1\section{Data, measurement and descriptive evidence}\label{sec:data}23\subsection{Sources and coverage}45Table~\ref{tab:sources} summarises the 143 sources. Over the 81-day window (12~August to 1~November 2026, 586 synchronisation cycles) the platform observed 273{,}265 distinct listings, of which 62{,}821 were active on the snapshot date. Dealer websites contribute 33{,}061 listings from 131 dealers; the two full-inventory portals contribute 107{,}903 listings from 929 named dealers, with exact dealer coordinates and VINs for 82~percent of Otogo listings; dealer groups add 17{,}162. The sampled portals (103{,}027 listings) and Kijiji private sellers (11{,}874) are observed but excluded from the analysis samples for the reasons given in Section~\ref{sec:background}. Figure~\ref{fig:coverage} shows the daily stock and flows of the reliable-source universe, the crawl intensity, and the regional distribution of the active stock. After the two portals come online, the reliable stock is stable at 44{,}000--49{,}000 listings, with roughly 1{,}000 entries and 1{,}000 confirmed exits per day and a pronounced weekly cycle in both.67\begin{table}[tbp]\centering\caption{Data sources, coverage and sample membership}\label{tab:sources}8\tabsize\input{tables/tab01_sources}9\begin{minipage}{\textwidth}\scriptsize\smallskip Notes: Dealers = distinct dealer identifiers (dealer websites and groups are one dealer each; portals identify the selling dealer by name). Active = active on 2026-11-01. Found/cycle = mean number of listings returned per synchronisation cycle. S1 = hedonic sample; S2 = dynamic sample (Table~\ref{tab:sample}). Sampled portals are capped sliding windows and cannot reveal exits.\end{minipage}10\end{table}1112\begin{figure}[tbp]\centering\includegraphics[width=\textwidth]{figures/fig01_coverage.pdf}13\caption{Coverage across time, region and source. (a) Active listings, entries and confirmed exits per day in the reliable-source universe (S2L); the two spikes are the connector start-ups of Otogo (day~6) and AutoUsagée (day~11), whose stock is treated as left-censored. (b) Synchronisation cycles per day. (c) Active canonical listings by administrative region on 1~November 2026.}\label{fig:coverage}\end{figure}1415\subsection{Variables}1617\emph{Asking price.} The platform logs the entry price and every subsequent change. Price logs are contaminated by two artefacts that any listing-based study must address: placeholder prices (2{,}164 listings at \$1 or less; 5 above \$1~million) and \emph{oscillations}, in which a connector alternately returns two variants of the same listing (3{,}451 listings exhibit at least one A$\to$B$\to$A reversion, some twice a day for weeks). We therefore work with a \emph{sustained price path}: a price level that lasts less than 48~hours and reverts to the previous level is deleted, listings with three or more such transients are flagged as oscillating and excluded from markdown analyses, and a \emph{markdown} is a sustained decrease of at least 1~percent and \$100. Under these rules the price log yields 76{,}261 cuts and 26{,}486 rises across all sources; among reliable sources cuts outnumber rises nine to one.1819\emph{Vehicle age} is measured in years from 1~October of the year preceding the model year to the listing's entry date; \emph{mileage} in units of 10{,}000~km. \emph{Powertrain} (combustion, hybrid, plug-in hybrid, battery-electric, diesel) is read from the fuel field and, when missing, recovered from the title or from the make--model(--trim--year) cell when at least 90~percent of known listings in the cell agree (84{,}753 recoveries; 11{,}099 listings remain unknown). \emph{Body type} is normalised to eight classes and recovered from the make--model modal class when missing. \emph{Region} is the administrative region reported by the source, recovered from the city or from the nearest known city when missing (13{,}073 recoveries; 12{,}434 listings, mostly on excluded sources, remain without region). \emph{Dealer} identifiers are the source for dealer websites and the normalised dealer name within portal for portals; coordinates are the listing's, or the dealer's median location when missing.2021\emph{Listing age} is days since entry; \emph{vehicle age} is distinct. \emph{Exit} is defined as the last observation of a listing that is inactive at the snapshot and has been absent for at least seven days; exits within the final seven days of the window are