“A new car loses 20% the moment you drive it off the lot” is one of the most repeated numbers in American life, and the pages that print it rarely say where it came from. This page follows it to the documents that do exist, rebuilds a depreciation curve from a peer-reviewed study that publishes its method, and says plainly which federal numbers are measurements and which are tax rules.
Last updated .
Start with the page that most others copy. Kelley Blue Book's depreciation advice, dated 24 July 2025, says the first year takes “the most significant depreciation hit,” with “many vehicles losing about 20% or more of their original value,” then about 30% over the first two years, 8–12% a year after that, and “about 55% of their original purchase price within the first five years.” It does name a method, which is more than most: “Kelley Blue Book Private Party Values and Manheim Auction data.”
Read that sentence twice. Kelley Blue Book's own footer describes it as “a Cox Automotive Company.” Manheim's own about page describes itself as “a Cox Automotive brand” and “the world's largest automotive marketplace.” The valuations and the auction data behind that figure are the same corporate group, and the page carrying it also carries buttons for an Instant Cash Offer, a free dealer price quote and finding local dealers. So that figure is published by a business that sets the values, runs the wholesale market those values are drawn from, and sells the introduction to a buyer. That does not make it wrong. It makes it a number from an interested party, and it should be read as one.
Now the lender's version. LendingTree answers it in a single line of an FAQ: “New vehicles depreciate quickly, losing 20% of their value in the first year.” No source. Its table of slow-depreciating models is footnoted “Based on data from Kelley Blue Book.” Its answer on which cars depreciate fastest is attributed to iSeeCars, a used-car marketplace. Its figure for accident damage is attributed to CarFax, which sells vehicle history reports. Three attributed claims, three companies that sell car data, a headline figure attributed to nobody, and no analysis of its own behind it.
We looked for the figure in federal consumer material and did not find it there — not in a regulation, not in an agency fact sheet, and not attached to a study we could follow. Every trail we walked ended at a company selling something. That is the honest answer to “where does 20% come from,” and it is worth saying before the rest of this page, because the rest of this page is about the few places where a real number does exist.
Kelley Blue Book on beating depreciation; LendingTree on new-car depreciation; Kelley Blue Book's about page; Manheim on Manheim.
There is a peer-reviewed paper that measures this and shows its working. Laura Roberson, Saurav Pantha and John Paul Helveston, “Battery-powered bargains? Assessing electric vehicle resale value in the United States,” Environmental Research Letters 19(5), published 10 May 2024. Its listings come from 66,641 dealerships and run from January 2016 to March 2022; the models are fitted to the 9,015,324 of them listed up to the end of 2019, the authors having censored the pandemic years out of the estimates.
Its measure is defined exactly: a retention rate is the “listing price divided by the original MSRP based on the vehicle make, model, trim, and model year.” The model is log-linear in age, which means value is assumed to fall by a constant percentage each year rather than a constant number of dollars. For conventional gasoline vehicles the paper reports an intercept of 79.31% and an annual rate of roughly 9.7%.
Those two numbers are a curve. Multiply 79.31% by 0.903 once per year of age and you get the whole thing:
Two things fall out of that. The first year really is the worst, and it is worst for a specific reason: about 20.7 points of the 28.4 points lost by the first birthday happen at the moment of sale, and only about 8 points are a year of aging. The cliff is not the driving. It is the step from a sticker price to a used listing. And KBB's “about 55% within the first five years” and this rebuild's 52.4% are close enough that the valuation companies are not making the shape up.
Three caveats, because they matter more than the decimal points. The paper's data are listing prices, and the authors say so: reliance on them “may not fully capture the nuances of final sale negotiations and discounts.” A listing price is a dealer's asking price and a private seller will not get it. Second, the model was fitted to vehicles between one and eight years old, so the age-zero figure is the intercept of a line extrapolated back — nobody measured a car on its first day, not even here. Third, nothing in this curve describes the market after 2019, and the authors cut those years on purpose because prices stopped decaying: the same paper reports mean listing prices in March 2022 running 37% above January 2020 for conventional vehicles and 39% above for non-Tesla battery electrics, in inflation-adjusted 2019 dollars.
Roberson, Pantha and Helveston (2024) in Environmental Research Letters; the record and abstract are also at OSTI.GOV. For where the used market as a whole has been, see our 73-year used-car price chart — that page is about the market, this one is about one car.
Three federal numbers get quoted as though they described resale. None of them does, and it is worth being exact about what each one is.
The MACRS schedule. IRS Publication 946 puts cars in a named class: “5-year property. Automobiles, taxis, buses, helicopters, and trucks.” Its own worked example for a car bought in May 2019 runs the percentages out in full — 20.0%, 32.0%, 19.2%, 11.52%, 11.52%, 5.76%. That is a statutory declining-balance schedule with a half-year convention, not an estimate of anything a buyer would pay.
It is also, for what it is worth, exactly 20.00% in year one. We cannot show that this is where the famous figure came from and we are not claiming it did. It is simply true that the depreciation schedule a business actually applies to a car writes off precisely that much in the first year, and that the number is a convention rather than a measurement.
