Used Car Price Data: Building a Valuation Model from AutoScout24 Listings
Anyone who values used cars needs used car price data that reflects what sellers actually ask today, not last quarter's guide value. Marketplace listings are the closest public signal: every ad has a price, a first registration date, a mileage reading and a spec. Put a few hundred of them in a table and a depreciation curve appears.
To show how this works, we pulled 100 Volkswagen Golf listings from AutoScout24 Germany and analysed them. You can download the same file from the free sample dataset page. If you want background on the platform itself, see our earlier piece on AutoScout24 automotive data.
Why listing data beats static price guides
Price guides are useful but slow. They are updated on a schedule, they average over regions, and they rarely tell you how many cars sit in each price band. Listing data gives you:
- Current asking prices, down to the trim and engine.
- Spread, not just averages. You see the cheap outliers and the premium examples.
- Segment detail. Fuel type, gearbox and power can be priced separately.
- Trend potential. Repeat the same search weekly and you have a time series.
The caveat is that asking prices are not transaction prices. Cars are usually sold for somewhat less than the sticker, and the gap varies by seller and segment. Treat listing data as a strong leading indicator, then calibrate it against your own sales results if you have them.
What our AutoScout24 sample shows
The sample covers 100 unique Golf listings from the first six result pages of the Volkswagen Golf search on autoscout24.de. We computed everything below directly from the CSV.
- Median asking price: €14,820 (mean €14,902). Prices range from €850 to €45,970.
- Median mileage: 65,617 km, with a mean of 86,267 km, so a few very high-mileage cars pull the average up.
- Price falls steadily with mileage. Cars under 25,000 km (12 listings) have a median ask of €22,717. For 25,000–75,000 km (44 listings) it is €17,305, for 75,000–150,000 km (26 listings) €10,845, and above 150,000 km (18 listings) €8,695.
- Age matters just as much. Registered 2024 or later (17 listings): median €22,985. Registered 2021–2023 (23): €17,400. 2018–2020 (16): €14,670. 2013–2017 (30): €10,470. 2012 or earlier (13): €6,450.
- Correlations are strong. Price correlates at 0.81 with registration year and −0.67 with mileage. A simple log-linear fit on this sample puts the price drop at roughly 6% per additional 10,000 km, though that figure overlaps with age effects and should not be read as a pure mileage effect.
- Automatic gearboxes carry a premium. 45 listings have an automatic gearbox, with a median of €17,220 versus €13,490 for the 55 manuals. Part of that gap is because automatics skew newer.
- Fuel mix: 77 petrol, 18 diesel, 4 petrol-electric hybrids and 1 fully electric car. Median diesel ask (€12,025) is below petrol (€14,790), but diesels in this sample are also older on average, so we would not call it a diesel discount without controlling for age.
A short excerpt (version text shortened):
| Version | First registration | Mileage (km) | Transmission | Asking price (EUR) |
|---|---|---|---|---|
| 1.4 Basis | 2002-09 | 303,000 | Schaltgetriebe | 850 |
| Plus (Klimatronic, PDC) | 2012-10 | 58,217 | Schaltgetriebe | 6,999 |
| Lounge BMT | 2016-10 | 79,000 | Schaltgetriebe | 12,999 |
| e-Golf VII | 2020-03 | 63,745 | Automatik | 16,990 |
| VIII 1.5 TSI OPF Life | 2023-02 | 29,600 | Schaltgetriebe | 15,950 |
The last row is a useful reminder: a 2023 car with 29,600 km sits at €15,950, below the 2020 e-Golf. Powertrain and trim shift the curve, so a model built on make and model alone will miss them.
A note on dealers versus private sellers
Our angle for this post was to compare dealer and private-seller prices. Our sample cannot do that: all 100 listings come from commercial dealers, because the default search ordering surfaced dealer ads on the pages we collected. We include a seller_type field so the comparison is possible, but we will not invent a gap that the data does not show. To study it properly you need a larger pull, filtered or stratified by seller type. We deliberately do not collect private sellers' names or contact details, only the seller type.
