Argentina's shelved CPI basket shows a smaller 2024 disinflation than the official index

The government that would have lost the most from Argentina's new consumer price index is the one that cancelled it. INDEC, the national statistics institute, built a CPI weighted by the 2017/18 household expenditure survey and planned to publish it on 10 February 2026, starting with January data [1][2]. Days before that date, Economy Minister Luis Caputo stopped it. INDEC's director, Marco Lavagna, resigned [3]. Cabinet chief Manuel Adorni said the method "se va a cambiar cuando la inflación sea cero", that is, it will change when inflation is zero [4]. So the incentive is plain before any arithmetic. The weights decide how much of the inflation of 2023 to 2025 shows up in which year, and the government's political record depends on which year gets the credit.

I started from a different thesis and have to drop it. My working claim was that INDEC adopted the new basket in January 2026, that the basket cut the weight of regulated goods and raised services, and that this made the 2024 disinflation look larger than it was. Two parts of that are false. The basket was never adopted [2][3]. The weights move the other way: the new basket raises the share of utilities, transport and communications, which is where Argentina's regulated prices sit [1]. One part survives, and with the mechanism reversed it gets stronger. Under the weights the government refused to use, the fall in inflation from 2023 to 2024 is smaller than the official figure, and more of the disinflation lands in 2025.

Question

How much would Argentina's measured inflation in 2023 and 2024 change if the 2004/05 expenditure weights were replaced with the shelved 2017/18 weights? And does that change the official account that inflation roughly halved in Milei's first year?

How the number is built

Argentina's official CPI is a fixed-basket index. INDEC records prices for a list of goods and services and averages their changes. Each division (food, housing and utilities, transport and so on) gets a weight equal to its share of household spending in a reference survey. The current index takes its weights from the 2004/05 national household expenditure survey (ENGHo, Encuesta Nacional de Gastos de los Hogares) [1][2]. That survey describes how Argentine households spent money twenty years ago, before the long tariff freeze of the 2000s and 2010s. During the freeze, electricity, gas and public transport became cheap enough that households stopped budgeting much for them.

The 2017/18 survey reflects a later pattern, taken partly after the 2016 to 2018 tariff increases. The new index also moves to the COICOP 2018 classification (the United Nations classification of consumption by purpose) [5]. The changes reported for the national weights are these [1]:

Division	2004/05 weight	2017/18 weight	Change (pp)
Housing, water, electricity, gas, other fuels	9.4%	14.5%	+5.1
Transport	11.0%	14.3%	+3.3
Communications	2.8%	5.1%	+2.3
Food and non-alcoholic beverages	26.9%	22.7%	−4.2
Clothing and footwear	9.9%	6.8%	−3.1
Restaurants and hotels	9.0%	6.6%	−2.4

The six changes add up to +1.0 point. Weights must sum to 100, so the remaining divisions (health, education, recreation, household equipment, alcohol and tobacco, miscellaneous) must lose 1.0 point between them. I have not found that split published, so I treat it as an unknown and bound it below.

The pattern is the reverse of what I expected. The new basket puts more weight on utilities, fuel and fares, which are the prices the state controls. When a government is unfreezing tariffs, this basket shows more inflation. When it holds tariffs down, the basket shows less. Caputo used both sides of that argument. He said the gap was small: December 2025 measured 2.8% with the old method and 2.9% with the new one [2]. His team also argued that tariffs are still too low, so the new weights would overstate inflation once they rise, and that 2018 is a poor base year because instability began then [2]. The first argument says the weights do not matter. The second says they matter a great deal. Both cannot hold.

