In terms of inequality dynamics, the Modi years of 2014-2023 can be divided into 3 phases: 2014-2017,2018-2020, and 2021 onwards. In the first phase, the economy was growing moderately fast and bothincome and wealth inequality continued to rise. In the second phase, from 2017-18 to 2020, growth slows down considerably and then plummets in 2020. In this second phase, we see top 10% income and wealth shares decline by 1-2 percentage points. The most likely explanation for which is the pro-cyclical nature of inequality, i.e. the rich tend to benefit disproportionately from boom periods and are isproportionately hurt during slumps, as Ghatak et al. (2022) also argue.41 This seems the most likely explanation especially given that we observe similar trends for both income and wealth during this phase. Moreover, as shown in Figure 2b, the wealth of the richest Indians as a share of national income also declined between 2018 and 2020. It is hard to think of other factors that concomitantly explain these trends for both incomes and wealth.42 Measurement error is of course a possibility, which we address shortly. Finally in the last phase, after the lock-downs were lifted and the economic effects of COVID-19 dissipated, we find top shares revert to their upward trend in 2021 and 2022, while bottom shares decline back to their 2014 level. By examining the growth incidence curve for incomes and wealth between 2014-2022, we find that the real beneficiaries in the recent years appear to be the super-rich, the top 1% and beyond (Figure 17b). This is particularly so for wealth concentration at the very top. This lends some support to political economy assessments that have characterized the economic system in India in recent years as “conglomerate capitalism”(Damodaran, 2020) and a “conclave economy” (Bardhan, 2022).The other interesting aspect to note is that both with income and wealth, the middle 40% seem to have grown slower than the bottom 50% during this period. This is likely to exacerbate the phenomenon of India’s “missing middle class” (Chancel and Piketty, 2019). The bottom 50%, in return, grew at the same rate as that of the average of the population, preventing an increase in their share of total income and wealth. We must, however, emphasize that this result might be overly conservative.
Recent ICE360 data indicates a significant decline in the bottom 50% income share over the period 2015-16 to 2020-21 (see Appendix Figure B.3), with growth rates well below the average. Enhanced access to household survey and administrative data sources is essential for a deeper understanding of these dynamics, as discussed subsequently.
7.3 Growing data challenges in recent years
During the last decade, various key data sources in India have either become unavailable or their quality has become suspect. This applies to all the key inputs that go into our inequality series: national income accounts, tax tabulations, and surveys. We briefly discuss these issues with the aim of drawing caution when interpreting the estimates for recent years.
National income accounts: Various concerns have been raised about validity of India’s national income accounts data in recent years.43 At least two detailed empirical exercises, one by an ex-chief economic advisor to the Government of India, point to possible over-estimation of GDP in the years post-2011 (Morris and Kumari, 2019; Subramanian, 2019a). Some concerns have also been raised regarding the possible mis-measurement of India’s GDP deflator (Subramanian and Felman, 2023).
More generally, the dated nature of the underlying data used to estimate GDP is very concerning –key inputs like the CPI, WPI, input-output tables, industry codes, consumption expenditure, etc. are currently based on data that might be 10-15 years old (Sapre and Bhardwaj, 2023). This is especially a worry for aspects relating to the informal sector of the economy. If it is indeed the case that GDP is being overestimated in recent years, that would imply that our inequality estimates would be slightly downward-biased. 44 43 As India’s ex-chief statistician clarified recently, the issue with India’s GDP estimates in recent years seems to be less about methodology and more about the severely outdated underlying data and unreliable proxies (Thapar, 2023).
44 This is because (70% of) per-adult net national income serves as the “control average” for the generalized Pareto 28
Tax tabulations
The British colonial administration introduced an individual income tax with the Income Tax Act, 1922. Since then, data on individual incomes began being collected and the colonial administration published this data in tabulated form on an annual basis. This practice which was continued by the Indian government post-independence. Between 1922 and 1998, annual publication of these ‘All-India Income Tax Statistics’ provided a vital source of information on top incomes, mobilizing which Banerjee and Piketty (2005) estimated the share of national income going to the top 0.01%, 0.1% and 1% during this period. There were naturally improvements to the methodology used to generate these tabulations over the years (partly owing to technological and computational improvements), but systematic and regular release of this data was not disrupted.
However, starting 1999 onwards, the government of India strangely stopped publishing these tax tabulations for reasons that remain unknown. For a whole decade when India experienced strong macroeconomic growth (2000-2010), no tax tabulations are available to date. Then in 2016, the government etrospectively released data but only starting 2011. For the next few years, data releases continued till retrospective data for 2017 was out, after which once again no tax tabulations were available. Finally in mid-2023, the government again retrospectively released data for the years 2018-2021. In short, the release of tax data has been highly erratic and incomplete in recent decades.
