How tauRank measures rankings by guest country
A ranking on Airbnb or Booking.com is not one number. It is the position a listing holds in one search, on one day, for one set of dates and guests, as seen by a searcher in one country. Change any of those and the position can change. This page sets out which of them tauRank holds fixed, which it varies on purpose, and what the city reports print as a result.
The search
tauRank runs the search a guest would run: the city, a flexible weekend for two guests, no filters. That is the unit every figure on the site reads. One unit, so a position in one city compares with a position in another, and a number read this week compares with the same number read last month.
The search runs once a day, from each country a customer tracks, through an ordinary residential connection in that country. The platform is asked nothing it would not be asked by a guest with a laptop in Munich or Manchester; it answers with the page a guest there would see, and we keep what it answered. We do not use any partner API for this, because none exposes ranking.
Why the country matters
Both platforms personalise search by where the searcher is. A guest in Germany and a guest in the United Kingdom looking at the same city for the same weekend are shown different orders, sometimes different listings, and prices in different currencies that are not always the same price. A host checking their own listing from home sees the home-country version only, which is the one version their foreign guests never see.
So the same search is run from several countries, and the gap between them is what tauRank reports: where a listing is for each country, and how far the countries agree.
How the city reports read it
A city page is built from the daily searches of the last seven days, and a platform appears on it only once it has been searched from at least three countries for at least seven days. Below that there is not enough to say anything.
For each pair of countries, two figures are read from the listings both countries returned:
- Agreement is Kendall's tau between the two orders, from 0 (no relation) to 1 (the same order). It counts, over every pair of listings both countries showed, how often the two countries put them in the same order. The page prints each country's agreement with the city's own country, because a host's home view is the reference they already know.
- Shared page one is how much of one country's top twenty is also in the other's top twenty, averaged over the days both were read.
Alongside those, the page prints what each country is shown: the median nightly price on its first page, and the median number of results the search returned.
What is not on the page
No listing is named. The reports are aggregates by design: a page republishes nothing the platform shows and hands nobody a list of properties to copy. A host who wants their own position tracks their own listing in the app; the city pages are for the shape of the market.
Nothing on a city page is modelled or estimated. Every figure is read from searches we ran, and the page says when it was last read. Our servers are in the EU, and the raw pages are kept so a figure can be re-read if the parser is corrected.
Caveats
A platform's order moves between two identical searches minutes apart, so a single reading is noise. tauRank reports over days for that reason, and the city pages use a week. Agreement is read over the listings both countries returned; a listing one country never shows cannot be compared, and shows up instead in the shared-page-one figure. And the unit is fixed: a city's ranking for a family of five staying a fortnight is a different ranking, and the city reports do not claim to describe it.