India s national income statistics are under a cloud. The Central Statistics Office (CSO) has released official revisions of the GDP estimates for the years 2004-05 to 2013-14 in the Manmohan Singh-led government’s tenure. The revisions make the Narendra Modi government s performance on the economic front appear better than that of its predecessor.
In 2015, when the CSO had first computed this set of estimates, the growth rates for the years in Dr. Singh s tenure were higher. The NITI Aayog rejected those numbers and blocked their release. Whether the CSO should have yielded to the NITI Aayog on a purely statistical matter is not clear.
In the recomputed estimates, which were released last month by the CSO under the guidance of the NITI Aayog, GDP growth does not exceed 9% even once during Dr. Singh s tenure. The fastest growth rate reached was in the year 2010-11 (8.5%). The growth rate for this year, before this revision, was estimated at 10.3%. The best year in Mr. Modi s term so far has been 2015-16 (when GDP growth reached 8.2%). In fact, the growth rates for the majority of years in Dr. Singh s term have been cut drastically. Besides 2010-11, the growth rate was slashed quite sharply for 2007-08, from 9.8% to 7.7%.
Naturally, these revisions have stoked a controversy. Besides the political duel between former Finance Minister P. Chidambaram and his successor Arun Jaitley, eminent statisticians have posed questions over the technical issues at hand. The CSO has offered no satisfactory answers. Even the Chairman of the Prime Minister s Economic Advisory Council, Bibek Debroy, is concerned about what he calls the use of less than perfect deflators by the CSO.
Statistical stunts
Mr. Jaitley is at pains to stress the CSO s credibility, and has emphasised that the revisions are compliant with the international guidelines, the System of National Accounts (SNA), 2008. The truth is, many of the proxies and techniques that the CSO has used are, in fact, not recommended by the SNA. At best, they are tolerated under the SNA system.
One particular statistical stunt that the CSO has introduced is a structural break in its back series in 2011-12. Let us understand this in detail.
Macroeconomic aggregates such as GDP and GVA (gross value added) are estimated every year at the prices of a selected year, the base year. Base years are periodically updated, and the GDP for every year all the way back to 1950-51 is then re-estimated. In 2015, the base year was updated from 2004-05 to 2011-12. Improvements in estimation methodology were also carried out. But there was a problem: non-availability of appropriate databases complicated the re-estimation backwards.
Three years ago, at the time of re-basing the GDP series, the forward computation was done using data sourced from the Ministry of Corporate Affairs MCA-21 database of balance sheets. Its use led to growth getting revised upwards substantially for the years after 2011-12, including for the last two years of Dr. Singh s term. But the MCA data are available only 2011-12 onwards. So, what was to be done for computing the series before 2011-12? This was the principal difficulty in backcasting the rebased series.
The CSO worked out a proxy. Its use would have led to growth rates getting revised upwards in the years before 2011-12. This was not agreeable to the NITI Aayog, and the back series computed with it was withheld. For three years, the CSO and the NITI Aayog could not resolve the problem. Now, the CSO, under the rather controversial guidance of the NITI Aayog, has for a proxy used data extracted from the Annual Survey of Industries (ASI), the database that was used for the earlier 2004-05 base year series. Combining MCA data with the ASI data is technically problematic. There is no statistically robust way of seamlessly linking these two datasets. Their coverage differs significantly.
Complications
Consider one of the complications introduced by the ASI s relatively smaller coverage than the MCA s. Take a company that has manufacturing establishments across the country. Such a company may also have non-manufacturing establishments as part of its set-up. The total corporate GVA should ideally be the sum of manufacturing and non-manufacturing establishments. While the MCA would cover both the manufacturing and non-manufacturing GVA of this company from 2011-12 onwards, the ASI has been found to cover only manufacturing establishments in such cases. Non-manufacturing GVA inside a manufacturing corporate enterprise is not captured by it. This GVA, in fact, does not get covered at all as the service sector surveys also leave it out.
Some of the GVA in the years before 2011-12 in the back series has escaped estimation altogether due to such holes in the ASI s coverage, making the year 2011-12 a point of discontinuity.
A structural break can be observed in the back series before and after the year 2011-12. The upgradation to MCA from ASI data 2011-12 onwards had led to upward revisions. Going backwards, the revisions are by and large sharply downwards.
Glossing over challenges
The statistical challenge before the CSO is to estimate the GVA that remained uncaptured by the ASI. This is a complex problem with no clear solution. Whether the holes led to an underestimation or overestimation and what the impact was on the growth rates is difficult to judge. Service offshoots can lead a conglomerate or can be a drag. The now-defunct Kingfisher Airlines associated with the business group United Breweries illustrates how a service establishment can ensure that a conglomerate s GVA goes up but it may also drag it down. The trouble is, rather than admitting to these challenges transparently, the CSO has sought to gloss over them.
One of India s most well-regarded statisticians, T.C.A. Anant, was not willing to replace the MCA database with the ASI database in the back casting exercise when he was Chief Statistician of India. Months after his retirement, the CSO went ahead and did exactly that.
Puja Mehra is a Delhi-based journalist