A Machine Learning Approach to GDP Nowcasting — Reserve Bank of India's Hybrid ML and Time-Series Model

India's quarterly GDP estimates are released with a significant lag, which is why the Reserve Bank of India's Department of Economic and Policy Research (DEPR) has invested in nowcasting: an earlier Dynamic Factor Model built from 6, 9 and 12 high-frequency indicators (RBI Working Paper WPS(DEPR) 03/2020) already aimed to give an early, real-time read on the current quarter's growth.

Building on that base, RBI economists Saurabh Ghosh and Abhishek Ranjan tested a hybrid approach that adds machine-learning methods, financial-market data and an economic-uncertainty index to the conventional time-series nowcast (SSRN working paper, December 2021; peer-reviewed in the Bulletin of Monetary Economics and Banking, 2023).

The combined machine-learning-plus-time-series approach reported improved nowcast performance over the pure time-series baseline at short horizons, with accuracy declining, as expected, the further out the forecast horizon extends.

The published materials describe this improvement qualitatively rather than through a single, independently audited error-reduction figure, so the evidence here should be read as promising internal and peer-reviewed research rather than a fully externally validated production system.

Read the full analysis: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3980188