EconomyPrices
What is inflation's trend once the noise is set aside?
Measures of underlying inflation strip out the most erratic prices. In August 2026 they ranged from 3.3% to 6.5%, against a headline of 4.8%. In our test, core excluding food and fuel, now 4.4%, has been the best single guide to inflation a year ahead.
Written for readers who want to know what the RBI means by underlying inflation, and how much to trust each measure. CPI to August 2026.
Every reading, side by side
Each dot is one measure over one horizon. The median of them is 4.4%; they run from 3.3% (excluding food, fuel, gold and silver) to 6.5% (headline, over 3 months). Readings over three and six months are seasonally adjusted and annualised.
Underlying inflation by measure and horizon, Aug 2026
Over 3 and 6 months: seasonally adjusted, annualised. Over 12 months: the change from a year earlier.
How to read it. Some measures run below headline for long stretches. After three years on the new basket, that habitual gap is removed before a reading enters the median. None has three years yet.
Loading chart…
Built from 358 items
The new basket lets us rebuild underlying measures item by item, with MoSPI's own weights. The weighted median is the price rise of the item in the middle of the basket by weight: 3.5% in August 2026, up from 2.1% in January 2026. A trimmed mean drops the fastest- and slowest-rising items and averages the rest.
Underlying inflation from the new basket
Change over twelve months, from January 2026, the first month the new basket gives yearly changes.
How to read it. When the trimmed means and the median sit below the headline, a handful of items are rising very fast. Watch the gap close or widen.
Loading chart…
The same idea on the old basket, 2012 to 2025
Base 2012, whose finest published breakdown is 23 groups (twelve food subgroups, eleven others). Coarse: a group holds many items. Not joined to the chart above.
How to read it. The old basket gives a long history but a blunt tool: with 23 groups, one big group can be the median. It ended at 2.6% in December 2025.
Loading chart…
The gold effect
Gold and silver jewellery are 0.9% of the basket by weight, but their prices have run far ahead of everything else. Core inflation excluding food and fuel was 4.4% in August 2026; leave out gold and silver as well and it was 3.3%, 1.1 points lower.
Core inflation with and without precious metals
Change over twelve months, new basket, from January 2026.
How to read it. The gap between core with and without gold is what precious metals add. It says more about world gold prices and the rupee than about Indian demand.
Loading chart…
The items adding most and least to inflation, Aug 2026
Percentage points of headline inflation over the year: the twelve items adding most and the six adding least.
How to read it. An item adds a lot when its price rose fast, when it is a large part of the basket, or both. Rent adds through weight; onions and precious metals through speed.
Loading chart…
Momentum: the last three and six months
A twelve-month rate is slow to turn: it still remembers prices from a year ago. The last three months, seasonally adjusted and annualised, show where prices are heading now. In August 2026 the three-month rate was 6.5% and the six-month 5.0%, against 4.8% over twelve months. For core, the three-month rate was 6.1%.
Headline inflation over three, six and twelve months
Three- and six-month rates: seasonally adjusted with our own factors, annualised. Monthly from 2019.
How to read it. When the short rates sit above the twelve-month line, recent months are running hotter than the year as a whole. The three-month line is noisy: one month of vegetable prices can swing it by several points.
Loading chart…
Core inflation, three months against twelve
Excluding food and fuel. Seasonally adjusted with our own factors.
Loading chart…
How broad are the price rises?
Breadth counts how much of the basket, by weight, is rising faster than the RBI's 4% target or its 6% ceiling. In August 2026 25% of the basket rose faster than 6%, against 10% in January 2026. On the old basket's 23 groups, the median share above 6% was 32%.
Share of the basket rising faster than 4% and 6%
By weight. 23 groups of the old basket to December 2025; 358 items of the new one from January 2026. The two are not comparable: a group averages away its items' extremes.
How to read it. A rising share means more of what households buy is getting dearer quickly. A high headline with a low share points to a few items; a high share to inflation that is broad.
Loading chart…
Which measure says most about the year ahead?
Before computing anything, we wrote down a test: forecast headline inflation a year ahead with each measure's value today, from January 2017 to August 2025, and compare the errors. Core excluding food and fuel did best. No measure beat it, and the median of the readings did 20% worse, more than the 10% we predicted.
| Forecast | Typical miss (RMSE) | Average miss | Against core: t | p |
|---|---|---|---|---|
| Headline today (no model) | 2.31 pts | 1.88 pts | 2.79 | 0.005 |
| Excluding food and fuel | 1.61 pts | 1.28 pts | – | – |
| Weighted median, 23 groups | 1.94 pts | 1.54 pts | 1.06 | 0.290 |
| Trimmed mean, 10% each tail | 1.95 pts | 1.59 pts | 1.52 | 0.129 |
| Trimmed mean, 25% each tail | 1.87 pts | 1.50 pts | 0.97 | 0.330 |
| Median of the readings | 1.92 pts | 1.51 pts | 1.50 | 0.135 |
How to read this. Each measure's value today is used as the forecast of headline inflation a year later. The typical miss is the root mean square error, in percentage points. A positive t means the measure missed by more than core; |t| below 2 means the gap is too small to tell from chance (p above 0.05). Forecasts overlap, so the test allows for that.
Only headline itself, used as its own forecast, did clearly worse than core (t of 2.8). Without the Covid months the ranking holds: core's typical miss was 1.54 points, the median of the readings 1.76. The RBI's own study of 14 core measures (George and others, RBI Bulletin, February 2024) also found exclusion measures held up best.
The test uses the old basket's 23 groups, since the 358-item measures start in 2026 and have no outcomes yet. They will be graded once three years of outcomes exist. The test's rules are in the pre-registration log (C70).
What the numbers can't tell you
- Short history on the new basket
- MoSPI links only the headline index across the two baskets. Every other measure starts again in January 2026: until it has three years of history on the new basket it is not put on headline's footing, and the 358-item measures cannot be tested yet.
- Our seasonal adjustment
- MoSPI publishes no seasonally adjusted CPI. Our factors come from the eight years before each month, on the old basket. For headline that is safe: in 2025, when both baskets were published, their monthly changes moved together (correlation 0.996). For core it is not: the two baskets' monthly core changes were unrelated (correlation −0.15). Read core's three- and six-month rates on the new basket as rough until it has seasonal factors of its own.
- Groups are blunt
- On the old basket the finest published breakdown is 23 groups. A weighted median of 23 groups can sit on one large group for months, so the long history of these measures is coarser than the new one.
- A test is not a rule
- Core did best on forecasts made from 2017 to 2025. Another stretch with different shocks could rank the measures differently.
The workings
The question
Data
Every chart has a table view, a CSV download and a link to its published JSON. Bundles are listed in data status.