C53, C55, E37
Keywords: Nowcasting, GDP, Machine learning, Payments data, COVID-19
Handle: RePEc:snb:snbwpa:2023-01