Mostly neither: 12,688 of 18,897 pages — 67.1% — show the same price on the first and last capture we hold, however many years apart. Of the 6,209 that did move, 3,667 went up and 2,542 came down.
The finding that surprises people is the first one: most prices never move at all. Not down in a sale, not up with inflation — the same figure, years apart. 67.1% of every page we hold two or more dated captures for shows an identical price on the first and the last.
| Pages | Share | |
|---|---|---|
| Same price, first capture to last | 12,688 | 67.1% |
| Higher than it was | 3,667 | 19.4% |
| Lower than it was | 2,542 | 13.5% |
| Pages compared | 18,897 | 100% |
Among the pages that did move, increases outnumber decreases by 1.4 to one. This runs against the intuition that electronics get cheaper, and it is worth being careful about why — see the limits below. But the direction is not subtle, and it is measured on the shops' own pages, not on a manufacturer's list price.
We hold 58,702 dated captures across 21,875 product pages, the oldest from 2014-07-09 and the newest from 2026-08-31 — 12 years of other people's snapshots of shops' own pages.
| Year | Captures | Distinct pages |
|---|---|---|
| 2014 | 3 | 3 |
| 2015 | 3 | 2 |
| 2016 | 9 | 2 |
| 2017 | 12 | 11 |
| 2018 | 11 | 11 |
| 2019 | 742 | 711 |
| 2020 | 1,249 | 1,216 |
| 2021 | 1,937 | 1,755 |
| 2022 | 3,883 | 3,581 |
| 2023 | 6,363 | 5,832 |
| 2024 | 13,290 | 11,325 |
| 2025 | 26,927 | 17,615 |
| 2026 | 4,273 | 4,048 |
Depth matters more than breadth for this question. Of our 21,875 pages, 18,897 have at least two captures, 1,396 have four or more, and 185 have ten or more. Only the first group can answer “did it move”; only the third can hint at when.
The thin years at the top of that table are not a gap in the market, they are a gap in the Archive. Web crawling in 2014 was sparser and e-commerce pages were a lower priority than they became. We publish the shape as it is rather than starting the table where it looks tidy.
The chart everybody wants here is a single line sloping across a decade with a confident percentage at the end. We could draw it. It would be the most cited thing on this site and it would be wrong, for a reason worth stating plainly.
The pages we hold are not a sample of the market; they are a sample of what the Internet Archive happened to crawl. That crawl favours pages that were linked, popular or long-lived, and it thins out sharply the further back you go — the early years in the table above hold a few dozen captures where recent years hold thousands. A line drawn through that would show the Archive's growth as though it were the market's, and no amount of smoothing removes the problem.
What survives contact with that limitation is the comparison of a page against itself. That question needs no representative sample, no weighting and no assumption about what the missing pages would have said: either this page's price changed between two dated captures or it did not. It is a smaller claim than a trend line, and it is one we can defend line by line, which is the trade this site makes everywhere.
The same restraint is why there is no score out of 100 on any product page. With roughly one archived reading per product, a score would dress a single data point as a trend — so we publish the verdict and the evidence, and leave the arithmetic visible.
Mostly they do not move at all. Of 18,897 product pages we hold two or more dated captures for, 12,688 — 67.1% — show an identical price on the first and the last, however many years apart. Among those that did move, 3,667 rose and 2,542 fell.
Partly inflation, and partly an artefact we can name but not size: a product that is discontinued stops being captured, and products are often marked down before they are discontinued. Survivors are therefore over-represented, and survivors are the ones that held their price. We will not pretend to know how much of the imbalance that explains.
No. “The same price as in 2014” means the same number of dollars. In real terms that is a substantial fall, and the page that makes that argument with dated evidence is what costs less today than years ago.
58,702 dated captures across 21,875 product pages, the oldest from 2014-07-09. The thin early years reflect how sparsely the web was crawled then, not how the market behaved.
No, and it should not be read as one. An index needs a fixed basket weighted by what people buy. This is every page we happen to hold twice, each compared only against itself. It answers “did this page's price move”, never “did the market move by X%”.