1Manager, Department of Payment and Settlement Systems
2MBA Student, Narsee Monjee Institute of Management Studies and Summer Intern at Reserve Bank of India
3Deputy General Manager, Department of Payment and Settlement Systems
*Corresponding author email id: rajput.naveen07@gmail.com
4The views expressed in the paper are those of the authors and do not reflect the views of the Reserve Bank of India
The recent worldwide development and widespread use of digital payment systems has provided an opportunity to explore new sources of data for the monitoring of macroeconomic activity. In India, digital payment systems data are available at a higher frequency (daily). In this paper, therefore, we analyse the usefulness of higher frequency data collected from different payment systems for nowcasting quarterly macroeconomic variables. We take advantage of the availability of such higher frequency data firstly, to forecast higher frequency payment system indicators, and use the forecasts to nowcast lower frequency economic variables such as Gross Domestic Product (GDP) and Private Final Consumption Expenditure (PFCE). We use new state of the art deep learning models like Prophet-with-XGBOOST-errors, Feed forward autoregressive neural networks and their weighted ensembles to forecast payment system data. We forecast payment systems like Real Time Gross Settlement (RTGS) which is typically a large value payment system, retail payment systems like Unified Payments Interface (UPI), National Electronic Funds Transfer (NEFT), Credit and Debit cards, Immediate Payment Service (IMPS), etc. through these methods. We use Autoregressive Integrated Moving Average (ARIMA) and Ordinary Least Squares (OLS) method for nowcasting GDP and PFCE.4
GDP, PFCE, Payment systems in India, Forecasting, Nowcasting, Machine (deep) learning, ANN, Supervised learning
