Satellite lights, Google searches cut state growth estimation errors by up to 62%: RBI study

A combination of satellite-based night-time lights and Google search data can improve estimates of state-level economic growth, reducing forecasting errors by up to 62%, according to a study published in the Reserve Bank of India’s October Bulletin.

The study, Nowcasting State-Level Economic Activity in India with Night-Time Light and Google Trends, developed a quarterly framework to track economic activity in Gujarat, Maharashtra and Karnataka using indicators that are available more frequently than official state economic data.

Gross State Value Added (GSVA) figures are typically released annually with a lag of 12-18 months, making it difficult to assess how state economies are performing in real time.


The researchers said the proposed framework could supplement official statistics by providing more timely estimates.

The model combined satellite imagery measuring night-time illumination with Google Trends search indices across 25 categories, including jobs and education, finance, shopping, real estate and automobiles.

It also incorporated past GSVA growth to estimate current economic activity.

The researchers compared the model’s performance with a benchmark based on past GSVA growth alone.

Adding both night-time lights and Google search data reduced the root mean squared error (RMSE), a measure of estimation error, by about 38% for Maharashtra, 54% for Karnataka and 62% for Gujarat. The combined indicators also improved estimates compared with using satellite data alone.

The study found a strong correlation between year-on-year growth in night-time lights and overall GSVA growth. The correlation was highest in Maharashtra at 0.90, followed by Karnataka at 0.79 and Gujarat at 0.66.

The relationship remained strong and statistically significant even after the researchers excluded quarters affected by the Covid-19 pandemic.

Night-time light data also provided predictive information about subsequent GSVA growth, while Google Trends data appeared to contribute mainly through contemporaneous signals.

The researchers cautioned that the framework had been tested only in three industrial states. They said it was intended to complement, rather than replace, official economic statistics.