India’s New GDP Series Is a Black Box Without the Underlying Data

India’s latest GDP estimates have opened a debate that goes beyond whether one quarter’s growth number looks too strong. At issue is the statistical architecture behind the estimate and whether the official system has disclosed enough for economists, companies and citizens to understand and replicate it.

The sharpest question concerns the deflator. Nominal GDP in April–June grew 10.3%, while real GDP rose 7.8%, implying an inflation adjustment of about 2.5%. That appears low against broader inflation measures. The new series uses double deflation, applying separate price indices to inputs and outputs rather than one output price to both. Internationally, this is regarded as a more rigorous approach. It is also enormously demanding: every sector requires reliable, granular prices for the goods and services it consumes as well as produces.

That distinction is central to Pronab Sen’s argument. India’s first Chief Statistician does not reject the method. He questions whether the underlying price database is sufficiently rich and, if it is, why it has not been placed in the public domain. Without such disclosure, he says, users are being asked to accept a black box. For users of economic data, this is not an academic dispute. The figures shape policy debates, investment decisions and assessments of living standards.

The controversy extends beyond deflation. A much larger-than-usual reduction in the level of current-price GDP after the base-year revision raises questions about past overestimation and future revisions. The composition of growth also matters. Fast-growing services coexist with a diminished manufacturing share, while quarterly estimates for non-corporate and unorganised activity still lean heavily on corporate indicators.

In this conversation, Sen tells Rajesh Mahapatra why sound methodology cannot substitute for visible data, why trust in official statistics ultimately turns on transparency, and why India needs to restore the institutional role of the National Statistical Commission.

The interview has been edited for clarity and length.

The Deflator

Q: The most contentious issue concerns the GDP deflator. Nominal GDP grew 10.3% in April–June, while real GDP grew 7.8%, implying a deflator of about 2.5%. That appears low compared with the broader inflation measures seen during the quarter. How should we understand it?

A: I wish I could decode it completely. One of the biggest changes made in the new GDP series is in the method of deflation. It is a substantial change.

Earlier, we had two broad sets of price data, the consumer price index and the wholesale price index. Depending on the sector, we used the appropriate prices. Agricultural prices, for instance, would be used to deflate the value of agricultural output.

The price of the final output was also used to deflate value added. In effect, the cost of the inputs was converted into units of the final product. Suppose a farmer produced one tonne of wheat and used inputs whose cost was equivalent to half a tonne of wheat. The net value added would be treated as half a tonne of wheat. This was done across sectors and then added up.

What is being done now is different. Separate price indices are being used for the output and the inputs. Conceptually, that is the recommended method. But it is extremely data-intensive.

Take wheat. Earlier, you primarily needed the price of wheat. Under double deflation, you need the price of wheat as well as the prices of seeds, fertiliser, pesticides and other inputs. For something such as an automobile, you could be dealing with thousands of inputs. The volume of price data required rises manifold.

If the National Statistical Office has collected all those prices, it should tell us. Why was this not disclosed when the series was introduced? We were told only when the GDP figures came out that input prices had been used for inputs and output prices for outputs. The question is whether the required price data actually exist.

Q: We have been told that the sources and methods will be disclosed. Would that address the concern?

A: Sources and methods give you the conceptual basis of the change. They do not necessarily give you the underlying data.

If this kind of price data exists, why does the NSO not make it available to other users? It would improve everybody’s work because we are currently relying on approximations. The lack of transparency raises a serious question: are the official estimates themselves based on even larger approximations than those used earlier?

Q: Is the producer price index also part of the problem?

A: There is no question that, in terms of international standards, producer prices are the right way to go. It is not that India has been unaware of this. We have been trying to obtain reliable producer prices for at least 30 years. The problem has always been getting the data.

Producers regard these prices as trade secrets. A manufacturer may sell the same product to different distributors and wholesalers at different prices. Wholesale or consumer prices are comparatively easier to collect because an investigator can go to the market and ask for a quotation. Producer prices have to come from a company’s records.

In many Western economies, the collection of statistics law is enforced rigorously. Companies are sent forms and are required to submit the information. India also has legal provisions for collecting statistics, but enforcement has been different.

If reliable producer-price data have now been assembled, that would be extremely valuable. It could tell us how much of the price paid by the consumer reaches the manufacturer and how much is taken by intermediaries. At present, we rely on rough assumptions about mark-ups.

Conceptually, therefore, I have no quarrel with what is being attempted. My question is: what data are being used, and why are those data not being made available?

Q: So, you do not dispute the methodology. Your concern is whether the database needed to support it exists.

A: It is not simply that the quality of the data may be suspect. A methodology should be changed only after you are convinced that the database needed to support it is in place. The data should precede the methodological change.

If the data exist, let us see them and assess their quality. We also need detailed documentation on how the prices were collected. How were producer prices obtained? How were the prices of the different inputs estimated? At present, users are not being given enough information to answer those questions.

Q: Is double deflation responsible for situations in which real manufacturing growth is higher than nominal manufacturing growth? Does that indicate negative inflation or a problem with the data?

A: Not necessarily. Suppose the prices of inputs rise faster than the prices of the final output. Nominal gross value added can then decline because the value of the inputs is rising faster than the value of the output.

Under double deflation, however, the inputs and the output are deflated separately. If there has been no change in the volume of either, real value added would show no decline. Real growth can therefore be higher than nominal growth. That is statistically possible and does not, by itself, mean that something is wrong.

Q: Should India abandon double deflation and return to the earlier method?

A: No, not if the statistical system can convince users that it has the data required to implement double deflation properly. It is a step forward conceptually.

