India's retail inflation has been back in the headlines since the RBI raised interest rates on 7 October, partly because it expects inflation to climb to 6% this quarter. If you build dashboards or write reports with CPI data, there is a trap to watch for: since February 2026, India's CPI comes from a new series with a new base year, a new basket and new weights. Mix the old and new series carelessly and your charts will show jumps that never happened. Here is what changed and how to handle the break correctly.
What changed in the new series
On 12 February 2026, the Ministry of Statistics and Programme Implementation (MoSPI) released the first figures on the new base, CPI (2024=100), replacing the 2012 base. January 2026 retail inflation came in at 2.75%. According to ThePrint and Drishti IAS:
Basket: 358 items (308 goods and 50 services), up from 299.
Classification: 12 divisions under the UN's COICOP 2018 system, instead of 6 groups.
Weights: derived from the Household Consumption Expenditure Survey 2023-24.
Collection: 1,465 rural and 1,395 urban markets, plus 12 online markets, with web-scraped data for items such as airfares and OTT subscriptions.
The weight shift is the part that matters most for analysis. Here are a few divisions, old weight versus new, from MoSPI's January 2026 release:
Food and beverages: 45.86% in the 2012 series, 36.75% in the 2024 series.
Housing: 10.07% before; now "housing, water, electricity, gas and other fuels" at 17.67% (fuel and light was a separate 6.84% group before, so this is not a like-for-like comparison).
Transport: 8.80% and health: 6.10% in the new series, both now standalone divisions.
Analogy: imagine a cricket team's batting average where the selectors swap a third of the players and change how much each innings counts. The average is still a batting average, but you cannot line up this season's number against last season's without some care.
Mistake 1: comparing index levels across series
Each series sets its own base period to 100. As the explainer at IndiaInflation puts it, "each series sets its own base period to 100, so the index levels are on different scales." An index value of 195 on the 2012 base and 104 on the 2024 base do not mean prices fell. They are measured from different starting points.
If you have a dashboard plotting the raw index, and it suddenly drops in January 2026, this is why.
The fix: linking
MoSPI handles continuity with a linking factor. Its January 2026 press release explains that the factor is calculated "to connect the two series over a common overlapping period", which is 2025, when both the 2012 and 2024 indices are available. Multiply old-series values by the factor and they are expressed on the new base. ThePrint reports that linking allows a back series to 2013.
A toy example with made-up numbers, just to show the arithmetic. Suppose that in the overlap year the old index averaged 200 and the new one averaged 100. The linking factor is 100 / 200 = 0.5. An old-series value of 180 from 2023 becomes 90 on the new base. Now 2023 and 2026 are on the same scale, and year-on-year rates computed across the break make sense.
In pandas, the idea looks like this:
import pandas as pd
old = pd.read_csv("cpi_2012_base.csv", parse_dates=["month"], index_col="month")["index"]
new = pd.read_csv("cpi_2024_base.csv", parse_dates=["month"], index_col="month")["index"]
lf = new.loc["2025"].mean() / old.loc["2025"].mean() # ratio over the 2025 overlap
linked = pd.concat([old.loc[:"2024-12"] * lf, new])
yoy = linked.pct_change(12) * 100Two cautions. First, use MoSPI's published linking factors and back series when they cover what you need, rather than your own estimate; the official numbers are what others will compare against. Second, compute linking separately for rural, urban and combined indices. They have different factors.
Mistake 2: comparing groups that no longer exist
Linking works cleanly at the headline level. Below it, the old six groups and the new twelve divisions do not map one-to-one. "Miscellaneous" in the old series (28.32% of the basket) has no single successor: its contents are now spread across several of the new standalone divisions, such as health, transport, and education services. If your report tracked "miscellaneous inflation" for years, there is no honest single continuation of that line. Say so in a footnote, start a new series from January 2025, and do not stitch them together.
Mistake 3: reading the headline as if food still dominated
With food at about 37% of the basket instead of 46%, a vegetable price spike moves the headline less than it used to, and services move it more. When you explain a monthly change, look at contributions (weight times division inflation), not just which division had the highest rate. A 6% rise in a 3% division adds less to the headline than a 3% rise in a 17% division.
A quick worked reading of August 2026
MoSPI's provisional release for August 2026 put combined inflation at 4.82%, with rural at 5.23% and urban at 4.31%, and food inflation (CFPI) at 5.95%. Notice that food inflation is above the headline, yet the headline is well below it: that is the lower food weight at work, along with other divisions rising more slowly.
A checklist for your CPI dashboards
Label every chart with the base year it uses.
Plot inflation rates, not raw index levels, across the January 2026 break, unless you have linked the series.
Use MoSPI's linking factors for rural, urban and combined separately.
Do not join old groups to new divisions. Start division-level series from January 2025.
When explaining a month's change, show contributions, not just rates.
Note in your methodology that provisional figures get revised.
Base-year changes happen roughly once a decade, which is exactly why they catch people out. A footnote and a few lines of code now will save you from explaining a phantom price crash in your next quarterly report.
