Time Series Nowcasting of India’s GDP with Machine Learning

Abstract

The GDP forms an essential metric in assessing the state of the economy. However, the GDP figures are only available with a certain lag whereas economists need the data on a timely basis for accurate predictions of economic growth. Nowcasting helps in addressing this problem. This paper explores several machine learning (ML) algorithms in nowcasting the nominal quarterly GDP of India for the period 4Q2014 – 2Q2022. The algorithms are trained over a number of years using a wide range of high frequency macroeconomic and financial indicators and the results are then compared to the ones obtained using a traditional autoregressive model, Vector Autoregression (VAR). According to our results, Huber regression gave the least error i.e. 3.67 % while VAR gave an error of 15.89%. ML models outperformed VAR in terms of predictive accuracy while nowcasting India’s GDP. In this paper, analysis has been carried out on Python using the Pycaret library.