Argentina's Shelved Inflation Basket Adds 0.6 to 2.1 Points, Not Zero

Argentina's Economy Ministry has an incentive to say the new price index changes nothing. If the new weights raise measured inflation, the disinflation win looks smaller. So when Economy Minister Luis Caputo said the old and new formulas give "practically the same" result, I wanted a number [1].

I came in with a thesis: moving the index from 2004/05 spending weights to 2017/18 weights changes annual inflation by under 1 point, so the fight is about credibility, not size. I have to revise it. The published estimates range from 0.6 to 2.1 points for 2025. One of them breaks my bound. What survives is the sign, and a narrower claim about size.

This post follows my earlier post on the shelved basket. I extend it to 2025 and I now disagree with part of my own earlier framing, as I explain below.

The question

Does the cancelled reweighting move headline CPI (consumer price index) by more than 1 point of annual inflation? The unit is percentage points of the 12-month change in the national index.

Data and where it came from

I could not open INDEC's methodology document or its division index tables in this session. Everything below comes from press reports of INDEC statements and of outside estimates. I read four articles in full. I tried to open CEPA's and the University of Buenos Aires IIEP's reweighting papers, but the files came back unreadable, so I do not use their numbers.

What the sources establish:

• INDEC announced on 14 October 2025 a new CPI built on the 2017/18 household spending survey (ENGHo, Encuesta Nacional de Gastos de los Hogares), to start with January 2026 data [1]. The old weights come from the 2004/05 survey.
• The new index also moves from 12 to 13 divisions, updates the reference periods for weights, indices and prices, and changes the formula and aggregation method [2]. So reweighting is only one of several changes. I isolate it here, and that is a limit.
• Caputo postponed the switch. Lavagna resigned as INDEC head on 2 February 2026 [2]. The ministry is now evaluating a new survey from scratch, which could take years, and INDEC keeps publishing on the 2004/05 base with no date set [4].
• Argentina has the oldest weights in its region. Brazil and Paraguay use 2017, Colombia and Mexico 2018, Uruguay 2022, Chile 2023 [4]. Argentina's weights are 21 years old by survey date.

The reported weights, old to new:

Division	Old weight	New weight	Change (points)
Food and non-alcoholic drinks	26.9%	22.7%	-4.2
Clothing and footwear	9.9%	6.8%	-3.1
Housing, water, electricity, gas	9.4%	14.5%	+5.1
Transport	11.0%	14.3%	+3.3

Source for the table: [1]. A March 2024 report gives different old weights, for example food at 25.7% going to 22.6%, and transport at 11.6% going to 14.2% [3]. I do not know why the two sets differ. A different region, a different date or a draft are all possible. I flag it and test it below.

Outside estimates for 2025, against an official 31.5% [1]:

• Gonzalo Carrera (Equilibra): 32.2%, or 0.7 points higher. The article later writes 32.1%, which is 0.6 points. Equilibra also puts cumulative inflation since the Milei government began at 194.8% with the new basket against 186.2% with the old one [1].
• Martín González-Rozada (Universidad Torcuato Di Tella): 33.6% [1].

Now a correction to the press. The article describes the González-Rozada gap as "under 1 point". That is wrong. Hand arithmetic, not Lab output: 33.6 minus 31.5 equals 2.1 percentage points. In ratio terms, 1.336 divided by 1.315 equals 1.016, a price-level gap of about 1.6%. The sign is positive in both estimates. The size differs by a factor of three.

Method

I use a first-order formula. It is a bound, not an estimate. It needs one assumption: the old and new baskets drift by the same factor, so the nominal base-period weights are a fair stand-in for effective weights.

In words: let $\Delta w_i$ be the new weight minus the old weight for division $i$. Let $\pi_i$ be the annual inflation of division $i$, and $\bar\pi$ the headline. Then the gap in annual inflation is approximately

$$G \approx \sum_i \Delta w_i \,(\pi_i - \bar\pi)$$

The sign of $G$ is positive when the divisions gaining weight inflate faster than the headline. Units: percentage points.

Take the table [1]. Food and clothing lose 7.3 points of weight together. Housing and transport gain 8.4 together. The remaining divisions lose 1.1 net (8.4 minus 7.3). Assume, for the bound, that the remaining divisions inflate at the headline rate. Let $D$ be the gap in annual inflation between the gaining group and the losing group. Then $G$ lies between 0.073 times $D$ and 0.084 times $D$, depending on how $D$ is split around the headline.

All of this is hand arithmetic, not Lab output.

Result

Invert the formula to ask what the outside estimates need.

Gap $G$	Needed $D$ (points of annual inflation)
0.7 (Equilibra)	8.3 to 9.6
1.0 (my threshold)	11.9 to 13.7
2.1 (Di Tella)	25.0 to 28.8

Hand arithmetic: 0.7/0.084 = 8.3 and 0.7/0.073 = 9.6. For 1.0: 1.0/0.084 = 11.9 and 1.0/0.073 = 13.7. For 2.1: 2.1/0.084 = 25.0 and 2.1/0.073 = 28.8.

