Is the price in the listings statistically reliable? Is the data sufficient to build a model?
In my market, there is no car valuation product like KBB, so I'm wondering about building it from the car classifieds data.
Hmm .. Interesting that you say your sector isn't being addressed. Wonder why ... But if so, then that's an opportunity.
I've spent the past 4 years specializing in probably THE hardest market for making accurate appraisals: the domain name industry. For me as an engineer, that has meant crunching numbers. So I'm intimately familiar with the issues involved in making inferences about value from partial list-price and sale-price data.
In fact, I spend much of each day compiling raw data and building database-driven tools to organize and analyze it.
So I don't know the auto industry at all, but I know a thing or 2 about figuring out the value of assets in obscure areas.
Certainly a model can be constructed for extrapolating value from list prices. It only remains to be determined how wide the margin of error would be and what kind of latent bias might exist.
If you'd like my help, just ask.
You would get more more accurate data from auction sites. Not sure about eBay, but there are usually local niched car auction sites on every market which do offer this service (valuations) to do generate traffic and business.
An example (sorry, you'll have to Google translate the Swedish) www.kvd.se
Listing sites try to do it, but then you have to control for not only on the spot price negotiations, but also that a share of listing drop in price and sometimes renewed under a new listing ID. Some of them publish these market prices as PR moves, but the value is really only relevant in comparing trends.
If you want access to data (and don't want to scrape or have access to search apis often built for the mobile apps) take a look at quikr who recently opened up theirs.