India’s Chief Statistician TCA Anant defends the Central Statistics Office's (CSO) GDP estimates and underscores the institutional challenges in improving the accuracy further.
Edited excerpts:
It has been said that the CSO’s estimates overstate economic growth. The main issue is that not all of the growth that the figures show is on account of real growth. Much of it is purely due to the increasein prices, which ought to have been deflated out adequately.
We had used CPI IW (Consumer Price Index - for Industrial Workers), the earlier CPI series. We have started using the CSO’s new CPI series. Other than this, whatever price information has been available is the same that has been available for the last 50 years.
This series (2011-12 base year GDP) has not done something different in the choice of deflators or prices from the past. CPI covers a smaller vector of activities and goods in the economy, so it will be appropriate to use for the large number of them, prices derived from the WPI. (GDP deflator is a constant used to offset the impact of rising prices on the GDP.)
What about the disconnect between the CSO’s estimates and the reality on the ground? People do not feel all of the growth.
When you have a whole segment of manufacturing seeing falling prices, that segment is not necessarily going to be feeling very well-off. Large chunk of WPI is negative. It has had peculiar implications.
There is a large segment of commodities which are there in WPI, and are not there in the CPI, which have seen a different price movement. When you have such sharp relative price changes there must have been real effects in the economy and they need not be always to the good.
One example which people have cited is you’ve seen slow growth in sales, in fact at times negative growth in sales in certain segments in the economy, but costs have fallen even more sharply.
A logical consequence is value-added has gone up. If my costs fall faster than my sales, I am still making an increase in real terms. In some sense we have picked that up.
While GDP estimates show manufacturing growth has picked up, the IIP doesn’t quite. Which of the two should we believe?
Output measurement the way it is done in the Index of Industrial Production uses the following: It takes a list of items, which it constructs in the base year. For each item, it identifies producing entities, which it also does in the base year.
Whatever structure they fix, they then freeze. They start getting output figures from these entities month over month from that they construct the index. In IIP, this fixed basket of items and entities introduces a peculiarity.
One, if exit of entities takes place their output falls to zero. But because the basket is frozen, new entities are not brought in, you can’t take account of new entities.
Secondly, we fix the basket of items also in the base year. But the item baskets themselves are volatile. I may use, smart phones, simpler mobile phones, I may use desktops instead of mobile instruments.
There may also be new items which come in. Maybe watch phones. I will be limiting myself to whatever was chosen in the base year. Something which goes out of fashion goes out. But something that has come in has not come in because the base year has not changed.
The selection of item and producing entities has remained the same from many years. There will be implications. IIP growth will have a certain directional bias. The recommendation always has been: try to make it more dynamic.
The recommendation was to revise it more frequently. So with that in mind the attempt has been to push forward the date of revision. We were earlier doing decade-wise revisions…in 1993-94 and so on. Now we are targeting the revision within seven years.
Ideally, we should go for more, quicker revisions. But revisions in a process where you have to physically collect data from entities and establish collection system, where no regular system of reporting production is available, is a cumbersome task. It is an institutional challenge.
IIP is a very old series. It’s one of the longest times series which we have been producing. It goes back to the 60s or the 50s.
These are not new limitations…have been pointed out in the past.
The difference between the IIP and the GVA is huge.
When you break it up in to its components and you look at each of these elements the fact is the part of this hugeness was already there in the old [2004-05 base year GDP] series.
Means the IIP is not reflecting macroeconomic trends…
IIP has limitations as a representative of aggregate growth in manufacturing. It is still useful—and the only measure available—for the micro picture for a broad sense of what is happening to different segments.
We continue to use IIP for estimating capital formation and construction because there is nothing else available as a leading indicator. It’s very useful when carefully used.
The 2011-12 base-year GDP series shows faster growth than the previous one (2004-05 base year). It has been suggested that, at least for a few years, CSO should publish GDP figures using the old method alongside those according to new method.
It is not recommended to generate two estimates because you cannot have two estimates.
Certain estimates are flawed in the old series. When we did the base revision exercise, we computed the GVA in trade and showed it to be substantially less than what had been computed in the old series.
Why? The estimate of trade is most robust in the base year because for the base year we have available an NSS survey in non-incorporated establishments. Large part of the trade does happen in the unincorporated segment.
These surveys are done infrequently. The last one was done in 2010-11. The next one is just getting completed. The previous survey of trade was from 1999-2000.
There was a ten year gap between the two. We found that we had overestimated trade by a very significant amount.
What does it mean to say continue the old series? Do I take account of this overestimation? This is just one contradiction, which is implied in this. We also showed that we have underestimated GVA in manufacturing once we examined the corporate data carefully.
One reason why the new series differs from the old is purely the use of a different data set. The second factor is corporate value-added differs from ASI (Annual Survey of Industries) by the notion of enterprise value-added versus the establishment value-added.
The ASI would capture information from a plant, while my data would come from the accounts filed by the company.
The latter could include data from services or trading at show-rooms. Earlier, a lot of this was not captured at all. Now we are picking up that also, admittedly showing it as part of manufacturing because we don’t have disaggregated corporate accounts.
There is a third element. The coverage of the MCA database is much more comprehensive compared to either the ASI or the IIP coverage. A much larger set of entities have been captured.
We are partly addressing people’s requirement by trying to make an attempt at back-casting the new series with the data that was available in the past.
That will give people a longer time-series. We will do it for about seven-eight years.
