spb/wp22_uqo

1\section{Quebec transaction data and institutional setting}\label{sec:data}23\subsection{Institutional setting}45In Quebec every transfer of immovable property is registered at the land registry (\emph{Registre foncier du Québec}), and the deed records the price. Deeds are published shortly after registration and redistributed by data providers. Municipal property assessment is governed by the \emph{Loi sur la fiscalité municipale}: each municipality (or its MRC) deposits a triennial assessment roll that records, for every assessment unit, its use code (CUBF), number of dwellings, physical link, floor area of storeys, lot area, year of construction, number of storeys and the assessed values of land and building. The roll is open data and is updated with a lag relative to construction. There is no multiple-listing requirement: a substantial share of sales occur without a broker, so registry data cover more of the market than brokered-resale indices do.67\subsection{Source, snapshot and provenance}89The data are the transaction table of \emph{Vrai Prix} (www.vrai-prix.com), a platform that ingests daily the ``Transactions immobilières'' feed of the Journal de Québec / Journal de Montréal (api.qub.ca), which republishes land-registry deeds, geocodes them, and matches them to the provincial assessment roll by address. We took one immutable SQLite snapshot of the production database on 21 October 2026 (SHA-256 recorded in the configuration) and never read the live file again. The snapshot contains \NRawTx{} sales dated 4 January 2021 to 7 October 2026 and \NUnitsStock{} roll units, of which \NUnitsMatched{} distinct units are matched to at least one sale. The provider's structural fields are the roll's (year built identical in 99.2\% of matched sales, floor area in 94.1\%); its assessed value is the value in force at the sale date, which we use only for a screening rule. A full data dictionary and a generated provenance report accompany the repository.1011Four earlier extractions of the same table (6 and 9 August, 10 and 17 October 2026) measure the publication process (Table~\ref{tab:vintages}). In every vintage more than \SharePresentMin\% of the sales that the final snapshot dates on or before the vintage's last observed date were already present, no row was deleted and no price revised; the feed publishes with a lag of \PubLagMin{} to \PubLagMax{} days and does not backfill older months. Consequently the only incompleteness in real time is the final two weeks, and the real-time experiments of Section~\ref{sec:estimation} model a ten-day publication delay with a partial last month. The table has one date per sale; deed signing, registration and publication cannot be distinguished, a limitation shared with every registry-based index.1213\subsection{Coverage}1415Monthly volumes (Figure~\ref{fig:coverage}) start low (\NJanTwentyOne{} sales in January 2021, a feed start-up), average \AvgMonthly{} per month over 2022--2025 with a strong June--July peak tied to Quebec's moving-day calendar, and end with a partial October 2026 (\NOctTwentySix{} sales in seven days). Volumes in 2026 run 8--14\% below the same months of 2025. The provider collects no sale below \$50{,}000, a hard floor that truncates the lowest-price rural markets. Sales are geocoded, and every one falls inside a municipal polygon of the provincial administrative boundaries (SDA, version 2026-07); the roll's municipality code agrees with the spatial join for \RollSpatialAgree\% of matched sales. The series are on 2026 boundaries; municipalities constituted or merged during the sample are listed in the repository.1617\begin{figure}[htbp]\centering18\includegraphics[width=0.95\textwidth]{figures/fig_coverage_monthly.pdf}19\caption{Registered sales per month in the snapshot, January 2021 to October 2026 (all property types, before exclusions).}\label{fig:coverage}20\end{figure}2122\subsection{Classification and population definition}2324Property classes are built from the roll rather than from the provider label: a residential use code (1000) with one dwelling and an ``integrated'' physical link is a condominium; one dwelling with a detached, semi-detached or row link is a house; two to five dwellings a plex; six or more a multi-unit building; use codes 1100 and 1211 are cottages and mobile homes. The provider label is used only when the roll is missing or incomplete; conflicts between a complete roll record and the provider label (4.3\% of rows) are flagged and the roll class kept. The headline population is houses, condominiums and plexes with roll characteristics. Table~\ref{tab:exclusions} lists the exclusion rules in priority order. Sales dated before the roll's year of construction are lots or unfinished structures (the roll is a current vintage, so a lot sold in 2021 and built on in 2022 now carries the house's attributes); vacant land, non-residential use codes, parcels sold more than ten times in six years (developer or shared parcels), same-day sales of the same parcel at different prices (undivided co-ownership shares and condominium projects), multi-parcel deeds (same day, price and street or 1-km cell), exact duplicates and prices above \$10 million are excluded. The price screen is relative: a sale is excluded when its log price-to-assessment ratio deviates by more than $\log 4$ from the median of its year, class and region, using the assessment in force at the sale. This is not the only outlier treatment; every estimator below also uses a robust likelihood. The headline sample is \NHeadline{} sales (\ShareHeadline\% of the raw table), \NHeadlinehouse{} houses, \NHeadlinecondo{} condominiums and \NHeadlineplex{} plexes (Table~\ref{tab:sample_by_year}).2526\input{tables/tab_exclusions}27\input{tables/tab_sample_by_year}2829\subsection{Characteristics, missingness and what is not observed}3031The available quality variables are floor area of storeys, lot area, year of construction, number of storeys, building genre (bungalow, two-storey, split-level, mansard, one-module), physical link and, for plexes, the number of dwellings. Within the headline sample their completeness is high and stable over time (Table~\ref{tab:completeness}; the raw-row version in Table~\ref{tab:completeness_supp} shows lower shares in 2021 because land and unclassified rows are more frequent then), which is the condition for a time-dummy model: a variable whose missingness drifts creates artificial index steps. One provider field, the building style, was backfilled over the sample (12\% present in January 2021, 87\% in January 2025) and is banned from every model. Bedrooms, bathrooms, parking, basement finish, construction quality, renovations, waterfront, occupancy, financing and arm's-length status are not observed, and no image or text features are used. The roll's neighbourhood code turned out to be a unit-level identifier and is unusable; fine location is instead a nested metric grid (Section~\ref{sec:framework}). Characteristics are a single current roll vintage: renovations after the sale are invisible, and new constructions of 2025--2026 may still be coded as land, which lowers new-build coverage at the end of the sample.3233\input{tables/tab_completeness}34\input{tables/tab_mun_bands}3536\subsection{Geography}3738The hierarchy is province, 17 administrative regions, \NMrc{} MRCs and \NMunWithSales{} municipalities with at least one sale (Table~\ref{tab:mun_bands}); metropolitan communities (CMM for Montréal, CMQ for Québec) and boroughs are carried for aggregation and diagnostics. Statistics Canada census metropolitan area boundaries could not be obtained from the agency's site (an automated bot check blocked every download), so the metropolitan communities stand in for CMAs in external comparisons; the difference in coverage is documented. Within municipalities, location is controlled by the finest cell of a 500-m / 1-km / 2-km / 5-km grid in the NAD83 Quebec Lambert projection that contains at least fifteen sales of the class, so that no cell is a singleton and changing sale locations cannot masquerade as appreciation.3940\input{tables/tab_vintages}41