India’s GDP roller coaster

Subhash Garg, a former finance secretary and an intrepid economics commentator since his retirement, recently created a storm in the placid and congenial waters of national economic statistics. He called out the unexplained significant downward revision of the nominal GDP for Q1 of the previous year in the data released on August 31 for the provisional GDP estimate for April-June, FY27.

The previous year’s nominal GDP is the “base” for deriving the current year’s GDP growth — the number which the public focuses on as a broad measure of “well-being” at least for the top 40% of income earners invested in the stock market. Reducing the “base” statistically, ex-post facto a year later, pushes up the growth rate in the current year. Increasing the base year GDP would have the opposite effect. Done without substantive reasons, this is fiddling with national accounts and is unwise.

The matter was the unexplained, massive decrease in the previous year’s nominal GDP. Subhash Garg wrote on September 2 in The Quint that the nominal growth in Q1 FY27 was 10.3% (a satisfactory number). But, he said, had the base year nominal GDP not been statistically decreased from Rs 86.05 trillion, as assessed last year, to Rs 80 trillion, GDP growth would have been only 2.6% — lower than the ongoing inflation rate — and thereby negative in constant terms after excluding inflation from nominal GDP. The last fiscal (FY26) data was published under the earlier 2011-12 time series while a new GDP time series (2022-23) was applied from Q1 FY27 but extended backward.

This startling revelation created a furore. The Congress party was delighted at getting a chance to bait the ruling Bharatiya Janata Party. Economists of all stripes were united that Garg, a career bureaucrat from the IAS, had put his foot in his mouth by violating a basic rule in economics — never compare data across time series of GDP. The framework of weights across components of GDP gets altered in each series to better reflect the changes in the underlying economy; for example, the changing relative importance of services versus manufacturing or IT versus textiles.

Avoiding comparisons across GDP series is sensible. But Garg was pointing to nominal GDP, not constant GDP. Variations should be minor in nominal terms because this is the “visible” GDP which is measured. Constant GDP is less obvious and “manufactured”, being derived from nominal GDP after factoring out inflation to make data across time reflective of “value addition” as well as productivity of the economy over a period.

Garg’s point is simple and remains unexplained. A reduction in nominal GDP of about 7% in the previous year’s estimate is huge. It should have been investigated for systemic errors, not simply derided on first principles of statistics — that GDP estimates for the same year may differ depending on the time series used. Most of the reactions were “cookie-cutter”, based on the assumption that the statistical “system” could do no wrong.

This is worrying. Over the next decade — as AI becomes agentic — human collaboration would be critical to remain competitive and aware of why a number was spat out by a model. Was it hacked? Sloppy modelling? Or runaway machine intelligence?

Budget execution data for Q1 FY27 are positive for growth. Revenue collection was about 36% of the annual target versus just 31% last year in the same period. Consequently, growth supporting capital expenditure was 37% versus just 31% of the annual budget in the previous year. In comparison, revenue expenditure is muted at 32% this year versus 31% last year. Fiscal deficit is lower at 27% of the budgeted annual amount versus 30% last year, reserving fiscal fire power for future uncertainties. All of this is salutary for growth by the year-end.

The ministry of statistics and programme implementation, helmed by Saurabh Garg (not to be confused with Subhash Garg), has had its hands full. It has bravely and simultaneously implemented a new GDP series (2022-23) to replace the ageing 2011-12 series. It has also adopted the double deflator method for the first time to more forensically assess value added directly from the difference between input value captured from producer prices and the value of outputs captured from prices paid by consumers.

This enormous data set is being created for the first time. Sharing it openly — suitably anonymised — for discussion and comment within the fraternity of statisticians prior to release of the series might have been a better way to go about it. Data transparency internalises “informed” dissent beyond those directly responsible for the output. Walls around data can only externalise dissent and force “outsiders” and “insiders” into opposing camps.

Disclaimer: The views expressed are the author’s own and do not reflect the official policy or position of Financial Express.