The reason for this remains unclear. One possibility is that the analysis and release of tax data falls low on the priority list of the Income Tax department. This stands in sharp contrast to the past when, for instance, government appointed committees specially provided recommendations on ways to better analyze and report data from income tax returns.45 Besides releasing all-India tabulations, the income tax department also used to release state-wise tabulations till 1998. These could potentially allow going beyond all-India analysis and shed light on the evolution of top incomes and inequality at the state-level. Given the size and population of individual states, larger than many European countries in many cases, this is an important endeavor. However, starting 1990 (to the best of our knowledge), state-wise statistics have not been released at all, even post-2011 when all-India statistics have been released. The non-availability of state-level data in recent years is strange, not only because it used to be released regularly before, but also because computerization and digitization of records in recent decades should make dis-aggregation and tabulation of returns at the state-level easier than before. This leaves the estimation of state-level income and wealth inequality an incomplete endeavor.
Income and consumption surveys
One of the key challenges when updating the income inequality series for the last decade is the absence of a comparable NSSO consumption survey after 2011-12.interpolation algorithm used to extract a distribution of top incomes from the tax tabulations. A lower control average would mechanically increase top income shares. To what extent this issue affects our estimates depends on the extent to which national income is being over-estimated.
45 As an example, the ‘Committee on Direct Tax Statistics’ recommended using a part-sampling and part-census approach for generating tabulations of income-tax statistics from 1974-75 onwards – all returns with incomes above INR 25,000 were to be covered by a census while those with incomes below INR 25,000 were to be sampled, with most states assigned a 10% sample, some 20%, and a full census in some union territories like Delhi (Directorate of Inspection, 1978). Incidentally, this was a time when the government of India was explicitly interested in curtailing the power of the elites.
As noted earlier, NSSO has historically steered clear of measuring incomes and instead focusing on consumption expenditures. Consequently, our measurement framework also relies heavily on these consumption surveys. The NSSO did conduct a round in 2017-18 but it was suppressed by the government. From 2017-18 onwards, the PLFS came to our rescue. As it turns out, even though it is primarily designed for labour market outcomes, it collects preliminary data on ‘usual’ consumption expenditures. By correcting these for comparability with past NSSO CES rounds, we are able to extend our income inequality series on an annual basis from 2017-18 onwards. However, this involves a correction that is bound to be only imperfect at best. This creates an additional degree of uncertainty around our estimates in the recent years. More importantly, we find that alternate data sources present contradictory trends for bottom incomes. For instance, based on percentile-level growth rates of per-capita incomes between 2015-16 and 2020-21 in the ICE360 survey, we find a steeply upward sloping growth incidence curve such that bottom 50% shares would decline from 14.4% in 2015 to 9.8% in 2020 (Figure B.3). This stands in sharp contrast to the trends in our benchmark series which suggest a relatively stable bottom 50% share over this period, besides a temporary and marginal increase (1 percentage point) between 2018 and 2020. Therefore, we see our benchmark estimates as a conservative scenario until better data emerges to improve our estimation.
Wealth surveys
It is also worth mentioning a couple of concerns relating to NSSO AIDIS that forms the basis for our wealth inequality series. First, as highlighted earlier, it appears that the issue of under-estimation at the top has worsened over the last three successive rounds in 2002, 2012 and 2018 – the total (net) wealth of USD MER billionaires in the Forbes list as a percentage of the total survey wealth increased from 1.26% in 2002 to 2.74% in 2012 to 6.01% in 2018. The issue of under-estimation and under-representation of the very rich and wealthy in sample surveys is not unique to India but the fact that the issue is getting worse over time deserves closer attention by the NSSO. More stratification and purposive over-sampling at the top could be ways to counteract the current trend of increasing non-representativeness of the right tail. Further, with all its surveys (CES, AIDIS, PLFS, etc.), NSSO should release non-response rates by some variable like, say, the ‘usual’ consumption expenditure variable that it could collect at the household listing stage – this would allow decomposing the non-representativeness of surveys for the right tail more clearly into response-related and measurement-related issues. It is also worth highlighting that we are likely to under-estimating wealth at the top of the distribution due to off-shore wealth. Of the total foreign owned off-shore real-estate in Dubai, 20% is owned by Indians (Alstadsater et al., 2024), amounting in total value to to 1.1% of India’s GDP (Alstadsater et al., 2022). The second concern relating to AIDIS relates to the timing of the release of the latest round of the data. Starting in 1961-62, these surveys were meant to be decennial surveys and indeed they were conducted every 10 years, in 1971-72, 1981-82, 1991-92, 2002-03 and 2012-13. It is unclear why the last round was conducted within a shortened gap of 6 years in 2018-19. If this is part of a broader plan of more regular AIDIS rounds, then it is a welcome change. If, on the other hand, this was the result of political considerations, then 30 there is a cause for worry. Coincidentally (or not), our estimates suggest that the top 10% wealth share may have been at its lowest during the last decade in 2018 .
