UQO Working Papers · WP36 · v1.0
The Hidden Dynamics of Used-Car Markets: Price Discovery, Inventory Liquidity, and Strategic Dealer Behavior
Evidence from a Daily Panel of 273,265 Quebec Listings
- Year
- 2026
- First release
- November 2026
- Latest revision
- November 2026
- Publication status
- Working paper
One used-car listing in five is marked down, the hazard doubles over three weeks, and neighbours reprice on the same day without reacting to each other
A daily panel of 273,265 Quebec dealer listings observed seven times a day for eleven weeks: nonlinear depreciation, median 2.8% markdowns on overpriced cars, regional premia within 4% of Montréal, exits that overstate sales by at least a tenth, and a five-component Automobile Market Liquidity Index that captures 56% of the common variation in regional turnover.
Abstract
The paper studies price discovery and liquidity in the Quebec used-car market with a daily panel of 273,265 dealer listings observed by the aggregator Auto-Ka between 12 August and 1 November 2026, with every asking-price change and listing disappearance recorded about seven times a day. Hedonic models with flexible age and mileage splines reject linear depreciation and improve dealer-held-out fit by three percent; profiles differ by powertrain. One listing in five receives a sustained markdown (median 2.8 percent of the entry price); the markdown hazard doubles over the first three weeks on the lot, and vehicles that entered above their quality-adjusted value or above their local competitors are cut sooner and more deeply.
A competitor event study with market-by-week and model-year-by-week fixed effects finds a same-day co-movement of price cuts among close local competitors (0.9 percentage points against a daily base rate of 1.4 percent) but no lagged response, and a within-make–model placebo yields nothing: this is interpreted as common shocks and synchronised repricing calendars, not strategic reaction. Quality-adjusted regional premia across Quebec's seventeen regions are economically small (all within four percent of Montréal) and jointly insignificant once trim is held fixed (p = 0.53), although coarser quality controls detect small but significant differences.
Listing exit is strongly predicted by mileage, price and listing age; cross-source matching by vehicle identification number shows that 94 percent of exits coincide within a week on two independent sources while at least 10 percent are followed by a relisting, so exit overstates sales. A transparent five-component Automobile Market Liquidity Index captures 56 percent of the common variation in regional turnover and, built without the exit component, predicts next-week exit intensity within regions under z-score standardisation, a result that weakens under rank standardisation; deep markdowns load as a symptom of illiquidity. No external forecasting claim is possible with an eleven-week window, and the paper documents what listing data can and cannot say about transactions.
Key findings
273,265
dealer listings observed about seven times a day, 12 Aug–1 Nov 2026
1 in 5
listings receive a sustained markdown, median 2.8% of the entry price
×2
markdown hazard over the first three weeks on the lot
0.9 pts
same-day co-movement of cuts among close competitors (base rate 1.4%), no lagged response
< 4%
quality-adjusted regional premia relative to Montréal, jointly insignificant with trim fixed (p = 0.53)
56%
of the common variation in regional turnover captured by the five-component liquidity index
Scientific contributions
- Nonlinear depreciation by powertrain
- Hedonic models with flexible age and mileage splines reject linear depreciation, improve dealer-held-out fit by three percent, and show profiles that differ across powertrains.
- Markdown dynamics on the lot
- A discrete-time hazard that doubles over the first three weeks; vehicles priced above their quality-adjusted value or above local competitors are cut sooner and more deeply, with the thirty-day predictability of repricing mostly reflecting dealer practice.
- Co-movement without strategic reaction
- A competitor event study with market-by-week and model-year-by-week fixed effects finds same-day co-movement and weekly harmonics but no lagged response, and a within-make–model placebo yields nothing.
- What exits do and do not measure
- VIN matching across two independent sources shows 94% of exits coincide within a week while at least 10% are followed by a relisting, so exit overstates retail sales; the median listing disappears after nine days.
- A transparent liquidity index
- Five components summarise more than half of the common variation in regional turnover; the non-exit components anticipate next week's exits under the preregistered z-score standardisation, less clearly under rank standardisation; deep markdowns load as a symptom of illiquidity.
Methodology
Methodology (5)
Hedonic splines
Flexible age and mileage splines by powertrain, evaluated on dealer-held-out samples against linear depreciation.
Discrete-time markdown and exit hazards
Hazards of sustained markdown and of listing exit as functions of lot age, mileage, price and mispricing relative to quality-adjusted value and local competitors.
Competitor event study with placebos
Market-by-week and model-year-by-week fixed effects, same-day and lagged responses, and a within-make–model placebo.
Regional price tests
Quality-adjusted regional premia with and without trim held fixed; joint significance tests across seventeen regions.
Preregistered liquidity index
Five-component Automobile Market Liquidity Index, built with and without the exit component, tested as a within-region predictor of next-week exit intensity under z-score and rank standardisations.
Research data
- Auto-Ka daily listing panel
- Every listing published by 131 dealer websites, two full-inventory portals and five dealer groups in Quebec, observed about seven times a day between 12 August and 1 November 2026, with each asking-price change and disappearance logged; immutable snapshot with an exclusion log.
- Cross-source VIN matches
- Vehicle identification numbers matched across two independent sources to date exits and detect relistings; 94% of exits coincide within a week, at least 10% are relisted.
- Regional and trim controls
- Seventeen administrative regions, make–model–trim and model-year identifiers used for quality adjustment, competitor sets and placebo tests.
Reproducibility
- Pipeline with immutable snapshot, exclusion log, preregistered plan and a claims ledger (CLAIMS_LEDGER.csv) that ties every statement to an experiment
- Decision log (DECISIONS.md) and experiment log (EXPERIMENTS.md) released with configs and figures
- Listing data are asking prices, not transactions; extension to registration transfers is specified as the next step
- Full LaTeX source of the 44-page paper
Research artifacts
Version history
| Version | Date | Description | File |
|---|---|---|---|
| v1.0current | November 2026 | Current manuscript as published on www.spboucher.ai (revised edition after the 2026-10-17 scientific review for papers 1–17). | PDF · 44 p. |