Brief Analysis of “India’s GDP Mis-estimation: Likelihood, Magnitudes, Mechanisms, and Implications” by Dr. Arvind Subramanian

When measuring a particular metric with a particular scale, leads you to cast some doubt with regards to an earlier measurement of the metric with a different scale; the verification of the measurement of the metric and/or the mis-estimation of the metric has to be done with a third different scale and not with the one that led you to conclude in the first place that there is mis-estimation.

For example, a friend of yours ‘Mr. A’ claims that he is 6ft high. To check if he is really that high, you use your measuring tape ‘tape M’ to measure Mr. A’s height. You measure with ‘tape M’ and conclude that the height of Mr A should be lesser and not 6 ft as claimed by him. Then, to verify the correct height of Mr. A and find the mis-estimation (if any) in his height, you will use another ‘tape Q’ to measure Mr A’s height. This will help verify if the measurement as per ‘tape M’ is true or if Mr A’s claim of 6ft is true; and also help us decide if the height needs further investigation with some other measuring scale.

In this case, once you cast doubts on Mr A’s height after measuring with ‘tape M’; will you again use ‘tape M’ to verify the doubt and compute the mis-estimation? Will Mr A accept to this way of verification or will he insist that the height needs to be investigated with a different ‘tape Q’?

The problem in this situation is the fundamental flaw in the paper “India’s GDP Mis-estimation: Likelihood, Magnitudes, Mechanisms, and Implications” by Dr. Arvind Subramanian. The author raises doubts over the GDP growth estimates based on the new methodology primarily because,

· many (11) of the 17 indicators like growth in passenger vehicle sales, international tourist arrivals, exports, imports, electricity consumption, credit etc which are expected to co-move with GDP growth seem to be differently correlated post the methodology change (2011–2017).

This casts a doubt on the GDP growth estimates based on the new methodology.

Post this, to verify this doubt (claim), the author of the paper,

· Uses 4 chosen indicators (subset of the 17 indicators above) [chosen only because they are easily available and for no other reason]; as variables in the regression equation to estimate the mis-calculation in GDP growth.

This is the fundamental flaw in this paper. Using the same scale that raised the claim; to verify the claim and compute the mis-estimation too.

This means the claims of mis-estimation to the extent that the paper claims is at best ‘an unverified claim’.

Now, though an unverified claim, let us run a few checks to see if the claims made in the paper stands on their own weight.

The claim of mis-estimation is fundamentally based on the changes in correlation between the growth in many of the 17 indicators and GDP growth; which had high positive correlations with GDP growth in the period 2001–2011 under old methodology turning to lower positive or negative correlations in the period 2012–2017 under the new methodology.

The underlying assumption here is that the correlation between the indicators and GDP growth doesn’t vary much or is range bound. Should this really be the case? Is the correlation between the growth of these indicators and GDP growth range bound? Do any structural changes in these indicators affect the correlation of theirs with GDP growth? Or in the extreme case, is there no significant correlation between these indicators and GDP growth and the co-move in 2001–2011 purely incidental?

Let us study the 4 indicators that the paper takes into account in the regression equation to understand better. The 4 indicators are –

1. Growth in electricity consumption

2. Exports growth

3. Imports growth

4. Credit growth

Growth in electricity consumption — Were there any structural changes in electricity consumption in the last 8–10 years that could have resulted in the change in its correlation with GDP growth? May be. UPA as well as NDA’s significant focus in universal electrification, NDA’s push for mass adoption of LED bulbs and energy efficient motors in farm sector are factors that could have affected the growth in electricity consumption significantly. The paper itself acknowledges that ‘universal electrification’ may have resulted in a change of this indicator’s correlation with that of GDP growth. So, the paper produces 2 results — GDP growth miscalculation with and without electricity consumption as one of the variables in the regression equation and the results is more than a full 1% apart.

So, the paper does acknowledge that change in correlation between the growth in chosen indicators and GDP growth affects the results/claims of the paper significantly. Let us check if there is cause to believe that there could have been structural changes in the other 3 indicators used in the regression equation.

Exports growth — Services sector (IT services) exports forms a big chunk of our exports every year. There has been significant changes in the tech space with the world moving away from IT services based tech sector to an increasingly products based one. Add to this change, the base effect on exports increasing with increasing exports each year. This means the growth in exports need not necessarily keep pace with GDP growth. When exports growth doesn’t keep pace with GDP growth, correlation between them are bound to change. The changes in the tech sector happened predominantly in the last decade.

Imports growth — The single largest chunk of our imports is oil. Crude prices dropped by 40–50% from USD 110–120 levels in 2011–2013 to about USD 50–60 levels in the period 2014–2017. Correspondingly, in the 2001–2011 period, crude prices increased from USD 40 level to a high of USD 140 and then to about USD 110–120 levels in 2011. With this level of shift in movement of crude prices (increasing 150%-200% in 2001–2011 and decreasing 40%-50% in 2011–2017), there is no way the correlation between growth in oil imports and GDP growth will hold up in these 2 time periods. With crude being ~20% of India’s imports, this change in crude prices will affect import growth — GDP growth correlation too and hence can be a factor in change in correlation between these 2 time periods.

Credit growth — In credit too, the fast growth in 2001–2011; was stymied in the 2012–2017 period, mainly owing to a significant number of loans turning bad (~INR 8–10 lac crores) which increased the banking sector’s capital requirements. With NPA crisis in the banking system, differentiating the conditions of operations of banks between these 2 periods; correlation between credit growth and GDP growth changing between these 2 periods is a natural consequence.

Even if the correlation between the growth of these indicators and GDP growth held themselves in other countries as the paper puts out; it does not necessarily mean it has to hold up in India. Each country is structurally (politically, socially, economically) very different from each other.

So, we can conclude that there is reason to believe that correlation between GDP growth and growth of each of these indicators could have changed owing to structural changes in each of these indicators. The change in correlation does not necessarily mean mis-estimation of GDP growth.

The paper crumbles under its own weight owing to these 2 factors –

1. using the same parameters that led to the claim, to verify the claim and estimate the metric

2. assuming that correlations between the indicators and GDP growth do not change over time

PS — In addition, export-import growth is affected by USDINR movements if INR export-import used for finding correlation with GDP growth.