1\section{Institutional background and data-generating process}\label{sec:background}23\subsection{The Quebec used-car market}45Quebec is Canada's second-largest provincial vehicle market. New motor vehicle sales in the province were 41{,}657 units in August 2026 (Statistics Canada table 20-10-0085-01), and the used market is conventionally two to three times larger in units. Used vehicles are sold by franchised dealers, independent used-car dealers, dealer groups operating under a common banner, and private sellers. Almost every dealer publishes its inventory on its own website, typically through one of a handful of website platforms (D2C Media, SM360, AMVOQ's Theme Central, EvalAuto, SyncAuto) that expose structured listings, and most dealers simultaneously syndicate their inventory to portals such as Otogo (operated by LC Media), AutoUsagée (operated by the dealer association AMVOQ), AutoTrader.ca, CarGurus and Kijiji Autos. Prices are posted in Canadian dollars, usually before taxes and fees; Quebec's consumer-protection law requires the advertised price to be the all-in price excluding sales taxes, which limits the scope for hidden fees and makes posted prices comparable across dealers.67Transactions are not public. The provincial registry (SAAQ) publishes annual vehicle-stock files but no ownership-transfer series, and the only Canadian transaction-based used-vehicle index (the Statistics Canada used-vehicle CPI component, built from J.D.~Power dealer transaction data since June 2022) is national and monthly \citep{statcan2022usedvehicles}. Consequently, the asking price and the listing's presence on the dealer's website are the only high-frequency, vehicle-level signals available to a researcher, as they are to a buyer.89\subsection{Auto-Ka and the listing panel}1011Auto-Ka (\url{https://www.auto-ka.com}) is an independent aggregator of Quebec used-vehicle listings built and operated by the author. It maintains a registry of 143 sources: 131 dealer websites, 5 dealer groups, 2 full-inventory portals (Otogo and AutoUsagée), 2 sampled portals (AutoTrader and CarGurus), the private-seller section of Kijiji and two small portals. A dedicated connector per website platform reads each source directly, normalises each listing into a common schema (make, model, trim, model year, asking price, mileage, transmission, fuel, drivetrain, body type, colours, VIN when published, stock number, dealer name and location), and hashes the content. A synchronisation cycle re-reads every source; over the sample the cycle took 3.3 hours on average, giving 7.3 observations per listing per day, with no gap longer than 13.6 hours for any source. Each cycle classifies every listing as new, changed or unchanged, appends every asking-price change to a price log, and increments a miss counter for listings that are no longer returned by the source; a listing is declared inactive after two consecutive misses, unless a drift alert (a fall in the source's volume below a quarter of its recent median, or a jump in missing prices) suspends removals for that source. Listings carrying the same valid 17-character VIN on several sources are de-duplicated in favour of the dealer's own site.1213The research dataset is a read-only copy of the production database taken on 1~November 2026 after the last synchronisation of the morning, from which the user-account table was removed before transfer. No scraping was performed for this paper and no public API was used; the platform's data-governance rules, the field-level audit, and the restricted status of VINs, URLs, free text and exact coordinates are documented in the replication package.1415Three features of this data-generating process matter for the analysis. First, the panel starts on 12~August 2026 for dealer websites, on 18~August for Otogo and on 23~August for AutoUsagée; the stock present on each source's first crawl is \emph{left-censored} (its true listing date is unknown) and is excluded from every duration analysis. Second, a listing's disappearance is an observed event on the source website; it may reflect a retail sale, a wholesale transfer, a withdrawal, or a temporary feed failure. We therefore require seven days of continuous absence before counting an exit, and we validate exits across sources and against relistings in Section~\ref{sec:dynamics}. Third, two portals are deliberately sampled by the platform (AutoTrader returns a sliding window of at most 2{,}000 of more than 4{,}000 listings; CarGurus is capped at 500) and cannot distinguish a listing leaving the market from a listing leaving the sample; they are excluded from all dynamic analyses and from the baseline hedonic sample.16