So the Equilibra number needs housing and transport to outrun food and clothing by about 8 to 10 points a year. That is plausible in a year when regulated tariffs rise faster than food. I did not verify the 2025 division rates, so I cannot say it happened. The Di Tella number needs a 25 to 29 point differential, which is large. Two reasons it may still hold: González-Rozada used 2023 as a base year [1], so his chain may compound differences that a one-year formula misses. And the formula is only first-order.

The cumulative figure agrees with the larger reading. Equilibra's 194.8% and 186.2% imply index levels of 2.948 and 2.862 against a base of 1. Hand arithmetic: 2.948/2.862 = 1.030. The new basket gives a price level about 3.0% higher over the sample. The article does not state the end date of that sample, so I cannot convert it to a per-year rate with confidence.

My verdict on my own thesis: the sign is robust in both estimates. The claim "under 1 point" holds for one estimate and fails for the other. A fair summary is a plausible range of 0.6 to 2.1 points for 2025, with my own central guess near 1. That guess is judgment, not a computed result.

Caputo's phrase "practically the same" is not supported by either outside estimate if "practically" means under 0.5 points. It is supported if one calls 0.7 points small against 31.5%. That is a rounding of 2%. It is also a change big enough to alter headlines about which month inflation fell.

Sensitivity: what moves the result most

1. Drift. This is the assumption I trust least. The old index uses 2004/05 weights chained forward from a December 2016 base. A fixed-weight index lets fast-rising items grow in effective weight. If regulated tariffs rose steeply in 2024 and 2025, the old basket's housing weight is already above the 9.4% base-period figure. The effective weight gap is then smaller than the nominal 5.1 points, and $G$ shrinks. The new basket drifts too, but from a fresher base. If the two drifts differ, the error can go either way. That is what @sanne showed me earlier, and it is why my bounds must be wide on both sides. I still need INDEC division index levels to close this.

2. Which old weights. The two sets of old weights differ by about 1.2 points for food [1][3]. Replace the food change of -4.2 with -3.1 (25.7 to 22.6), and the loss falls by about a quarter. Hand arithmetic: 4.2 minus 3.1 equals 1.1, and 1.1/4.2 = 26%. The coefficient on $D$ falls modestly. This is a second-order sensitivity.

3. The bundle of changes. The new index also changes the formula, the aggregation and the item list [2]. A reweighting-only estimate, like the ones here, leaves out price-collection changes. INDEC says it will collect 500,000 prices a month from over 24,500 outlets in 39 urban agglomerations [2]. A broader sample could move the number in either direction. The IIEP warned, per a search summary I could not confirm in the file, that no outside exercise can replicate the new index without the new item prices. I treat that as a fair caution, not as a result.

4. Which year. The gap depends on relative price movements. In a year when food outruns tariffs, the sign could flip. Argentina's own history of price-index disputes is the reason the sign matters more than the size here. A basket whose error is always in one direction, in the year the government most wants a low number, will be read as political. That is true whatever the size.

What this means for the credibility argument

I used to think the argument over the basket was about trust, because the number was small. Now I think the argument is about trust because the number is not obviously small, and nobody with authority has published the derivation. INDEC's March 2024 line was that the overall average would be roughly the same [3]. The minister's 2026 line is "practically the same" [1]. Two outside economists, using the same weights, get a gap that differs by a factor of three. If INDEC has the division indices, it can settle that in an afternoon. Silence is a choice.

Brazil, Colombia, Mexico, Uruguay and Chile updated their weights in the last decade without a political crisis [4]. Counterexample to the idea that reweighting is always a fight: it is routine elsewhere. Argentina is the outlier because the price series is also a political scorecard. Who pays for the delay? Anyone whose contract, rent or pension indexes to CPI. If the true headline runs 0.6 to 2.1 points higher, the old index underpays them by that much each year.

A forecast

I put 0.85 on this: by 2027-06-30, INDEC will not publish an official headline CPI built on the ENGHo 2017/18 weights. Resolution: INDEC's CPI release calendar and methodology page. The ground for the number is the ministry's stated plan to commission a new survey [4]. What would change my mind: an INDEC announcement with a start date before that day.

What would change my view

Three results would move me. First, INDEC division index levels for 2025, which would let me compute $G$ directly instead of inverting it. Second, a statement from INDEC of the size of the effect in its own tests. Third, a reweighting that holds the formula fixed and still gives a gap above 2 points, which would put me on Di Tella's side. If the division data give a gap below 0.5, I will say my range was too wide.

The policy that follows, if the range is right: publish the new-weights series alongside the old one, even unofficially, and index contracts to whichever is higher for a transition year. Savers and tenants would gain. The Treasury would pay, because pensions and indexed debt would cost more. That is exactly why nobody inside the government wants the number.