The luxury-auto caps. Revenue Procedure 2026-15 sets what a taxpayer may actually deduct on a passenger automobile placed in service in 2026: with the section 168(k) first-year allowance, $20,300 / $19,800 / $11,900 and $7,160 each succeeding year; without it, $12,300 / $19,800 / $11,900 / $7,160. Look at the second row. Without bonus depreciation the second year's cap is larger than the first year's, which no real car has ever done. That is what a dollar ceiling laid over a declining-balance schedule produces, and it is the clearest possible sign that these tables describe tax, not value.
The mileage rate. Notice 2026-10 sets the 2026 business standard mileage rate at 72.5 cents and then states, in section 4, that “the portion of the business standard mileage rate treated as depreciation is … 35 cents per mile for 2026.” The series is 26 cents for 2022, 28 for 2023, 30 for 2024, 33 for 2025, 35 for 2026. That is the federal government putting a per-mile price on depreciation — just under half of the whole allowance — and, unlike the industry figures elsewhere on this page, it comes from a body with nothing to sell.
What none of them says: what your car will fetch. They are cost-recovery rules. A taxpayer who drives 15,000 business miles in 2026 must reduce basis by $5,250 whether the car held its value or not.
IRS Publication 946; Rev. Proc. 2026-15; IRS Notice 2026-10.
The Bureau of Economic Analysis has to depreciate every vehicle in the national accounts, so it maintains a table of rates. Its method statement is the most useful sentence any federal agency has written on this subject: “BEA bases its depreciation patterns on empirical evidence of used asset prices in resale markets wherever possible. For most asset types, geometric patterns are used because the available data suggest that they more closely approximate actual profiles of price declines than straight-line patterns.”
Geometric means a constant percentage of what is left, every year, forever — which is the same shape the Environmental Research Letters model fits. For light trucks, including consumer-owned trucks from 1992 on, BEA prints a rate of depreciation of 0.1925 against a service life of 17 years and a declining-balance rate of 3.2725.
For autos, the row is blank. All four columns are dots. The footnote explains why: “Depreciation rates for autos are derived by BEA from data on new and used auto prices.” The one place in the federal statistical system where you would expect a single number for a car declines to print one, and says the rate is derived from price data rather than fixed.
Run the truck rate on a $40,000 pickup and the reason the curve feels like a cliff becomes obvious. The percentage never changes; the dollars collapse. Year one costs $7,700. Year two $6,218. Year three $5,021. By year six it is $2,644. Same rate, one fifth the money. Nothing dramatic happens at the end of year one — it is just the year with the most value left to lose.
BEA Depreciation Estimates (PDF).
Miles. The IRS figure above comes from a body with nothing to sell, and it is a tax convention rather than a market price — say that out loud before using it. Against it, NHTSA publishes what a car is actually driven. Its survivability and travel-mileage report estimates a passenger car covers 14,231 miles in its first year, 13,028 in its fifth and 11,193 in its tenth. At 35 cents a mile that is roughly $4,980, $4,560 and $3,920 of depreciation a year — the federal government's price for those miles, not the market's.
Body. Two federal documents agree that trucks are assumed to last longer. NHTSA puts lifetime travel at 152,137 miles over 25 years for a passenger car and 179,954 miles over 36 years for a light truck. BEA gives light trucks a 17-year service life and gives cars none at all. Neither document compares a car's resale curve with a truck's, so a longer assumed life is all this establishes.
Powertrain. The Environmental Research Letters model reports an annual rate of about 9.7% for conventional vehicles, 10.0% for hybrids, 14.5% for plug-in hybrids, 16.1% for non-Tesla battery electrics and 13.2% for Tesla battery electrics. A hybrid is, on that evidence, a conventional car for depreciation purposes.
Condition. We found no figure for the resale cost of accident history, wear or service records that was not published by a company selling history reports or appraisals, so this page does not carry one. That gap is real and it is worth knowing about: the thing buyers argue over hardest is the thing with the least independent published data behind it.
NHTSA, Vehicle Survivability and Travel Mileage Schedules (DOT HS 809 952, January 2006); IRS Notice 2026-10; Roberson et al. (2024).
This is the part of the subject most contaminated by interested parties, so here is the peer-reviewed estimate first. In the same Environmental Research Letters model, battery electrics start lower and fall faster: an intercept of 71.24% across all battery electrics, 68.45% excluding Tesla, against 79.31% for conventional vehicles, and an annual rate of 16.1% for non-Tesla battery electrics against 9.7%. Run both curves to three years and a non-Tesla electric is at about 40% of sticker where a gasoline car is at about 58%.
Tesla is the exception and runs the other way. The paper reports Tesla battery electrics “retaining 97.8% of their MSRP upon initial sale compared to only 68.5% for Non-Tesla BEVs,” and, unusually, earlier model years holding value better than newer ones. For other electrics the trend is improving: newer model years and longer-range cars retain significantly more than older, shorter-range ones.
Then the finding no marketplace would publish. A $7,500 federal subsidy on a new battery electric is associated with lower resale retention for the same model — 3.3% lower for non-Tesla battery electrics and 3.4% for Teslas. The discount does not stay with the first buyer; part of it is handed to the second one.
