In a provocative cover story, the reputed newspaper, The Economist (January 13-19, 2018), and a newspaper I have read and admired over the last 40 years, reached the conclusion that there was a missing middle class in India, and that “income inequality in India had reached historically high levels. In 2014, the share of national income accruing to India’s top 1% of earners was 22%, while the share of the top 10% was around 56%.” The Economist offers a caveat about this radical conclusion: “Some have doubts about Piketty’s methodology. But other surveys suggest pretty similar distribution patterns”. As it happens, inequality is a subject I have worked on for decades, and besides several academic papers, have written two books on the subject—Imagine There’s No Country (2002), and The New Wealth of Nations (2017). I not only have doubts about French economist Thomas Piketty’s methodology, I have grave doubts. Piketty’s results for India just do not stand up to any scrutiny, and fail several smell tests (A smell test is akin to a duck test—if it doesn’t walk like a duck, and if it doesn’t quack like a duck, it is not a duck!).
In 2014, Piketty produced a masterpiece, Capital in the Twenty-First Century, and this book is reputed to have sold over 3 million copies, substantially above my favorite Le Carre book, Tinker, Tailor, Soldier, Spy. His work should not be taken lightly. He, along with his associates, has changed our understanding of inequality in the advanced economies of the world—they are more unequal than was conventionally thought (especially US inequality). The major Piketty innovation (again, it is a body of work with several scholars, but as shorthand, I will refer to Piketty only) was the recognition that tax payments give much better information on income than household survey responses—a very reasonable inference. Typically, household surveys only capture 25-50% of national income. In the US, since 2006, IRS publishes data on tax compliance of households, and they report that over 80% of tax that is due to the IRS is paid. In other words, the income estimate from the tax data is close to 80% of national accounts data.
In the World Inequality Report, Piketty and his colleagues, as the title implies, go several steps further than just analysing income distribution data for advanced economies. They estimate it for almost all the major countries of the world; for India, they estimate inequality for 92 years, 1922-2014. The method used by Piketty et al is broadly the same as that used for an advanced country like the US—i.e., use household survey data on consumption and income, and “marry” these data with tax payments. There will be other occasions to discuss the major flaw in the assumption that tax data in India has the same information as tax data in the US. In summary form, it does not. Tax compliance in India, in 2014, was close to 25%, not the 80-plus compliance in US (and presumably Europe). This article is about duck-smell tests. First, something curious has happened with the computation of income shares of the top 1% of the population for India. In an earlier Piketty paper (with Abhijit Banerjee, Top Indian Incomes, World Bank Economic Review, January 2005; hereafter BP for Banerjee & Piketty), Piketty reported the share of the top 1% in India for 1999 as 9%, i.e., the top 1% had 9% of the income. The NSS consumption shares for FY1999-2000 for the top 1% was 6.3%.
Thus, comparing NSS consumption and BP income shares, we obtain the result that in FY1999-2000, BP income share of the top 1% was about 2.7 percentage points (ppt) higher than the NSS consumption share. This result is very typical of inequality calculations for different countries. Typically, the Gini coefficient of inequality for income is about 6 ppt above that for consumption. According to the NCAER-University of Maryland IHDS survey for FY12, (and the survey used by Piketty with Lucas Chancel, Indian Income Inequality 1922-2014: From British Raj to Billionare Raj, hereafter CP, for Chancel & Piketty), the top 1% had a consumption share of 7.8%, and an income share of 11.6%, with the difference in consumption and income shares being 3.8 ppt, i.e., close to the difference of 2.7 ppt observed in FY1999-2000 between NSS and BP.
The first smell test failed by Piketty is that, for FY1999-2000, his new work (CP) has the share of income accruing to the top 1% as 14.7%, some 6 ppt above his own earlier estimate. A revision of an earlier estimate is not at all uncommon, and is indeed to be welcomed. It is not the revision that is a problem—it is the fact that in the CP paper, it is not outlined as to why there is such a large difference between the two Piketty estimates for the same year, 1999, and the same country, India. But there are other, more serious problems with the Piketty analysis for India. It is that the implied savings behaviour of the Piketty estimates does not conform to any known model of savings behaviour. Not the life-cycle model (the share of consumption in income stays flat over a lifetime as individuals borrow from future income to pay for present (higher than income) consumption). Not the permanent income hypothesis, PIH (savings a constant fraction of income, regardless of any change in permanent income). Nor the modified PIH which states that in developing economies savings rise with the level of income; this modified PIH is the most applicable to developing economies like India.
The average estimates of income, consumption, and savings for five categories of individuals – entire population, the top 1%, the top 10%, the middle 40% (50th to 90th percentile) and the bottom 50%, is computed in the accompanying graphic. These categories are the same as that used by Piketty.
The method of computation is straightforward—consumption shares are taken from NSS consumption survey for FY12, the income shares from Chancel-Piketty. Given national income accruing to households (obtained from Piketty World Income Distribution database, WID), and the shares of each group in national household income (again, obtained from WID), one can obtain the absolute value of nominal income in command of each group. Consumption of each group (top 1%, top 10%,etc) is obtained from the NSS consumption survey for FY12. The difference between income and consumption is the savings of each group.
The implied Piketty savings results are out of any known ballpark. The middle 50% had a savings rate of 18% in FY1999-2000; in FY76 (NCAER Income Distribution survey), they had a savings rate of less than 10%. In FY12, according to the Piketty data, the savings rate of the middle 50% collapses to zero, indeed marginally negative. Let us ponder a bit about the Piketty misfortunes of the middle 50%. This group was collectively saving Rs 1,229 billion in FY1999-2000; just 12 years later, with nominal household incomes galloping ahead at a compound annual rate of 12.5%, the absolute aggregate savings of these individuals decline! It is this lack of correspondence with most realities is what makes the recent Piketty calculations (with Chancel) very suspect. Note that the same illogicality does not apply to the earlier Piketty calculations with Abhijit Banerjee. And it is this illogicality that makes any calculations, or inferences, based on Chancel-Piketty, e.g., the Economist article on India’s middle-class, extremely suspect. Far from missing, India’s middle-class in 2016 is more than double its FY1999-2000 level of 27%. But details of middle-class calculations will be contained in a future article (the interested reader can gain some advance knowledge from The New Wealth of Nations).