But if the necessary data cannot be demonstrated, then there is a case for returning to a method based on data that users can access and replicate. Under the present arrangement, users are being given a black box.

Base Revision

Q: Another objection concerns the revision to the previous year’s April–June GDP. At current prices, the estimate was around ₹86 trillion under the old series and is about ₹80.4 trillion under the new series. Can a change in the base year alter current-price GDP by as much as ₹5 trillion or ₹6 trillion?

A: Yes, it can. When the base year is changed, very intensive data collection is undertaken for that year. That degree of data collection cannot necessarily be replicated every year. As you move forward from the base year, several datasets have to be estimated using approximations.

Take services. For many services, the estimate may be based on the number of workers multiplied by an assumed productivity figure. That can go wrong because productivity can change considerably. A new base-year exercise can reveal that the earlier assumptions were inaccurate.

So, current-price GDP can also change when the base year changes.

Q: But when the base was changed to 2011–12, GDP declined by a little over 2%. This time, the reduction is closer to 7%.

A: That is the important point. By and large, previous base-year changes have resulted in relatively small declines, usually around 2% to 3%. This time, the decline is more than twice as large.

The big question is what we were doing from 2015–16 onwards that created such a large discrepancy, which is now being corrected through the base change.

MoSPI should be given credit for bringing this discrepancy to light. It is important. But the implication is that GDP may have been consistently overestimated for 12 or 13 years.

Q: In other words, the economy was not as large as the earlier estimates suggested.

A: That is correct. If the revision is carried through consistently, something similar to what happened after the previous base change will occur.

When the base year was changed in 2015, the historical data were revised from 2004 onwards, and the estimated growth rate for the 2004–2011 period came down substantially. Something similar may happen now, possibly to a greater extent.

Q: Would that mean that the growth rates reported after 2014 may also be revised downwards?

A: Yes. If the new methodology and database are applied consistently to the historical series, growth rates of 6.5% or more reported during that period may have to be revised downwards.

The current April–June estimate of ₹88.27 trillion could also be revised as more complete data become available.

Trust Deficit

Q: At the heart of the controversy is a problem of trust in government data. Why has that problem arisen?

A: Transparency.

Until the mid-1990s, official data were used primarily by the government. The private sector did not use them extensively. After the 1991 reforms, as the private corporate sector expanded, companies became major users of official statistics.

Until around 1999, there was very little transparency. Statistical estimates were released without the underlying data. Subsequently, pressure for greater disclosure increased, and transparency improved dramatically. Technology also made it possible for ministries to place detailed datasets on their websites.

That transparency now appears to have declined.

Q: You were India’s first Chief Statistician. What can be done to make the system more transparent and rebuild credibility?

A: Transparency is not really in the hands of the Chief Statistician. It is ultimately a political decision. Without political approval, greater disclosure will not happen.

When I became Chief Statistician, the National Statistical Commission was also established. It acted as an auditor of official data. Statistical releases were not cleared unless the commission had approved them.

The commission played that role from around 2006 until 2019. Since then, it appears to have dropped out of the picture. Despite the present controversy, I have not seen the National Statistical Commission’s assessment of the new series.

Q: The post of Chief Statistician is also vacant, with the Statistics Secretary carrying out the functions of the office. What does that indicate?

A: The functions may be carried out administratively, but that is not the same as having an independent Chief Statistician. Not appointing one effectively weakens the autonomy of the statistical system.

The same is true if there is no functioning National Statistical Commission.

Growth Mix

Q: There is also a question about the composition of growth. Finance, IT and professional services account for almost 27% of GDP and grew by about 12.1%. Trade, hotels and transport grew 8.5%, while construction grew 7.7%. Manufacturing’s share, meanwhile, has fallen to around 13%.

Should we be concerned that the fastest-growing sectors may support the livelihoods of only a relatively small part of the population?

A: Not necessarily. We need to examine the sectors individually.

Finance is almost entirely corporate. It is dominated by banks and insurance companies, with some smaller components. That has always been the case.

Trade, however, is almost entirely non-corporate in the Indian context. When the data refer to hotels and restaurants, they do not mean only large hotels. They also include small restaurants and dhabas. A substantial part of that activity is non-corporate.

Transport is also largely unorganised, apart from the railways, which are usually treated separately. IT, on the other hand, is highly organised and corporate.

Q: But for quarterly GDP estimates in sectors such as hotels, restaurants and trade, are we not still relying heavily on corporate data?

A: Yes, and that is a different issue. Corporate data are being used to extrapolate the performance of the non-corporate sector. That is where the problem arises.

The sector itself may contain a large amount of non-corporate activity. But the quarterly estimate may still be based primarily on the much smaller corporate segment for which timely data are available.

Q: The broader concern is employment. Finance and IT are relatively capital-intensive, while manufacturing, from which greater employment generation is expected, is not doing as well. Where does that leave the questions of equity and livelihoods?

A: The bigger issue is the role of micro, small and medium enterprises in the growth process.

Outside agriculture, a large proportion of employment is in the unorganised and MSME sectors. We have some data on those enterprises, but the information is available annually through the survey of unorganised-sector enterprises.

For quarterly GDP, the statistical system essentially uses the limited corporate data available at that point and extrapolates from them. When the annual survey results become available, those data can be incorporated. Until then, we do not really know what is happening in the MSME and unorganised sectors.

Q: The first requirement for rebuilding trust, then, is greater transparency about how the data are collected and GDP is computed. The base-year revisions also suggest that the size of the economy, and possibly its historical growth, may have been overestimated. Is that a fair summary?

A: Yes. The reduction in the estimated level of GDP has already happened. If the new series is carried backwards consistently, the historical growth record may also have to be revised.