Underlying economic data, not forecasting errors, blight investment decisions

There’s a strong case to invest more in the collection of primary data.

Economic data continues to make news in more ways than one. The base year for GDP is being moved from 2011-12 to 2022-23 with improvements in estimations being targeted, including aligning them with the WPI, CPI and the IIP, a frequent source of discord in the past. The Consumer Price Index (CPI) is also being revised and base year moved from 2012 to 2024.

The new RBI governor is reportedly reviewing RBI’s internal forecasting tools for inflation and growth so as to minimise projection errors. These include adding a wide spectrum of data sets and the use of machine learning to pre-empt price fluctuations in volatile items such as food. Having taken flak for getting estimates wrong many times, this is understandable.

Underlying data is the real problem

But the larger problem is with actual economic data itself being flaky, not so much forecasting. World Economics, a data analytics firm, has a "C" rating assigned to India which means its GDP data should be used with caution.

Questions arising from change in GDP base year

The issues with GDP data are old and well known. Some of them were addressed during the previous base year change in 2011-12. An important change then made was the move from volume to value, with the method of calculating value added itself drastically changed. This became controversial as some indicators improved dramatically. In manufacturing, the change from IIP/ASI data to corporate balance sheets led to value-add growing faster than actual output.

Real GDP as a whole grew at a compound annual growth rate (CAGR) of 5.7 percent from fiscal 2011-12 to 2022-23, while IIP went up by only 3 percent per annum, raising questions of credibility. In services too, using indirect taxes to measure changes in value resulted in higher growth rates that seemed inexplicable.  But old shortcomings persisted.

The IIP, still in use as a high frequency volume series, had less than 20 percent value-based items even though GDP was itself a value-added measure. The IIP also failed to adequately capture the unregistered sector. Further, it did not take into account improvements in capacity or technological changes and its fixed weights (as of 2011-12) meant that it missed newer products such as smartphones.

The other crucial data peg is the Annual Survey of Industries (ASI) which has also been problematic. Annual GDP estimates use ASI data instead of IIP, which are used for quarterly estimates. The ASI is a product of the 1960s industrial period and covered only units set up under the Factories Act, which left out a large number of industries outside the Act.

Data for unregistered manufacturing (non-ASI) was collected only once in five years on a sample basis and the IIP was used for extrapolating the benchmark estimates.  Also, ASI data came delayed with a lag of 2-3 years which meant that GDP data also got constantly revised.

The famed double-deflator versus the single-deflator issue (which is currently used to deflate nominal data) is another known problem, but this has tended to be downplayed as being an acceptable feature and not a bug.

For at least two important segments of the economy, namely the informal sector and services, primary data availability has been poor, necessitating derivations based on productivity estimations and labour force data. For instance, in many of the service sectors, gross value added (GVA) is derived as the product of estimated labour force and value add per worker, the latter obtained from surveys, some done only once in five years. For the years other than the survey year, estimates were simply extrapolated using appropriate price and quantity (IIP) indices.

With the informal sector more than 70 percent of the economy and services contributing over half of total GDP, a significant chunk of GDP could be said to be “derived” actuals based on estimations from surveys. Clearly firmer data is required for the services and perhaps an Annual Survey of Services, on the lines of the ASI could help.

CPI’s issues are conceptual, not data-related

As regards measuring inflation, the CPI is a critical tool for both monetary policy and for deflating nominal GDP to real terms. But the considerations here are about conceptual issues, such as the treatment of free goods and services, rather than data. The Government seems to be clear about the intended usage of CPI when it says that being a measurement of inflation and an input into monetary policy, the scope of the index should be restricted to monetary transactions only.  Here it is worth understanding how inflation tracking works in other economies.

The US tracks retail inflation through two measures viz. the personal consumption Expenditure (PCE) index and the CPI. The PCE reflects changes in the complete prices of goods and services purchased by consumer, not just the out-of-pocket expenses, while the CPI deals with only actual out-of-pocket spends by consumers. In items such as healthcare, this difference can be significant with actual health care costs incurred by the consumer and the economy being different.

In essence, the CPI is based on a survey of what households are buying and the PCE on what businesses are selling. This also reflects in the different weights in the two indices. The important difference is in their purpose, one for macro policy, the other about living standards. The US Fed prefers the PCE to gauge inflation at a macro level as it reflects substitution effects better when prices change, while the CPI reflects changes in actual living standards of consumers. The recent debates in India over including food prices in CPI inflation when making monetary policy decisions, comes to mind. Perhaps having two indices could help, though some have argued that the WPI is in some sense similar to the PCE.