The golden age of labour market research
Theories juxtaposed with big data will give birth to a new generation of policy-relevant models
There has been a plethora of articles in the public domain following our report titled “Towards a payroll reporting in India”. Some constructive. While we have addressed all the issues raised (sometimes raised multiple times), we believe it is fair to address them again.
First, why is there a need to create a payroll report in the Indian context? We believe labour markets in India have long been characterised by information asymmetry due to the reliance on surveys rather than actual data. Economic policies thus get built on theoretical grounds but have little or no validity from a micro point of view.
Governments, researchers, and the private sector in India have traditionally relied on household surveys to assess labour market conditions. While these surveys were indispensable earlier, they are being questioned nowadays—they are costly, dated and limited in scope, coverage and depth. Economists all over the world (India included?) are exploring the use of big data and machine learning to understand the labour market on a real-time basis and in the process deal with large semi-structured data (like EPFO, ESIC and NPS) rather than a small survey. This real-time strategic intelligence offers crucial insights, such as how many jobs were created in total, which jobs were most in demand, the specific skills employers needed, and the career directions that offered the highest potential for workers. This is exactly where we wanted to make a beginning.
Real-time labour market information (LMI) can not only offer a much better understanding of the current job scenario, but it also exploits the rise of online job search and job matching by “scraping” or “spidering” the web for job postings and résumés. The scraped data is then structured into occupations, skill groupings, career pathways and many other combinations to provide an up-to-date picture of the labour market that can facilitate decisions by job counsellors, educators and even job seekers and students.
Researchers and governments have started using such scraped data to study the impact of unemployment, the spatial mismatch of jobs, and firm demand for skills. For example, the United States Department of Labor and the World Bank recently announced the use of big data and LMI for assessing the job market scenario on a real-time basis in addition to the usual methods. The rise of mobile technology has made these approaches extremely relevant in lowerand middle-income countries where services like Souktel, Ta3mal, and Duma Works bring employers and job seekers online to search for both formal and informal work.
Having explained our rationale of using EPFO data for understanding payroll reporting, let us now highlight some more points related to our report.
First, according to the Economic Survey, from social security perspective the stock of EPFO, ESIC and government employees is estimated at 75 million, nearly equivalent to our stock at 77 million (excluding amnesty and zero contributions). Additionally, according to Income Tax, the total number is at 127 million, much higher than our overall estimates at 102 million (based on social security). Most importantly, the stock of formal to informal could be now as high as 53 per cent.
Second, the EPFO contribution is for those whose average monthly wage/salary is ~15,000. The ESIC contribution is in firms whose monthly wage/salary is below ~21,000. Hence, these jobs are mostly low paying. The total workforce entering is 15 million, graduating is 6.6 million and non-graduating is 8.4 million. The rest 10 million (40 per cent drops out of 25 million) drops out of workforce on account of marriage or gets into agriculture. It is not that any of them are fully unskilled but this is mostly irrelevant. EPFO is also meant for the graduating workforce and others too, as over ~15,000 is purely voluntary contribution. The moot point is, we must skill the people based on payroll addition across 190 industries.
Thus, the argument that India creates formal jobs at 7 million, which is more than the number of workers entering the workforce every year (this is an effective universe of 18.4 million and not 6.6 million as claimed) is a case of mistaken data interpretation.
Third, the US generates three times more non-farm jobs than India. Again, a careful misrepresentation of words. According to official payroll data, the USA created 2.17 million skilled nonfarm jobs in 2017 and India added 6.6 million skilled graduates to the labour force. There is a difference between job creation and labour force addition.
Fourth is the argument why consumption growth is not significantly impacted even as we are adding 7 million jobs in a year. To begin with, we don’t agree with this argument because consumption has expanded at a healthy 6.8 per cent in the last five years. Even if we buy this argument, a low pay of less than ~15,000 is not going to increase consumption dramatically in the ultimate analysis.
Fifth, we should not confuse stock and flow. For example, the total stock of bank deposits in India since Independence is ~109 trillion and even if we assume a 10 per cent deposit growth, the flow comes to 11 trillion—not a small number! Hence on a 92 million stock, there is no logic why flow cannot be at 7 million or 7.5 per cent.
Sixth, what we indeed show is that it is a job-based growth, but skilling is not happening in the right place and thus policy makers do not know whom and what to skill. Thus, in a majority of the cases it does not end up being a well paid job. This means that GDP growth and job growth may be least correlated! Interestingly, even China created more than 13 million jobs last year, when the GDP growth was at its lowest!
We are clearly entering a golden age for empirical labour market research. It must be a priority for social scientists to work together with data scientists in finding a common ground where theories juxtaposed with big data give birth to a new generation of policy-relevant models.