1\section{Introduction}\label{sec:intro}23Used cars are the largest durable good that most households trade, and the market that reallocates them is organised almost entirely through posted asking prices. A dealer lists a vehicle at a price, waits, cuts the price if the vehicle does not move, and eventually the listing disappears, either because the car was sold, wholesaled to another dealer, or simply withdrawn. Economists have studied the resulting equilibrium with transaction data from auctions and registration records \citep{genesove1993adverse,gavazza2014usedcar,larsen2021bargaining}, but the price-discovery process itself, the sequence of asking prices and the decision to cut them in the face of local competition, has remained largely hidden, because no public source records the daily asking-price paths of a whole market. This paper opens that black box for the Canadian province of Quebec.45We assemble a daily panel of 273{,}265 used-vehicle listings observed between 12 August and 1 November 2026 by Auto-Ka, an aggregator that re-reads the inventory pages of 131 dealer websites, two full-inventory portals and several dealer groups roughly seven times a day, logs every asking-price change, and records when a listing disappears from its source. The data are asking prices, not transaction prices, and listing disappearance is not a verified sale. We take these limits seriously throughout: every dynamic outcome is defined as an observed event on the dealer's website, and the paper's central methodological contribution is to show what can and cannot be learned about market liquidity from such listing data alone.67The empirical programme was preregistered before any model was estimated (Section~\ref{sec:method}; the full plan is in the replication package) and has six parts. First, we estimate hedonic asking-price models with flexible age and mileage profiles and ask whether depreciation is nonlinear and powertrain-specific (H1). Second, we model the hazard of a first sustained markdown and the size of the cut as functions of listing age, local competition and the vehicle's relative price (H2). Third, we estimate an event study of a dealer's repricing around the price cuts of close local competitors, with market-by-week and model-year-by-week fixed effects, treating the estimates explicitly as associations (H3). Fourth, we test whether quality-adjusted asking prices differ across Quebec's seventeen administrative regions (H4). Fifth, we estimate the hazard of listing exit and validate exit as a sale proxy using vehicles observed simultaneously on a dealer's own site and on a portal, and by detecting relistings of the same vehicle identification number elsewhere (H5). Sixth, we construct an Automobile Market Liquidity Index (AMLI) from five transparent components and validate it internally with a held-out-component design (H6).89Five findings stand out. (i) Depreciation of asking prices is strongly nonlinear in both age and mileage: the linear specification is rejected at any conventional level and splines improve dealer-grouped out-of-sample fit by about 3~percent, a gain that gradient boosting only modestly extends. Age profiles differ by powertrain, with battery-electric vehicles depreciating faster than comparable combustion vehicles at young ages. (ii) Markdowns are frequent but shallow: one listing in five receives at least one sustained cut, the median first cut is 2.8~percent of the entry price, and the daily markdown hazard rises steeply over the first three weeks of a listing before plateauing. Listings that enter above their hedonic value are cut more often and more deeply. (iii) Repricing is synchronised rather than sequential: a close competitor's cut is associated with a 0.5--0.9 percentage-point higher probability that the dealer cuts the same day, against a daily base rate of 1.4~percent, but the association vanishes on the following days and reappears only at seven-day harmonics; it survives dealer-by-weekday fixed effects at $k=0$ but not as a lagged response. A placebo that permutes competitor series within make--model cells produces no effect. We read this as common local shocks and shared repricing calendars rather than strategic reaction, and we do not claim causality. (iv) Quality-adjusted asking prices are remarkably integrated across Quebec: with make--model--trim fixed effects and dealer-clustered inference, the seventeen regional premia are jointly insignificant ($p = 0.53$) and all lie within $\pm 4$~percent of Montréal. This is a precisely estimated null. (v) Listing exit is strongly predicted by mileage, price and listing age, and exit is a reasonably consistent event across sources: for vehicles observed both on a dealer's site and on a portal, 80~percent of exits coincide within one day and 94~percent within a week; 10~percent of exits are followed by a relisting of the same vehicle within 30~days, 6~percent at a different location, which bounds the share of non-sale exits from below. The AMLI's first principal component explains 55~percent of component variance, and an index built without the exit component predicts next-week exit intensity within regions. We are explicit that the 81-day window does not allow any test of the index's forecasting value for external economic variables.1011The paper contributes to three literatures. To the economics of durable-goods markets \citep{akerlof1970lemons,hendel1999adverse,stolyarov2002turnover,gavazza2014usedcar}, it adds listing-level evidence on how dealers search for the price at which a used car clears, the counterpart on the sell side to the buyer-search evidence of \citet{murry2020colocation} and the intermediary margins of \citet{biglaiser2020intermediaries}. To the literature on asking prices and time on market, developed mostly for housing \citep{knight2002listing,han2016askingprice,anenberg2016frictions,guren2018momentum}, it brings a market with far shorter spells, near-complete price histories and many close substitutes. To the measurement of market liquidity, where industry days'-supply and days-to-turn statistics \citep{cox2022dayssupply,cbb2026index} exist but are opaque, it offers a reproducible, component-based index with documented validation limits.1213The remainder of the paper is organised as follows. Section~\ref{sec:background} describes the Quebec used-car market and the data-generating process of the aggregator. Section~\ref{sec:lit} reviews the related literature. Section~\ref{sec:data} documents the data, the quality controls and the sample construction. Section~\ref{sec:method} sets out the econometric framework and identification. Sections~\ref{sec:prices}--\ref{sec:amli} report the results on price formation, dealer dynamics and the liquidity index. Section~\ref{sec:robust} presents robustness checks and limitations, and Section~\ref{sec:conclusion} concludes.14