How to build a depreciation curve from listings
A workable first model takes about an hour:
- Clean the key fields. Convert first registration to a car age in months, keep mileage numeric, and drop listings with missing price or mileage.
- Segment first. Split by model generation, fuel type and gearbox before fitting anything. Mixing a 2005 Golf with a 2024 GTI in one regression gives a poor fit.
- Fit on log price. Depreciation is closer to a percentage decline than a fixed euro amount, so regress the log of price on age and mileage.
- Check the residuals. Large positive residuals are often trim or equipment effects (for example a GTI or GTE); large negative ones can be damaged cars or incorrect ads.
- Back-test against sales. Compare the model with your own transaction prices and apply a stable asking-to-sold adjustment.
With only 100 rows this is a demonstration, not a production model. A few thousand listings per model family gives much tighter estimates.
Use cases for car listings datasets
- Dealers and remarketing teams price trade-ins and stock against live competing ads.
- Leasing and finance companies estimate residual values and watch for early signs of falling prices in a segment.
- Insurers check replacement values against what comparable cars cost right now.
- Price-comparison and valuation apps show buyers whether an ad is high or low for its specs.
- Analysts and investors track used-car pricing as a proxy for demand, fleet turnover or the effect of new-model launches.
Which fields matter and why
- price_eur is the target variable. Check how the site treats VAT and conditional prices before comparing across sources.
- first_registration gives age. It is a better predictor than model year because it is exact to the month.
- mileage_km is the second pillar of any valuation.
- version, fuel_type, transmission, power_kw separate trims and powertrains that would otherwise blur together.
- seller_type and city support seller-segment and regional comparisons.
- listing_id and listing_url let you deduplicate across runs and track when an ad's price changes or the ad disappears.
Common challenges in automotive data scraping
Pagination limits. Search results typically stop after a fixed number of pages, so a single search cannot return a whole market. Splitting by make, model, year or price band is the usual workaround.
Anti-bot measures. Large marketplaces protect their pages. Reliable collection needs managed infrastructure, sensible request rates and rendering for pages that load data in the browser.
Duplicates and sponsored slots. The same car can appear on several pages or as a promoted result. Deduplicate on the listing ID.
Messy free text. Version names contain dealer marketing text such as equipment lists and service claims. Normalise them or map them to a trim taxonomy before modelling.
Layout changes. Sites change their markup and data structure from time to time, so parsers need monitoring. This is a large part of what separates a one-off script from a maintained feed.
Missing values. In our sample, one listing had no registration date because it was a new, unregistered car. Decide up front how to treat such rows.
How often should you refresh used car price data?
For valuation, weekly is a sensible default: prices move slowly enough that daily pulls add cost without much extra signal, yet monthly pulls miss promotions and stock turnover. Dealers who reprice stock actively may want daily runs for a handful of models. Keep every snapshot with its timestamp so you can calculate days on market and price cuts later.
FAQ
Is scraping AutoScout24 listings legal? We only collect publicly visible vehicle data, and we leave out personal data such as private sellers' names. Legal requirements depend on your use case and jurisdiction, so check the platform's terms and take legal advice for commercial use.
Can I get more than one model or the whole of Germany? Yes. The sample covers one search. A full dataset can span multiple makes and models, with the splits needed to get past pagination limits.
Are asking prices reliable for valuation? They are a good indicator of the market, but not the final sale price. Calibrate with your own transaction data.
What format do I receive the data in? CSV, ready for Excel, SQL or Python. Other formats and delivery options can be agreed for larger projects.
Get a sample for your own target
The file behind this article is free. Download the AutoScout24 sample and test your valuation approach on it. If you need a different make, region or source, request a free sample of your own target. Datasets from $49; see pricing. For more on why structured web data pays off, read web scraping for competitive advantage.
Author: DataHarbor Team