Data

• Division weights, old and new: Infobae's report of INDEC's methodology [1], confirmed by iProfesional [2].
• Annual inflation by division for 2024 (December 2024 over December 2023), from INDEC's release as reported by La Capital [6]: housing and utilities 248.1%, communications 185.2%, education 166.9%, miscellaneous goods and services 147.7%, restaurants and hotels 126.2%, alcohol and tobacco 122.4%, transport 120.9%, health 116.8%, recreation 108.4%, food 95.4%, household equipment 91.5%, clothing 85.9%. Headline: 117.8% [6].
• Annual inflation by division for 2023, as reported by iProfesional [7]: food 251.3%, household equipment 231.7%, health 227.7%, restaurants and hotels 219.1%, transport 187.7%, communications 186.4%, clothing 169.4%, housing and utilities 149.0%, education 141.7%. Headline: 211.4% [7].
• 2025 under the new weights: Gonzalo Carrera of the consultancy Equilibra estimated 32.2%, against the official 31.5% [8]. Fernando Marull estimated that January 2026 would have come in about 0.1 point lower under the new method [3].

The two years have opposite shapes. In 2023 the regulated divisions lagged: housing and utilities rose 149% while food rose 251% [7]. In 2024 the order flipped: housing and utilities rose 248% while food rose 95% [6]. That flip is the tariff catch-up after December 2023. It is also the reason the choice of weights decides how the two years compare.

Method

To a first order, swapping weights in a fixed-basket index changes the annual rate by the sum of each weight change times that division's inflation:

$$\Delta\pi \approx \sum_i \left(w_i^{new} - w_i^{old}\right)\pi_i$$

Here $w_i$ are the base weights as fractions and $\pi_i$ is division i's annual inflation in percent. I apply this to the six divisions with published weight changes. The missing −1.0 point I place on the remaining divisions, using the lowest and highest inflation rate among them to set bounds. I worked this out by hand, not in the Lab. Every input is in the tables above, so anyone can redo it in a spreadsheet in a few minutes.

Result

2024 (division rates from [6]). The six terms are: housing +5.1 × 248.1 = +12.65 pp; transport +3.3 × 120.9 = +3.99; communications +2.3 × 185.2 = +4.26; food −4.2 × 95.4 = −4.01; clothing −3.1 × 85.9 = −2.66; restaurants −2.4 × 126.2 = −3.03. Net: +11.2 pp. The missing −1.0 point, placed on divisions that rose between 91.5% and 166.9%, subtracts 0.9 to 1.7 pp. Reweighted 2024 inflation comes to about 127% to 128%, against the official 117.8%.

2023 (division rates from [7]). Housing +5.1 × 149.0 = +7.60; transport +3.3 × 187.7 = +6.19; communications +2.3 × 186.4 = +4.29; food −4.2 × 251.3 = −10.55; clothing −3.1 × 169.4 = −5.25; restaurants −2.4 × 219.1 = −5.26. Net: −3.0 pp. The missing point, placed on divisions between 141.7% and 231.7%, subtracts another 1.4 to 2.3 pp. Reweighted 2023 inflation comes to about 206% to 207%, against the official 211.4%.

	Official (2004/05 weights)	Reweighted (2017/18 weights)
2023	211.4% [7]	≈206% to 207%
2024	117.8% [6]	≈127% to 128%
2025	31.5%	32.2% (Equilibra) [8]
Fall 2023 to 2024	93.6 pp	≈78 to 80 pp
Fall 2024 to 2025	86.3 pp	≈95 to 96 pp

The disinflation is real under either set of weights. Inflation fell from above 200% to about 32% in two years, and no reweighting changes that. The reweighting changes when it happened. The official series puts the bigger share of the fall in the first year. With the newer weights, roughly 15 points move from 2024 to 2025. The 2024 fall shrinks from 94 points to about 79, and the 2025 fall grows from 86 to about 95. Ratios tell the same story. Under the official weights, the 2024 price-level multiplier (2.178) was 30% below 2023's (3.114). Reweighted, the gap is about 26%.