Against that, the industry figure. Kelley Blue Book's page says an electric vehicle “typically experience[s] a steep drop of 35%-40% in the first year before easing down to 45%-50% of their original value after five years.” Steeper than the academic estimate at year one, gentler at year five, published by a valuation company owned by an auction group, with no study cited. Both point the same way; only one shows its working.
One limit on all of it: the estimates are fitted to listings that stop at the end of 2019, and the paper's descriptive prices stop in March 2022. It cannot describe what used electric prices have done since, and nothing here should be read as though it does. Our guide to EV battery life covers the part of this that is about the battery rather than the market.
The widely-cited analysis of color and resale is from iSeeCars, a used-car marketplace — a company with a direct stake in used-car shopping. Its method is stated: over 1.2 million model-year 2022 used cars observed from August 2024 through May 2025, with each car's MSRP adjusted for inflation to 2025 dollars and compared with its list price, aggregated by color and body style.
The result: the average car lost 31.0% over three years, yellow lost 24.0%, orange 24.4%, green 26.3%, and gold 34.4%. The spread across colors is 10.4 percentage points, narrower than the 12.4 points the same study found in June 2023.
Now the problem, which is visible in the study's own table. Yellow depreciates 24.0% and that is $13,667. Orange depreciates 24.4% and that is $9,951. Nearly the same percentage, $3,716 less money — which can only mean the cars in the two buckets started at very different prices. The study aggregates by color and body style; it does not hold the model constant, so a color is standing in for whatever cars happen to wear it. Read as published it says certain colors appear on cars that depreciate slowly. It does not show that painting a car yellow slows it down.
Note also that the two best sources on this page disagree about the three-year number: iSeeCars reports a 31.0% average loss at three years, while the Environmental Research Letters model rebuilt above gives 41.6% for a conventional vehicle. Different samples, different years, different definitions of the starting price. Both are stated here rather than averaged into one figure that belongs to neither.
Every depreciation curve eventually runs into a floor it cannot go below, because at some point the car is priced as parts and metal rather than as transport. NHTSA's survival schedule says roughly when that population arrives: an estimated 78.7% of passenger cars are still registered at ten years, 50.9% at thirteen, 20.3% at seventeen and 2.3% at twenty-five. Somewhere around its thirteenth or fourteenth birthday, half of a model year has left the road.
Below that floor the percentage is meaningless and the inputs change completely: weight, the catalytic converter, which parts are worth pulling and what it costs to come and get the car. We have written that market out in full — what a junk car is actually worth, the scrap steel index under every junk offer, and what cars actually weigh. A twelve-year-old car with a dead engine is not on a depreciation curve at all.
And the honest answer on what a seller can do about any of this: almost nothing. Look back at what the sources above actually identify as drivers. Body style, powertrain, brand, trim and color are all fixed at purchase. The subsidy effect is fixed by policy. The market-wide swing is fixed by the market — our new versus used chart shows how much of it is nothing to do with your car. That leaves mileage, which is reduced only by driving less, and timing, which is the one lever genuinely in a seller's hand.
So the practical value of this page is not a tactic. It is knowing the shape: the largest single drop happens at the moment of first sale and is not recoverable, the percentage lost each year afterwards is fairly steady, and the dollar figures shrink fast because there is less left each time. A seller who knows that stops waiting for the car to stop falling, which it does not, and starts deciding when the remaining value is worth more to them than the car is.
Roughly, and one peer-reviewed study is where the evidence actually is. Roberson, Pantha and Helveston (2024) fit a model to 9 million used-car listings and report conventional vehicles retaining 79.31% of MSRP at age zero - a 20.7% drop. Two honest caveats: that is a listing price against a sticker price, not one transaction against another, and the model was fitted to cars aged one to eight, so age zero is an extrapolation. Nobody has measured a car on its first day.
About 9.7% of its remaining value a year for a conventional gasoline vehicle, on the same study. That is a constant percentage, not a constant amount, which is why the dollar loss shrinks fast: on a $40,000 vehicle at BEA's published light-truck rate of 19.25%, year one costs $7,700 and year six costs $2,644.
Not a single one. The Bureau of Economic Analysis publishes a rate of 0.1925 and a 17-year service life for light trucks, but leaves the auto row blank, footnoting that auto rates “are derived by BEA from data on new and used auto prices.” The IRS schedules - 20/32/19.2/11.52/11.52/5.76 under MACRS, and the dollar caps in Rev. Proc. 2026-15 - are cost-recovery rules for tax, not estimates of resale value.
On listings through 2019, yes, and the gap was large: an intercept of 68.45% and a 16.1% annual rate for non-Tesla battery electrics against 79.31% and 9.7% for conventional vehicles. Tesla ran the opposite way at 97.8% retention on initial sale. The same paper finds newer, longer-range electrics retaining significantly better than older ones, and its estimates stop at the end of 2019.
Very little, and the sources say so themselves. Body style, powertrain, brand, trim and color are decided when you buy. Federal purchase subsidies are associated with lower resale prices, not higher. That leaves driving fewer miles - the IRS values those at 35 cents each in 2026 - and choosing when to sell.