Measuring Indian GDP: Arvind Subramanian Can’t Be Taken Seriously

Estimation of gross domestic product of an economy comes from a painstaking grass- roots process. An econometric exercise, such as the one conducted by former chief economic adviser Arvind Subaramanian, can never be a substitute.

The Indian statistical agencies have a long history. Stalwarts associated with them set up robust processes to measure gross domestic product, even in a difficult economy like India with poor data availability for a large informal sector. Periodic revisions are required to keep pace with the changing structure of the economy. The shift to the base 2011-12 was a major one, and therefore has generated a healthy debate.

Use of larger, newer and more representative databases, more consistency across categories, conceptual improvements including for the informal sector and shift towards international practices have also been undertaken. These were all required.

For example, the much larger MCA-21 database of companies, arising from statutory filings under the Companies Act, is being cleaned out for dummy companies. But no one thinks we should go back to the earlier sample of 2,500 firms.

Problems arising from deflators using CPI or WPI in a period when the two had large divergences are shrinking as the two converge.

Therefore Arvind Subramanian’s recent argument that average growth is over-estimated by 2.5 percent since 2011 is difficult to accept. Moreover, it suffers from serious conceptual and econometric flaws.

Using regressions to predict GDP based on a few macroeconomic indicators that are found to predict well, but are not produced by government statistical agencies involved in the estimation of gross domestic product, he gets a predicted value of Indian GDP that is lower than the Central Statistics Office estimations after 2011. He finds, however, for a set of other countries these indicators are able to predict GDP. The indicators are export and imports of goods and services and domestic credit to the private sector.

We examine a few flaws.

First, when variables are growing, a regression in levels can give spurious results. He should have done his panel regression in growth rates.

Second, to ensure his results are robust he should have checked that his indicators did not underestimate growth for periods before 2011 also. And that they did not under or overestimate growth for other countries that had not undergone a change in estimation methodology. The robustness exercises he does are inadequate.

Third, the variables he uses are inadequate for India especially, and are biased in favour of his result. Credit growth was very low after 2011 because of non-performing assets in banks and a macroeconomic policy squeeze. India’s credit/GDP ratio has always been relatively low. As a major oil importer, both India’s imports and exports are vulnerable to oil price crashes, such as the one that occurred in 2014. India’s growth story has not been an export-led growth story. Export and import are tightly linked since it is hard for India to finance a large current account deficit.

Other variables useful to predict growth such as government final consumption and labour force growth are also available independently of government statistical agencies but were not used. Government expenditure multipliers often exceed unity. India has favourable demographics, while decline in working age population is causing economic stagnation in many countries.

Fourth, in his framework any structural change which leads to less reliance on external sector for growth, would be attributed as overestimation/ underestimation. He notices this for China but sets it aside arguing Chinese have a long history of overestimating GDP.

Fifth, although he repeats his regression for different types of country sets classified by income, oil export shares etc., he does not use a set of countries with the same dominance of the service sector that India has. Since this may be the deeper reason for India’s relative growth differences it should be controlled for.

Sixth, he neglects or assumes productivity growth differences to be constant across countries, which is inadequate in a growth regression, and unfair to India whose productivity growth differential was rising in this period.

Therefore, his is not the right framework to predict gross domestic product let alone estimate it.

We repeated Subramanian’s results using similar variables and country sets. But we find that export and imports of goods and services and credit to private sector are not able to predict growth pre-2011 (2006-10) also unlike his findings. Moreover, they underestimate GDP. Therefore it is incorrect to attribute the post 2011 change to estimation errors.

He is also incorrect when he argues higher estimated growth rates seriously misled policy. Policy-makers knew that there was excess capacity in industry and investment growth had slumped, while the changing structure of growth made the new growth estimates higher. Yet they kept real interest rates higher than warranted, and imposed tight restraints on the budget, because they were focused on inflation in the newly introduced and strictly interpreted inflation targeting framework.

The Indian statistical process is robust, independent and continually improving. More can and should be done to strengthen it and its advisory committees but bringing in outside experts who do not understand the economy can be counter-productive.

Ashima Goyal is professor at IGIDR, and a member of the Prime Minister’s Economic Advisory Council. Abhishek Kumar is a doctoral student working with Ashima Goyal.

The views expressed here are those of the authors and do not necessarily represent the views of BloombergQuint or its editorial team.