‘Seasonal adjustment is not a simple mechanical task’

THIS IS with reference to a piece that appeared in Business Standardon July 11, 2017, and a further comment related to that piece that appeared on November 14, 2017, by Mahesh Vyas (“In defence of estimates of job losses”). In that piece, he made two observations referring to me by name and it is important that a statistically correct position on these two issues is placed on record:

The basic issue of the July 11 piece and the comments on that thereafter are: Analysing CMIE data, Vyas claimed that there could have been as many as 1.5 million jobs lost post demonetisation, during January to April 2017. He arrived at this conclusion by comparing the January to April employment number from a CMIE survey with an earlier figure from September to December 2016. He implied there and repeated in his column on November 14 that this was very likely due to demonetisation. In this connection, he refers to my observation to him that before attributing this change to any causal factor it would be important to examine it in light of what we know about the seasonal dimension of employment change. In this connection, he may recall that I had drawn his attention to data in public domain pertaining to NSS 68th round survey conducted during 2011-12. NSS breaks its annual survey into four sub-rounds, corresponding to four quarters, starting the first subround during July to September 2011 and the fourth sub-round between April to June 2012. An examination of this data, in a manner similar to what Vyas has done with CMIE data, would suggest that the number of workers on current weekly status declined by 19.6 million between January-March 2012 and October-December 2011 when there was no demonetisation. This calculation is summarised in Tables 1 and 2. The Rural male Rural female Rural male+female Urban male Urban female Urban male+female Rural+urban male Rural+urban female Rural+urban male+female Rural male Rural female Rural male+female Urban male Urban female Urban male+female rural+urban male rural+urban female rural+urban male+female Rural male Rural female Rural male+female Urban male Urban female Urban male+female Rural+urban male Rural+urban female Rural+urban male+female magnitude of seasonal fluctuation revealed therein was clearly more than the change noted by Vyas in his column.

My second observation relates to Vyas’ comments on even the CSO not doing seasonal adjustment and Surjit Bhalla’s comments on looking at year-on-year change which may at best be described as statistical legerdemain. Year-on-year growth calculations as done by the CSO with its Consumer Price Index and Industrial Production Index are less affected by seasonality as compared to month-over-month or quarter-overquarter changes, as done by Vyas, because in general the seasonal element is common in the same months in both the years. It is true that it would be better to do month-on-month or quarter-onquarter comparisons with due adjustments for seasonability than yearon-year. However, undertaking seasonal adjustments is not a simple mechanical task that can be applied through a software programme; it requires careful accounting for idiosyncratic festivals like Diwali, Eid etc, which do not have a predictable annual calendar. The difficulties have been analysed in academic papers with suggestions for solutions. You may, for example, see the paper by Bhattacharya et al “Seasonal adjustment with Indian data: how big are the gains and how to do it” accessible online at Ajay Shah’s blog1. A perusal of the paper would reveal that care needs to be taken in generating seasonal estimates. A leading financial journal had noted: “The calculation of