Estimation error?: Government estimates go awry when growth trends down

Historical analysis shows that the government tends to underestimate the growth in agriculture every year; as growth has slipped in the last few years, it has fallen prey to overestimation errors

On May 31, the Ministry of Statistics and Planning (MOSPI) released its provisional GDP estimates for 2020-21 and fourth-quarter GDP figures. Given the muted growth in the third and fourth quarter, the country’s GDP, according to MOSPI, contracted 7.3 per cent—the first time in four decades. However, the numbers were still a surprise as the second estimates released in February had indicated an 8 per cent contraction for the economy.

While projections from RBI and others highlight a bounce back in 2021-22, the data for 2020-21 is subject to three more revisions over the next three years.

And, historical analysis shows that there can be stark differences between the preliminary estimates released by the government and the final numbers released after three years.

Most countries follow a shorter cycle of revisions—the US has only three iterations released within three months. But in India, the GDP calendar cycle runs for three years. (Read). The first advanced estimates are released in January, followed by second advanced estimates in February and provisional estimates in May. Then the government releases three revised estimates every January. So, the final data for 2020-21 would only be available in January 2024.

And, these are not just minor revisions. A BS analysis shows a 6 per cent difference between the first estimates of GDP at market prices released in 2007-08 and the third revised estimates released three years afterwards. The difference in gross value added for the year was an even higher 7 per cent. While this difference has decreased over time, the data is still subject to substantial revisions. For instance, for 2017-18, there was a 2.78 per cent difference between first advanced estimates and third revised estimates and a 4 per cent difference in gross value added.

The difference exists even between the first advanced estimates and preliminary estimates, even though both are released in the same year.

Moreover, the GDP growth and the extent of revisions are negatively correlated. So, in years the GDP is increasing, the revisions are smaller and in years GDP growth declines, the modifications are more significant.

In the context of the Coronavirus (Covid-19) pandemic, the provisional estimates may not really show the true picture of the economy, as data will keep showing substantial revisions until January 2024. This will also distort how we compute growth for coming years, as the base keeps getting revised repeatedly.

The problem does not end there. A deeper analysis of data reveals that the government falls prey to the same estimation errors every year. Barring just two years, in the last 13 years, the government has always under-estimated the growth in agriculture, forest and fishing. And, the problem has become worse over the previous three years for which revised estimates are available. While the revised estimates have, on average, come out 7.3 per cent higher than the first advanced estimates over thirteen years, in the last three years, the revisions have averaged 13.2 per cent.

The data also highlights that between FY18 and FY20, when the growth dipped, the government overestimated its value-added component (giving out much more optimistic estimates at the start of the year and then revising them subsequently). In all other years, it was underestimating the value-added numbers across categories.

So, mining and quarrying were overestimated by 12 per cent in FY20; manufacturing was overestimated by 6 per cent. Construction was overestimated by 5 per cent.

In all the years, save three, public administration, defence and other service spends have been overestimated.

The expenditure side throws up some interesting conundrums. While exports and imports have been estimated nearly perfectly by the government in recent years, its estimation of its own expenditures goes haywire each year. While it has improved its track record in estimating final private consumption—in FY20, the difference between the first revised estimate and first advanced estimate is just 0.01 per cent—in government expenditure, a five-year average shows an overestimation of 7 per cent.

A methodology change will go a long way in correcting these anomalies and addressing India’s GDP data concerns.