Estimating is not fudging: Decoding the needless jump from 'GDP is estimated' to 'GDP can't be trusted'

Some non-experts contest India's GDP data - such as the recent release of Q1 data for FY27 - because they themselves don't feel conditions are good. That is a subjective assessment of what should be objective. But some experts have a more serious complaint: that too much of India's GDP is based on obsolete modelling rather than actual measurements.

There is just one problem. No country actually measures GDP. They all estimate it, not just India.

No country counts every haircut, samosa, consultancy assignment or vegetable sold. But India poses an exceptionally difficult statistical challenge because hundreds of millions work on farms, tiny shops, construction sites and in household enterprises that do not produce audited quarterly accounts. So, a significant part of GDP calculations is based on projections and imputations.


Non-experts will say, 'Don't confuse me. Just tell me how much of GDP is measured and not estimated?' Officials say privately that a rough decomposition of annual GVA suggests that 45-50% is based predominantly on relatively hard administrative data, company accounts and regulatory records. Another 30-35% is constructed substantially from current surveys or measured physical production. And 15-25% depends heavily on imputations, ratios, commodity flow calculations and other indirect methods. The boundaries between these are fuzzy.

Consider manufacturing. Large companies file accounts. Listed companies report quarterly results. GST supplies enormous amounts of sales information. Annual Survey of Industries covers factories. This is about as close as national accounting gets to measurement.

Formal finance is easier. RBI knows an astonishing amount about their deposits, loans and income. But it knows not much about huge informal lending.

Government output comes largely from government accounts. Electricity production is metered. Coal and steel production can be counted (save for illegal coal mining).

Now, consider your neighbourhood chaiwala. He produces no quarterly P&L account for MoSPI. Nor does your plumber, roadside barber or tiny garment workshop. Yet, GDP needs to somehow include their output.

Until recently, this was a serious weakness. Under the old 2011-12 GDP series, benchmark surveys of unincorporated enterprises and employment were used to project output for years or even decades.

That could be bad methodology after a major shock. Demonetisation, GST and Covid might devastate informal enterprises, while formal-sector indicators used as proxies behaved differently. Caution: do not take the official GDP figure for the demonetisation quarter too seriously.

MoSPI has addressed some of these issues in its new GDP series with 2022-23 as the base year, introduced in February. It now uses annual data from two new sources: Annual Survey of Unincorporated Sector Enterprises (ASUSE) and Periodic Labour Force Survey (PLFS). MoSPI says, 'Every year, direct estimates will be generated.'

Bravo. But notice the wonderful statistical oxymoron: 'direct estimates'. ASUSE does not count every paanwala in India. It surveys a sample, and grosses the results up to represent millions of enterprises. That is vastly superior to extrapolating an ancient benchmark, but it remains an estimate. Consider these two sectors:

Agriculture

We estimate acreage and crop production, multiply output by estimated prices, then subtract estimated expenditure on seeds, fertiliser, diesel, electricity and other inputs. This is calculated every year based on annual data, not extrapolated from an ancient benchmark. But it is still an estimate, and a rather iffy one.

Construction

How do you measure the value added by millions of contractors, masons and labourers? MoSPI now uses current cement production, current steel consumption and sundry other indices to estimate construction GDP. This is not a very accurate methodology. But it's still based on real current data.

Some other parts of GDP are simply imputed. If you live in your own house, you pay yourself no rent. Nevertheless, national accountants calculate the rent you might have earned from letting it out, and add that to GDP. Nothing was bought or sold. GDP nevertheless increased.

This sounds absurd until you consider the alternative. Two identical houses could contribute differently to GDP merely because one was rented and the other owner-occupied.

So, if 45-50% of Indian GDP is based on actual measurement, another 30-35% is deduced from current data, and 15-25% is just extrapolation from benchmark years or pure imputation, is that good or bad? For a developing country, that is fairly good. It can still be improved.

The story does not end here. We have so far focused on the accuracy of annual GDP. But most breathless TV debates take place over quarterly growth data. And quarterly GDP is much fuzzier.

MoSPI says explicitly that quarterly GDP uses the benchmark-indicator method. Annual GDP provides the benchmark. High-frequency indicators are then used to quickly estimate what happened quarter by quarter. These include GST sales, listed-company results, crop production, output of major manufacturers and minerals, railway freight, airline passengers, port traffic, bank credit, and government accounts. High-frequency indicators are not very accurate but are available quickly. They are revised significantly when full annual indicators are available. Do not waste much time on first reports.

Some critics jump from 'GDP is estimated' to 'GDP cannot be trusted'. Nonsense. Your blood pressure, unemployment rate and election opinion poll are estimates, too. What matters is whether the sample is good, proxies sensible, methodology transparent and revisions unbiased.