How We Calculate Depreciation

Last updated July 2026

Every depreciation figure on this site is computed from real used-car listings — not estimates or manufacturer projections. This page explains exactly how the numbers are produced and where their limits are.

Data sources

Listings come from vehicles.dev, which aggregates active dealer inventory across the United States. For each listing we capture the VIN-decoded original base MSRP, the asking price, the odometer reading, the trim and powertrain, accident and owner history, and the seller's location. A typical model is backed by several hundred to a few thousand listings spanning roughly fifteen model years.

Expansion pages that are marked Observed 2021 use the Craigslist Used Cars and Trucks corpus assembled by Austin Reese and published on Zenodo under CC BY 4.0. We retain valid VIN-bearing U.S. listings from model years 2015-2020, deduplicate them, and take a deterministic sample of up to 120 listings per model-year. These pages are historical benchmarks and never describe the 2021 observations as current inventory.

Original MSRP

vehicles.dev rows use a VIN-decoded OEM base MSRP. Historical corpus rows do not contain that field, so they use model-year and, when a listing trim can be matched confidently, trim-level MSRP references from CarAPI. A trim match receives more weight than a model-year median; both receive less confidence than a VIN decode. The source and confidence travel with the computed record instead of being hidden.

Mileage normalization

Within a single model year, odometer readings vary enormously — a three-year-old car might show 20,000 miles or 80,000. Comparing raw median prices across years would conflate age with mileage. To prevent that, every listing's age is measured at its source observation date, then its price is adjusted to a standard 12,000 miles per yearfor the vehicle's age, using a slope of about 0.5% of value per 1,000 miles(0.75% for electric vehicles, which lose range-related value faster). The adjustment is capped at ±25% so a single extreme odometer reading can't distort a year.

Asking price vs. sale price

Listings show asking prices, which run slightly above final transaction prices. We apply a flat 5% asking-to-saleadjustment so the depreciation figures approximate what cars actually sell for, not just what dealers list them at.

The retention curve

After normalization, each listing's retention (sale-adjusted price ÷ original MSRP) is winsorized by capping extreme tails, then a single smooth curve — retention = e^(α + β·age)with β < 0 — is fit across every listing for the model. This guarantees the headline curve is monotonic (older always means more depreciation) and stable. Each model year's figure blends this curve with that year's own median, leaning on the curve when a year has few listings. A confidence score reflects sample size and fit quality.

Headline vs. per-year figures

The 1, 3, and 5-year figures we lead with come from the smoothed typicalcurve. The per-year table shows each model year's actual cohort, which can run above or below the typical curve depending on market conditions (a redesign, a supply shortage, or a used-price spike).

Limitations

  • MSRP references describe the factory configuration, not every installed option. Absolute dollar loss can therefore differ for heavily optioned vehicles.
  • Pages reflect the observation date shown on the page and move with active inventory. A small number of model-years still rest on the licensed 2021 corpus, and say so on the page. Neither cohort is a transaction-price forecast.
  • Very low-volume model years carry wider uncertainty; we surface a confidence indicator rather than hiding thin data.

External sources & standards

We combine market data with authoritative public datasets:

Editorial standards

Every figure on this site is computed from data, not opinion. We do not accept payment to alter depreciation numbers or rankings, we cite our sources, and we show the observation date so a historical benchmark cannot be mistaken for a current quote. Where a number is uncertain — a thinly-listed model year, for example — we surface a confidence indicator rather than presenting it as settled fact. This is information to help you research a purchase, not financial or investment advice.

Reviewed by CarResaleValue Editorial. Last reviewed July 2026. Found an error? Check the official source and let us know.

This site provides information, not financial advice. Always inspect a vehicle and review its history report before buying.