The median collector car made nothing, and since 2021 it lost money

People talk about collector cars as an asset. The auction houses publish record results, the magazines print the big numbers, and somewhere in that conversation the idea takes hold that the cars go up.

We can test it, and not by comparing this year's batch of cars to last year's. We matched cars to themselves.

What we did

Every sale in our database carries a VIN where the auction house published one. We grouped sales by VIN and kept every car that sold twice, to a different buyer, at least 180 days apart. Where a car sold three or more times we used its first and last sale.

That gives 1,501 cars across 24 models, from a first sale in January 2016 to a last one in October 2026. For the money part below we use the 680 cars that sold both times at Bring a Trailer (BaT), because that is the one house whose full buyer and seller fee schedule we could read off its own published pages and apply to both ends of the trade with no guessing. On the wider 1,141 pairs where both houses' buyer premiums are known, the median comes out at minus 3.0% net. On all 1,501 with Bring a Trailer's fee schedule assumed throughout, it is essentially zero. Every cut of the data lands in the same place: positive before fees, at or below zero after.

This is the cleanest way to measure what a collector car did, because market composition cannot move the answer. It is the same car, sold to a different buyer, twice.

What it shows

On those 680 cars the median holding period was 2.1 years. The median car brought $66,250 the first time and $72,952 the second: up 4.3% before costs, or 2.3% a year. Just over half of them, 58.2%, sold for more than they had brought before.

Then we charged the fees, from BaT's own published schedule: 5% to the buyer with a $250 minimum and a $7,500 cap, and a $99 listing fee to the seller.

The median round trip costs 4.9% of the money you put in. Not 4.9% of the gain. Of the purchase price. In dollars, the median car paid $3,412 in fees across the two transactions, and the median car is a $66,000 car.

After that, the median car is down 0.7%. Up 4.3% before costs, down 0.7% after. The 95% confidence interval on that net figure runs from minus 2.6% to plus 1.2%, so the honest reading is not "cars lose money" but "cars do nothing, and the fees are why."

The individual-car version is the part people will feel. 396 of the 680 cars went up before fees. Only 329 went up after. Sixty-seven cars made money on paper and lost money in the transaction. That is one car in ten that a paper gain turned into a real loss, purely from the cost of buying and selling it.

A quarter of the cars lost more than 13.8% after fees and a quarter made more than 20.6%. There is real money made and lost in there. The middle of the distribution is just flat.

When you bought it is the whole story

This is the finding that surprised us, and it survives every check we could throw at it.

Split the same 680 cars by the year of the first purchase:

The early group had a 95% confidence interval on net return of +25% to +40%. The later group's runs from minus 6.3% to minus 2.8%. Those intervals do not overlap. The people who bought before 2021 did well. Everyone who has bought since 2021 has been treading water and paying fees to do it.

We checked whether that was just a change in which cars were selling, and it is not. Within each of the six models with enough pairs to test separately, the early buyers did better than the later ones: the Bronco by 28 points, the Mustang by 39, the 993 by 51. Six for six.

By exit year, the recent pattern is remarkably steady. Cars sold in 2023 came out at minus 3.6% net, in 2024 at minus 3.8%, in 2025 at minus 3.9%. Cars sold in 2026 were the first positive year since 2022, at plus 2.8%.

Where it was not flat

The average hides a real split, and the split is not the one people expect.

The cars that made money are the ones where a specific thing is scarce: an air-cooled 911, a 993, a 355 at the end of its run, an early G-Wagen before they became a fashion item. The cars that lost money are the ones that got popular, got restored in volume, and then got common.

The Bronco is the cleanest illustration of the problem this whole method exists to solve. Its model-level median sale price went from $41,525 in 2017 to $75,000 in 2026, which looks like a market that nearly doubled. The same trucks, resold, came out at minus 6.2%. Both numbers are correct. What changed was the cars: the Broncos crossing the block in 2026 are restored and modified in a way the 2017 ones were not, so the median rose while the asset did not.

The Mustang shows the same trap from another angle. Compare recent Mustang sales to older ones and you get minus 5%. Compare cars of the same model year and body style across the two periods and you get plus 12%. Both are honest arithmetic on the same data, and they disagree because the mix shifted: more coupes, fewer fastbacks, and the coupe is the cheap body. Match each Mustang to itself and you get minus 1.7% over 270 cars. That is the number, and it says the car went nowhere.

What this does not say

It does not say collector cars are a bad buy. It says the median car, bought at auction and sold at auction, is a flat trade, and that a car bought since 2021 has been a slightly negative one.

It also does not count what we cannot see: transport, storage, insurance, detailing, and the repairs every old car needs. Those are all on the wrong side of the ledger. Breaking even after auction fees is the optimistic version of this.

Two limits on the method. We only see cars that came back to auction inside our data, so a car sold privately in between, or sold once and kept, is not here. And a car bought in 2016 and sold in 2026 had a full decade to recover from a bad purchase in a way a car bought in 2024 has not, which is part of why the early cohort looks better and is worth remembering before anyone reads that table as a forecast.

How to use it

The useful thing here is not the average. It is the 4.9%: the cost of a round trip is roughly five percent of the car, every time, and it is the same whether the car costs $30,000 or $300,000.

That reframes the question. You are not asking whether collector cars go up. You are asking whether the specific car in front of you is cheap enough that a five percent round trip plus however long you hold it still leaves you ahead. That is a bid problem, and it has a number.

That is the job at BidBetter: read the comps that actually sold, adjust them for how this car differs, and arrive at the most you should pay rather than the most you could. A model page tells you what a car like it sells for. A report gives you the number for the one in front of you.

• What collector cars actually sell for, model by model
• How we value a car
• Value a specific car

Figures are medians of completed auction sales in US dollars after our cleaning rules, as of 1 November 2026. Repeat-sale pairs are the same VIN sold at least 180 days apart, to a different buyer; where a car sold three or more times, the first and last sale are used. Fee arithmetic uses Bring a Trailer's published schedule — 5% buyer's fee with a $250 minimum and $7,500 cap, $99 seller listing fee — applied to the 680 pairs that sold there both times. Cars sold at Barrett-Jackson are excluded from the fee figures because its seller commission is not published in a form we can apply consistently; on the wider 1,141 pairs where only the buyer's premium is known at both ends, the median net result is minus 3.0%. Estimates only, not appraisals, and not investment advice.