Where auto retailers make their money, and what the market pays for it

2026-11-03T00:00:00Z

The auto retail study puts all eight listed US auto retailers of any scale on one basis — six franchised dealer groups and two used-only retailers, $211 B of revenue and $118 B of market value between them. This is the short version: three things it establishes, and what each one is worth to somebody deciding whether to own any of them.

One: the group is not one group

The five franchised dealers a public buyer can still own trade between 6.3× and 7.5× EBITDA, on 7.6× to 9.7× trailing earnings — every one of them on a single-digit price-earnings ratio, and all of them within about a turn of each other. The sixth, Penske Automotive, is being taken private by its controlling shareholder, so its 11.5× is a transaction price rather than a market view.

The two used-only retailers are priced on a different basis altogether. CarMax is on 27.0× trailing earnings against the thinnest EBITDA margin of the eight at 3.5%, and Carvana on 31.0× EBITDA — the only multiple worth printing for it, because a deferred-tax release leaves it with no meaningful earnings multiple at all.

The takeaway. Any statement that begins "auto retail trades at" is already wrong, because this set runs from 6.3× to 31.0× on the same measure. The useful question is not what the sector is worth but which of these businesses you are actually buying, and the gross profit mix below answers that better than the sector label does.

Two: gross profit comes from the back of the shop

New vehicles are about half of a franchised dealer's revenue and roughly a fifth of its gross profit. Parts and service runs the other way: a tenth to a sixth of revenue, and 41% to 48% of gross profit. Finance and insurance is single-digit revenue and a quarter to a third of gross profit, because it is booked net of its cost.

Share of gross profit by line of business, latest fiscal year; “Finance” is a captive lender’s own margin.
Retailer	New	Used	Parts & service	F&I	Finance
GPI	21%	10%	44%	26%	—
ABG	20%	8%	48%	23%	—
AN	13%	9%	48%	30%	—
SAH	16%	8%	43%	33%	—
LAD	20%	13%	41%	26%	—
KMX	0%	68%	0%	16%	17%
CVNA	0%	59%	0%	41%	0%

Penske Automotive is left out of that table because it does not publish the same split; the used-only pair are in it, and they are the shape of the argument rather than an exception to it — selling only used cars empties the parts and service column, and used vehicles plus finance carry the whole of the gross profit.

The takeaway. Anybody ranking these businesses on revenue mix is ranking them on the half of the business that barely pays for itself. A dealer's earnings are a service annuity with a car dealership attached, and the right question about a new-vehicle sales number is not how big it is but how many cars it is putting into a service bay for the next decade.

Three: the windfall has mostly gone, and the share count hid it

Across the six franchised dealers together, revenue is up 82% since 2018. The EBITDA margin went from 3.9% in 2018 to 7.2% at the 2022 peak and back to 4.9% — 69% of the gain returned.

Earnings per share still compounded at 14% to 22% a year over seven years, and that is not a contradiction. It is the buyback. AutoNation has the slowest revenue growth of the six over that window and the fastest earnings per share, on a diluted count falling 12% a year.

The takeaway. The operating cycle and the per-share record point in opposite directions here, and which one you read decides whether these look like melting ice cubes or compounders. Both readings are honest. Neither is complete without the other, which is why the study prints three years and seven side by side rather than choosing.

The full workings — the basis, the adjustments that make floor plan and captive finance comparable across all eight, every chart and the downloadable workbook — are in the auto retail study. Figures are struck on the September 2026 quarter and priced at the 30 October 2026 close.

Why book value still matters for Berkshire, and only for Berkshire

2026-10-21T00:00:00Z

Book value fell out of fashion for a good reason. For most modern companies it measures the wrong things: it capitalises factories and ignores brands, software, research and customer relationships. A company whose value is its engineering team carries almost none of that value on its balance sheet, so the ratio of price to book tells you nothing except how capital-light the business is.

Berkshire is the exception, and it is worth being precise about why — because the exception does not generalise, and the moment you use the multiple carelessly you are back to comparing a railroad with a software company.

The reason is composition, not accounting

Roughly a quarter of Berkshire's book value is Treasury bills and cash. Another large slice is a portfolio of quoted equities carried at market value, marked every quarter by the market itself. For those assets, book value is not an estimate. It is the number.

That is what makes the multiple usable here. When most of what you own is marked to an observable market, book value becomes something close to a floor — not because accounting is generous, but because a Treasury bill is worth what a Treasury bill is worth.

Twenty-six years, and almost no help from the multiple

Here is the argument the rest of this note rests on, in one picture. Book value per share and the share price, from the 1999 year end to the latest close.

Book value per share and the share price Per B share, $ · 1999 to 2026

Book value runs to Q2 2026, the last quarter filed; the price runs to the close on October 22, 2026. The vertical gap between the two lines is the price-to-book multiple. Source: Berkshire annual reports and quarterly filings; year-end closing prices.

Book value per share compounded at 10.5% a year over 26.5 years, from $24.55 to $348.26 — a 14.2× increase. The share price compounded at 10.6%, from $36.60 to $546.97, a 14.9× increase.

Those two numbers are a tenth of a point apart, and that is not a coincidence. A share price is book value multiplied by whatever the market will pay for it, so the return a shareholder earns decomposes into exactly two parts:

price growth = book value growth × change in the multiple

The multiple did almost nothing. It was 1.49× book at the end of 1999 and it is 1.57× today — a drift of 5.4% in total, which spread over the whole period is about 0.2% a year. Set beside a 14.2× rise in book value, it rounds to nothing.

Which leaves only one candidate for where the money came from. Berkshire pays no dividend, so every dollar of profit it keeps lands in book value and stays there. Book value grew 14.2× because retained earnings were reinvested and the returns on them were retained in turn. Essentially the entire twenty-six-year return to a Berkshire shareholder is the compounding of retained earnings inside the company, not a re-rating of the stock.

That is a stronger claim than it may look, and it is worth saying what would falsify it. If the multiple had gone from 1.5× to 3×, half the return would have been the market changing its mind, and the next twenty-six years would start from a much worse place. It did not. Berkshire's shareholders were paid by the business, at roughly the rate the business earned, and they are not sitting on a multiple that has to keep expanding to make the record look good.

It is also why the multiple is worth quoting here at all. A price-to-book ratio is only informative when book value tracks the thing being valued. On this record it has — for twenty-six years, within five per cent.

Where it still understates

The operating businesses are the other half, and there book value does what it does everywhere else: it understates. BNSF sits on the balance sheet near its 2010 acquisition cost. GEICO's economics — a cost advantage built over fifty years — appear nowhere. Fifteen years of retained earnings reinvested inside the manufacturing businesses show up as plant, not as the earning power that plant produces.

So the multiple of book is a deliberately asymmetric measure: roughly accurate on the financial half, systematically low on the operating half. That is why the scenario range on the main valuation lands between 1.38× and 1.58× book rather than at 1.0×, and why the gap has been reasonably stable for twenty-six years.

How to use it

Use it as a cross-check, never as the valuation. The question it answers is narrow but useful: is today's price unusual relative to what the market has historically paid for this particular mix of assets? It cannot tell you whether that historical range was itself sensible.

For that you need the sum of the parts — and an opinion about what Berkshire does with its cash.

Intrinsic Value Per Share — Research notes