Do the GDP overestimation claims of India’s former CEA stand up to statistical scrutiny? Look closely

His central exhibit was a chart of 17 economic indicators, 11 of which had turned negatively correlated with GDP after 2011. The starkest case was the Index of Industrial Production (IIP) for manufacturing, at minus 0.78.

Extend the data by seven years using the same method. IIP-manufacturing now correlates with GDP at plus 0.92.

The vanishing indicators: Of the 11 indicators flagged by the 2019 paper as negatively correlated, 10 have not merely turned positive, but strongly positive. The flagship of the case has reversed.

In March 2026, Subramanian, Abhishek Anand and Josh Felman published a new paper under the Peterson Institute’s banner. It claims Indian GDP has been overestimated by 1.5 to 1.9 percentage points a year.

The conclusion is a close cousin of the 2019 claim; almost none of the evidence is.

In his 2019 acknowledgements, Subramanian thanked Felman for “first alerting me to the issues, and helping at every stage.” Two of the three 2026 authors have been behind this research programme since before 2019. The author list has grown; the evidence, not so much.

‘Inconvenient’ indicators, 13 of the 17, have been dropped. The 2026 paper does not defend the 2019 case by testing its indicator set on extended data. It replaces it.

Put less charitably, the 2019 result was window-specific. It hinged on six years that happened to see structural changes in the Indian economy. After that window, the ‘breakdown’ reverses. Tractor production is the only holdout and it tracks monsoon rains more than aggregate activity. The 2026 paper does not mention any of this.

The deflator argument: One new 2026 exhibit is a wedge between corporate sales growth and gross value-added (GVA) growth, presented as independent corroboration. Corporate sales are deflated using the core Consumer Price Index (CPI). GVA, on the other side, is deflated substantially using the WPI.

When the CPI and WPI diverge, as they did sharply after 2014, the gap between the two deflated series widens automatically. The wedge the paper discovers correlates with the CPI-WPI gap at 0.81, which the authors describe as moving in the “expected direction.”

But the same correlation would arise in any economy that deflates two series with diverging indices, regardless of measurement quality. The test cannot distinguish their hypothesis from basic arithmetic. Treating the wedge as “independent evidence” is circular.

The paper further proposes deflating construction with the food-heavy general CPI. What does construction have to do with food?

A survey comparison that isn’t one: The informal-sector growth number, worth 0.4 to 0.8 points of the overall overestimation claim, rests on a 3.3 percentage point gap between two surveys: the National Sample Survey 73rd Round (2015-16) and Annual Survey of Unincorporated Enterprises (2023-24). These two surveys are not the same.

The instrument changed and so did the sampling frame and coverage. And there is a gap of nearly a decade between the readings, during which the Indian economy changed structurally. A compound growth rate drawn across that decade (which included covid) between two mismatched surveys cannot produce a reliable estimate. The paper uses it anyway.

Other statistical problems: The 2026 paper treats its five headline correlation changes as qualitative evidence of a broken link between GDP and real activity. In an April 2026 note for the Economic Advisory Council to the Prime Minister, Virat Singh and Aditya Sinha applied the Fisher r-to-z test to those differences. None is significant at the 5% level, or at 10%.

In simple words, the changes cannot be taken as meaningful. ‘Noise’ could have produced the results. A single observation flips the GVA-exports correlation the paper leans on most; adding back two covid observations excluded by the paper changes the story further.

Several 2019 positions were reversed in 2026. On double deflation, the 2019 paper said its absence “immediately induces a bias” but the 2026 paper calls this absence “a relatively minor problem.”

In 2019, the authors looked at informal manufacturing, about 5% of GVA, and called it “not an explanation” for their results. In 2026 the entire unorganized economy, around 44% of GVA, is a primary explanation.

The 2019 paper excluded tax indicators on the ground that the tax-to-GDP relationship was “unreliable.” In 2026, tax is one of the headline indicators, with an appendix to explain the parts that contradict the thesis.

On GDP overestimation magnitude, the 2019 paper offered 2.5 points as its central estimate, with a 95% confidence interval of 1.5 to 3.5 points. The 2026 paper lands at 1.5 to 1.9 points, which is entirely within the lower half of its confidence interval, without this downshift being flagged.

An analysis of both papers reveals a pattern. Indicators that fail on new data readings have been replaced and statistics that are not statistically significant have been presented qualitatively. The conclusion has stayed within a narrow decimal band for seven years while almost every piece of backup evidence has been replaced.

This reads like a conclusion in search of support, not empirical research.

These are the author’s personal views.

The author is associate professor, finance, at S.P. Jain Institute of Management & Research.