Methodology

Every statistic is computed over the inventory TuLugar continuously monitors in each country: the listings published on the site plus residential inventory we track for statistical purposes only, which never appears in search results. That is why a page's "monitored listings" figure can be much larger than the browsable inventory — and why our medians rest on samples several times larger than what is publicly visible.

The base is cleaned before anything is computed: the same unit listed by several agencies counts once (group deduplication), listings with price or currency error signals are excluded, and each metric applies its own plausibility filters, detailed below. The counts printed next to each figure ("N listings analyzed") are always that figure's actual sample — the evidence behind the number, never the area's total inventory.

Since 6 September 2026 every count and every median is computed over unique units. In markets where non-exclusive mandates are the norm the same property is listed by two, three or more agencies, and where we follow it by more than one route the same unit reaches us repeated. Measured in Argentina, about a third of the inventory has an exact twin elsewhere in the base, and the large majority of those duplicates are the same unit in different agencies' hands, not re-posts.

There is one rule and it is applied in one place, before any statistic: two listings are the same unit when country, operation, property type, city, currency, exact price and floor area (rounded to the metre) all agree. Where no area is stated, the exact coordinate decides; never a generic office pin. Neighbourhood and bedroom count may not contradict each other: where they do, the listings are treated as distinct units, and where two groupings are possible none is chosen. Developments listing dozens of identical units are not collapsed: they are real units. One representative survives per group, the listing with the finest geography and the most complete data, and already published series are not recomputed: the basis change is dated here.

Almost all Latin American inventory is published in two currencies at once: the local one and the dollar. Whoever publishes picks which, and a share of listings arrives with the wrong label — a guaraní amount tagged as dollars, a Chilean sale tagged in Argentine pesos. Converted blindly, a $133,000 plot reads as a $529,000 one, and a single misread figure is enough to move a whole city’s median.

Since 8 September 2026 every price has one dollar reading, decided from two references: the currencies a listing may legitimately be quoted in for that country (legal tender, the indexed unit where one exists — Chile’s UF — and the dollar), and that country’s plausible price range, recomputed nightly from the data itself, separately for sale and for rent.

The rule has a single, conservative principle: believe the source unless what it says is impossible. If the amount is plausible in the stated currency, that is how it reads, even when the currency is foreign to the listing’s country — an Argentine border property quoted in guaraníes by a Paraguayan agency is exactly that. Only when the amount cannot be true as stated do we try another, and only if exactly one alternative is plausible and the two currencies are a factor of a hundred or more apart, so there is no room for confusion. Between close currencies we never reinterpret.

A mislabel is corrected; a typo is not. If a listing asks 2,600,000,000,000 guaraníes for a house, we do not divide by a thousand to fix it: the digits are wrong and we cannot know by how much. That listing leaves the price figures and stays in the inventory count. The same goes for one-dollar placeholder prices, and for cases where two readings are equally plausible: we would rather publish no figure than an invented one.

The price as the source published it is never modified: the dollar reading is a derived value, and each listing also stores the reason for its reading, so any published figure can be traced back to the decision behind it. Across the inventory of 8 September 2026, 99% of listings read in their stated currency, some twelve thousand read in another, and some nine thousand were left out of the price figures.

The median price per square meter (USD) of apartments for sale. Apartments only, and the reason is units: an apartment's m² is built interior area, while a house's usually includes the lot — for some sources it IS the lot. Mixing them does not produce a "more complete" average; it produces a units error. Houses and land get their own metrics, in their own units.

Filters: active sale listings with valid price and area (20–1,000 m²), a 200–8,000 USD/m² plausibility band (below is almost always a currency error; above is not a Latin American residential apartment), no detected price anomalies, grouped duplicates counted once. The index is published at country, region, city and neighborhood level — always the same definition, with minimum sample floors at every level.

Houses: the median TOTAL price in USD, not $/m² — a house's published m² mixes construction and lot depending on the source, so total price is the only honest unit, and it is also how the Latin American market itself quotes houses. Plausibility band: 10,000 to 5,000,000 USD.

Land: the median price per m² of the plot. For area we take the larger of the published area and the recorded lot size (the known data-entry errors always store values smaller than the truth, so the larger of the two is the best available estimate of the plot), require plots of at least 300 m², and apply a 1–2,500 USD/m² band that discards hectares mistakenly recorded as m².

For regions and cities, the reference rent is the TuLugar Rent Index (next section). For neighborhoods — where samples are too thin to mix-adjust — we publish per-bedroom-cut medians, each with its own minimum sample floor, alongside the per-bedroom breakdown where the sample allows. Monthly rentals only (short-term rentals have their own section with their own methodology), with a plausibility band and anomaly filters.

A raw median over all rentals together moves when the listing mix changes, not when prices change — a month with more 3-bedroom listings "raises" the average without a single rent going up. That is why we do not publish one.

Published on the price maps (regions and cities): a mix-adjusted typical rent in the methodological family of the Zillow Observed Rent Index (ZORI). It is computed over all monitored residential rentals (apartments and houses, monthly), grouped into six cells by type and bedrooms; each area's value is the weighted sum of its cell medians, with fixed weights reflecting the composition of the country's monitored rental stock. Every area is thus priced as if it had the same housing mix, so differences between areas reflect prices — not composition.

One difference from Zillow we prefer to state rather than hide: Zillow weights to census housing stock; we weight to our monitored stock, the best available proxy in markets without that public statistic. An area publishes its index only when its cells cover at least 60% of the weights with a minimum number of observations; where the sample falls short (most neighborhoods, for example), we publish the per-bedroom medians instead — the same data threshold decides, never a different methodology.

Sample floors: no figure is published with fewer than 5 observations, and aggregate levels require higher floors (a median computed over three listings is not a median). Where the sample falls short, the area shows no figure — we prefer a map with honest gaps to a complete map of weak numbers.

Published months are not rewritten, with two exceptions, and both are dated and explained on this page. A market surveyed by monthly census publishes its month from that census and republishes it if the census completes or is corrected. And a methodological correction that changes how already-published data reads is applied to the affected month, rather than leaving two bases side by side in one series: that is how Argentina’s and Paraguay’s August 2026 were restated on 8 September, on the corrected currency reading. Outside those cases a series does not change retroactively, improvements ship as new series with their date and definition, and downloadable datasets only ever add columns — existing ones keep their meaning.

Cadence: page figures are recomputed daily over the nightly snapshot; monthly series freeze on the 1st of each month.

Other methodologies

Not all TuLugar data lives on this page: each data product publishes its methodology where it is used. This is the index.