IN DEFENCE OF ESTIMATES OF JOB LOSSES
In July 2017, I wrote in this column that the number of people employed during January-April 2017 was 1.5 million less than those employed in September-December 2016. The number of employed fell from 406.5 million to 405 million. Employment fell further to 404.6 million during May-August 2017.
I had conjectured that the 1.5 million fall in employment during January-April 2017 could be attributed to the November 2016 demonetisation. I had also added a caveat that the fall could also be because of seasonality but since we do not have a long time series it was not possible to adjust the fall for seasonality before attributing it to demonetisation.
This conjecture has been criticised by Bibek Debroy (Moonwatcher’s Logic,
The Indian Express, October 19) as post hoc, ergo propter hoc fallacy. Later, it was criticised by TCA Anant that if I adjust for seasonality then I will not see any fall in employment. Surjit Bhalla (“No Proof Required: Demonetisation and its Contents”, The Indian Express, November 8) has compared the January-April 2017 employment data with the January-April 2016 data and shown that employment has gone up and so, employment has gone up post-demonetisation. There could be many more criticisms but I know these three from very eminent economists.
Bibek Debroy’s criticism that I cannot attribute an observation to a phenomenon just because the observation came after the phenomenon is facile. Demonetisation was a huge shock to the economy. It was expected to have a severe short-term impact upon the economy. The long-term impact is a gamble but the short-term impact was expected by everybody. Even the Prime Minister asked for a little time to solve a big problem. Newspapers narrated countless stories of job losses for weeks after demonetisation. The CMIE-BSE real-time measurement of unemployment gave us the opportunity to observe its impact in real time. Eventually, we will be able to do more rigorous natural experiments to understand this impact better. But, for now, we have a measure of the immediate impact.
Several eminent economists have told me that a job loss of 1.5 million postdemonetisation is possibly an underestimate. Possible. 1.5 million is a net number. Gross job losses were larger and these were offset by job gains elsewhere. This is a regular affair. What matters is the net increase or fall in jobs.
TCA Anant was mentioning what I had already pointed out too, in my piece in Business Standard on July 11, that the employment numbers should be seasonally adjusted. His critique was that I had made only a fleeting mention and not emphasised this sufficiently. I may plead guilty to that. He also said that once the numbers are seasonally adjusted then I may see no fall in employment. This could be a fair conjecture and I do not disagree that impact of seasonality must be investigated when we have more data. But, then why doesn’t he, as Chief Statistician of India, produce seasonally adjusted series for the IIP and other fast-frequency indicators generated by the official machinery. Why pick on a very young series generated privately?
Surjit Bhalla is a statistician-wizard, a numbers magician besides being an eminent economist. He addresses the problem of seasonality by making yearon-year comparisons instead of sequential comparisons as I do to say that 1.5 million jobs were lost after demonetisation. He uses the BSE-CMIE estimates to show that jobs grew by 4.2 million in January-April 2017 compared to January-April 2016. Since these two are like periods, he seems to suggest that there is no seasonal vitiation. Further, since demonetisation happened between these two periods, we can say that shockingly, jobs grew after demonetisation.
Surjit is, of course, wrong. He is wrong for two reasons. First, he is wrong on an elementary statistical problem. A year-on-year comparison is not a seasonal adjustment or a correction for seasonality in any way. A year-on-year comparison ensures we are comparing like months, that’s all. This is not a seasonal adjustment.
Further, while it is true that 4.2 million jobs were added between JanuaryApril 2016 and 2017, these were not added post-demonetisation. As many as 5.7 million jobs were added during MayDecember 2016 (a period that is mostly before demonetisation) and 1.5 million jobs were lost during January-April 2017, that is post-demonetisation.
Secondly, to understand the effect of an event (such as demonetisation) we have to see the outcomes (such as employment) before and after the event. A seasonally adjusted sequential comparison is the right way of doing this, not a year-on-year comparison.
While 4.2 million jobs were added from January-April 2016 to 2017, these were not added post-demonetisation