right-censored, as are listings still active. The exit time is placed at the last observation plus half the source's median crawl interval, which treats exit as interval-censored between two crawls.2223\subsection{Sample construction and quality controls}2425Table~\ref{tab:sample} reports the attrition from the raw snapshot to the analysis samples. The hedonic sample S1 (136{,}393 listings) keeps canonical automobile listings from reliable sources with model years 2005--2026, mileage between 1{,}000 and 400{,}000~km, an entry price between \$2{,}000 and \$250{,}000, and non-missing make and model. The dynamic sample S2 (80{,}856 listings) further excludes the sampled portals and Kijiji, six connectors with documented failures or flicker (up to 94~percent of their exit events were later reversed), 53 oscillating listings, and the 47{,}235 listings that were already present on the first crawl of their source. The intermediate sample S2L retains the left-censored stock and is used as the competitor universe and for sensitivity analysis. Every excluded listing is written to an auditable exclusion log with the rule that removed it.2627\begin{table}[tbp]\centering\caption{Sample construction and exclusions}\label{tab:sample}28\tabsize\input{tables/tab02_sample_construction}29\begin{minipage}{\textwidth}\scriptsize\smallskip Notes: Sequential exclusions from the 2026-11-01 snapshot. S1 = hedonic sample (reliable sources); S2L = dynamic sample including the left-censored stock; S2 = dynamic sample used for hazards, competition and the liquidity index.\end{minipage}30\end{table}3132Three further quality controls are reported in the replication package: a per-source reactivation share (the fraction of exit events later reversed, which motivated the seven-day confirmation rule and the exclusion of flickering connectors), the stability of listing identifiers within sources (only 1.7~percent of Otogo VINs appear under more than one identifier, almost all genuine dealer-to-dealer transfers), and the distribution of crawl gaps (median 3.1~hours, maximum 13.6~hours).3334\subsection{Descriptive evidence}3536Table~\ref{tab:desc} reports descriptive statistics for S1 and S2 and by region. The median listing in S1 asks \$25{,}627 for a 5.7-year-old vehicle with 68{,}629~km; sport-utility vehicles make up 57~percent of listings, sedans 21~percent and pickups 11~percent; battery-electric vehicles are 6~percent and hybrids 4~percent. Completed listing spells in S2 are short: the median confirmed exit occurs 8.9~days after entry (interquartile range 1.0--20.6 days), 44~percent of completed spells end within a week and 87~percent within 30~days; 46~percent of S2 listings are right-censored. One listing in five (19.7~percent) receives at least one sustained markdown and 6.6~percent receive two or more; the median first cut is 2.8~percent of the entry price (mean 4.6; three quarters are 5~percent or less, 7.6~percent exceed 10~percent). Figure~\ref{fig:agecuts} displays these distributions and Figure~\ref{fig:paths} shows six representative price paths selected by a deterministic rule described in the caption.3738\begin{table}[tbp]\centering\caption{Descriptive statistics by sample and region}\label{tab:desc}39\scriptsize\input{tables/tab03_descriptives}40\begin{minipage}{\textwidth}\scriptsize\smallskip Notes: Entry price in Canadian dollars; vehicle age in years at entry; duration = completed spells (confirmed exits). Exit within 30~days is computed on S2 entrants observed for at least 37 days. ``Region missing'' = listings whose region could not be recovered.\end{minipage}41\end{table}4243\begin{figure}[tbp]\centering\includegraphics[width=\textwidth]{figures/fig03_listing_age_and_cuts.pdf}44\caption{Listing durations and markdowns (S2). (a) Empirical distribution of completed spell lengths. (b) Size of the first sustained markdown as a percentage of the entry price. (c) Share of listings with at least one sustained markdown, by source class.}\label{fig:agecuts}\end{figure}4546\begin{figure}[tbp]\centering\includegraphics[width=\textwidth]{figures/fig04_price_paths.pdf}47\caption{Asking-price paths of six anonymised listings. Selection rule: S2 listings with at least three sustained cuts and a duration between 30 and 70 days, in the three largest body classes; within each class the two listings with the lowest SHA-256 digest of their anonymised identifier. Only make, body and model year are shown.}\label{fig:paths}\end{figure}48