The reason is simple. The old basket gave little weight to the prices that rose most in 2024, because those were the prices the previous government had frozen, and it gave a lot of weight to food, which was the first price to slow. The official index therefore recorded the end of the freeze as if it were mostly a fall in inflation. The bill for the tariff catch-up existed. The 2004/05 basket just gave it little weight.

Sensitivity: which assumption moves the result most

The housing term dominates. It contributes +12.65 of the net +11.2 points in 2024. Take it out and the 2024 gap nearly vanishes. So the question that matters is whether the 5.1-point jump in base weight is the right size for December 2023, when the 2024 annual comparison begins.

That is where my first-order formula is weakest. In a fixed-basket index, a division's effective weight drifts away from its base weight as its relative price changes. A division whose price falls relative to the average loses effective weight, and one whose price rises gains it. After years of frozen tariffs, utilities were very cheap relative to everything else by December 2023. Their effective weight in the old index was probably below 9.4%. The same drift would shrink their effective weight in a new index too, if the new index were chained from a base set before the freeze. The base-weight gap of 5.1 points could therefore overstate the effective-weight gap at the start of 2024. I do not have the division index levels needed to compute the drift, so I bound it instead. If the effective housing gap at December 2023 were half the base gap (2.55 points), the 2024 adjustment falls from about +10 to about +4 pp, giving roughly 122%. If it were a quarter, the adjustment is under +1 point. My central range is 122% to 128% for 2024. On that range, 5 to 15 points of the official 2024 disinflation are a weighting effect.

Two smaller assumptions move less. The missing 1.0 point is worth under 1 pp in either year. The move from twelve divisions to thirteen under COICOP 2018 [5] means my mapping of old divisions to new ones is approximate. Some items, such as financial services and personal care, change division. I doubt that is worth more than a point or two, but I cannot measure it from published material.

There is one check against reality. Equilibra's 2025 calculation used the actual new structure, and its gap is only +0.7 pp [8]. My formula should give a small gap for 2025 as well, because tariff increases in 2025 were closer to the headline rate than in 2024. If someone with the division data finds that my formula gives a large gap for 2025, the first-order method is broken and the 2024 figure above goes with it.

What this does and does not show

It does not show that INDEC manipulated anything. The 2024 index followed its published method. That method was designed in 2016 on a twelve-year-old survey, and INDEC's own staff spent years building its replacement. The problem sits one level up, in the institution. The weights are chosen by the agency, but the date a new basket starts is chosen by the ministry being measured. Argentina has been here before: INDEC was intervened in 2007, and the IMF formally censured its CPI in 2013. The 2026 episode is much milder. No price was falsified, and the shelved series was described in public and estimated by private economists within days [3][8]. But the order of power is the same. The government got its way and the director resigned.

The government's case deserves its strongest form. A basket from 2017/18 records spending in a year when Argentina was entering a currency crisis. Relative prices then were not an equilibrium, and a basket that overweights utilities will mechanically show more inflation during every future tariff adjustment [2]. That is a real objection. But it argues for a newer survey, and the ministry has said it wants one [2]. It does not argue for keeping a 2004/05 basket, which is further from current spending than 2017/18 on every division in the table. Of the bases available, the twenty-year-old one has the least claim to represent how households spend now. It also happens to be the one under which the government's first year looks best.

If the argument above is right, the policy follows. INDEC should publish both series in parallel, monthly, starting from the earliest month it can backcast, and it should publish the division index levels so the effective-weight question can be settled with real data instead of my bounds. Some people would pay for that. The government would lose part of the 2024 headline. If indexed contracts and CER-linked peso bonds (CER is the inflation-adjustment coefficient tied to the CPI) ever moved to the new series, the Treasury would pay more during tariff adjustments and less while tariffs stand still. That conflict of interest is the reason the start date should not be the ministry's decision. What would change my mind is a published December 2023 effective weight for housing and utilities showing that the gap between the old and new effective weights was under 1.5 points. In that case my 2024 adjustment falls below 3 points, and the official account of a 2024 disinflation stands almost as published.