indianotes

On 12th February, the government announced a new consumer price inflation (CPI) series. The earlier CPI used 2012 as the base year; the new one uses 2024.

This was much overdue. Consumption patterns have changed dramatically since 2012. The old series still tracked stuff like VCRs and cassettes, which are pretty much extinct. Digital expenditure, such as OTT subscriptions or mobile data, was absent earlier. The new series is based on an expenditure survey conducted between August 2023 and July 2024, and better reflects how Indians spend today.

This is all to the good.

The media has discussed and dissected headline changes in weights – so I won’t repeat that. However, my random digging uncovered a few quirks about inflation measurement – which I thought might interest you guys.

Geometry vs Arithmetic

Prices of an individual item are collected from multiple outlets. The average price is not the arithmetic average we’d normally think of, but a geometric mean that a bloke called Jevons devised in 1871.

Let’s say two shops reported the price of sugar as ₹30 and ₹60, the arithmetic average would be ₹45, but the geometric mean works out to ₹42.4 [calculated as the nth root of the product of n numbers – in this case, the square root of 30×60].

This makes a substantial difference when numbers diverge wildly. If prices are ₹40 in both outlets, both the geometric and arithmetic means are ₹40. If one price falls to ₹20 and the other rises to ₹60, the arithmetic average is still ₹40, but the geometric one falls to ₹34.6.

Apologies for subjecting you to this random math, but wanted to illustrate how Jevons reflects price changes and helps smooth out isolated spikes.

Chain vs base year comparison

One of the biggest changes in the new series is the base for comparison. Under the old method, 2025 prices were still benchmarked against 2012 (base year). The new CPI uses month-on-month chain linking, so prices are compared with the previous month.

Hence, missing prices or one-off shocks affect only that month rather than permanently distorting the index. The earlier system allowed errors and outdated patterns to linger and bias inflation long after conditions had changed.

Quality improvement looks like inflation

Price collection struggles to reflect product improvements. For example, consider a car, say the Maruti Swift. Every year, there is incremental improvement; new features, electronics, safety or efficiency enhancements, etc. Some of the price increase is due to this. Yet, it is very difficult to isolate actual price inflation from the cost of new features.

This is the case for most manufactured products. Quality upgrades or new models often tend to exaggerate inflation.

Outlet selection can matter more than item selection

Typically, prices are collected from a fixed set of outlets or providers. This does not matter much in the case of products, where price differences across outlets are not too high. But in the case of services, the impact can be substantial.

For example, a haircut can cost less than ₹100 or many multiples of that. Price increases at one outlet may not sync with those at another. Depending on which barbers have been sampled, the inflation can look very different.

Consumption changes are not reflected

In January 2026, the price of silver jewellery went up 159% yoy. Many may have deferred their purchases (and even sold their silver jewellery). Yet the CPI assumes the same level of consumption, pushing up measured inflation – and this could persist until prices fall, or consumption revives.

This is not uncommon. When one vegetable becomes expensive, consumers switch to substitutes. When telecom data prices collapsed, usage surged. CPI weights don’t adjust for these shifts until the next expenditure survey. Even gradual changes, such as the disappearance of cassettes, can distort readings if base revisions are infrequent.  

_______________

Life is complicated, so there is no reason why calculating inflation should be simple. Added to the statistical limitations, the complexities of collecting representative data from such a large and diverse country are mind-boggling.

However, each iteration helps improve the measurement. Let’s hope that the next upgrade won’t take another 12 years!