Small versus big data, to choose a used car

There are a few websites such as Edmunds that gather data to tell you the recommended price for a used car, based on brand, model, year, car history, and age - and maybe even zip-code since harsh Midwest winters can hurt cars. This is big data: collecting data from all used car sales in US, and computing trends and aggregates. This involves a bit of data plumbing - making sure that the data is accurate and correctly captured and summarized, in time. It does not involve much predictive modeling.

But what about a different, small-data approach, using just a few metrics, maybe the best predictors of car value: the color, age and brand. I bought a 6-year old used Honda Accord with a gold color (the only model available on-the-spot in my price range when I showed up at the car dealer - I guess nobody wanted it), and I was later told that the only people buying such cars are old ladies who are going to take great care of their vehicle. It was definitely a great value, and doing better than my wife's car - the same Honda Accord model (grey rather than gold), one year younger than mine, and brand new when bought. My wife's brand new car was priced $36K, my 6-year old used car $18k when I purchased it. Mileage is now similar on both cars, but the used car is doing better (much less repairs needed).

In short, can the color of the car be a better predictor of resale value, than anything else, given age, model, car history and mileage? For used cars only, of course, as color is a proxy for the kind of driver who own it before, which in turn is a proxy for how well the car was cared for. Obviously, people driving red cars are different from people driving gold cars. And in my case, I am a red/yellow car lover who don't mind driving a gold car, if it saves me thousands of dollars on the purchase price. Just thoughts for food.

DSC Resources