How the Cyclesite UK Bike Retail Price Index is calculated

Methodology last revised 2026-07-30. Revisions are listed in the changelog at the bottom of this page.

What the index measures

The index tracks the mean advertised (asking) price of used bikes in the UK, by category, by calendar month. It is an asking-price series: it measures what sellers advertise, not final sale prices. Asking prices react to market conditions faster than confirmed-sold data and are the figures buyers actually see; this is the same convention Autotrader uses for its published Retail Price Index. The median is computed alongside the mean as a robustness check and published in the CSV download.

Data sources

Observations come from used-bike listings advertised across the UK marketplaces we track: around two dozen sources, including the largest general marketplaces (eBay is the largest single contributor), specialist used-bike platforms, and Cyclesite’s own marketplace. Retailer RRP feeds, trade-in and wholesale buyback quotes, and non-UK sources are catalogued but excluded from the index panel: they are not asking prices on used stock.

Source coverage has grown over time. The largest expansion landed in March 2026, when high-volume marketplace scraping came online; the chain-linking step below exists to stop that expansion reading as a price move.

Inclusion rules

• Advertised or recently-delisted used-bike listings from tracked UK sources.
• Priced above £200, to exclude parts, frames and non-runner listings.
• Identified to a specific catalogue model. Listings our matcher cannot resolve to a known bike model are excluded, because unidentified listings skew heavily toward incomplete and junk entries and would drag the series down.
• Listings our data-quality gate rejects (spam, miscategorised, malformed) are excluded.
• Each listing counts once per month per source: repeated scrapes of the same listing are collapsed to their median observed price, so heavily re-scraped listings cannot over-weight the series.

Statistical treatment

• Winsorisation. Within each category and month, prices below the 2.5th or above the 97.5th percentile are excluded before aggregation, so a single £15,000 superbike or a mispriced listing cannot move the headline.
• Minimum sample. A category-month with fewer than 10 qualifying observations publishes no index point at all. Months with small-but-publishable samples carry a reduced confidence score.
• Margin of error. The page’s commentary treats a monthly move as flat when it is smaller than the month’s ±95% sampling margin of error (estimated from the month’s interquartile range and sample size). We do not narrate noise as a trend.

Mix adjustment and chain-linking

Two adjustments separate real price movement from changes in what happens to be for sale:

• Laspeyres mix adjustment. For each category we hold a reference basket of its most-listed models, weighted by their share of listings over a trailing 12-month base window, and compute a fixed-basket price alongside the raw mean. This isolates price movement from composition shift and is published in the CSV as laspeyres_gbp.
• Chain-linking across source changes. When the source panel changes materially (as in March 2026), months observed only under the older, narrower panel are rescaled by the ratio between the two panels measured over their overlap months (geometric mean across up to six overlap months). This is the standard overlap-chain-linking technique used by published price indices. The chart, the data table, and month-on-month changes all use this chain-linked series; the raw unadjusted mean remains available in the CSV as mean_asking_gbp.

Confidence and quality flags

Every published point carries its sample size, a quality flag, and a confidence score in the CSV download:

• confidence = min(1, n/30) × a dispersion penalty that scales down as the month’s interquartile range grows relative to its median (floored at 0.5). It is a heuristic ranking signal, not a statistical confidence level.
• quality_flag: reliable; low_sample (fewer than 10 observations); high_dispersion (IQR > 60% of the median, common in used-bike categories, where one label spans £250 hybrids and £8,000 superbikes); withdrawn (see revision policy; never published).

Revision policy

• The index is recomputed nightly over a trailing 24-month window, so recent months can be revised as late-arriving listings are ingested and matching improves. The live (incomplete) month is never published.
• A revision sentinel compares each recompute against the previously published value: any closed month whose mean moves by more than 2% is flagged internally for review before the change is accepted as an improvement rather than a defect.
• A category-month that stops meeting the publication bar is withdrawn from the series (it renders as a gap), but the previously computed row is retained internally: historical records are flagged, never deleted.
• Months that age beyond the 24-month recompute window are frozen as last computed.

Known limitations

• Asking, not sold. The series measures advertised prices. Achieved sale prices typically sit below asking; the gap varies with market conditions.
• Source concentration. eBay is the largest single source. The multi-source panel and chain-linking mitigate but cannot eliminate single-marketplace effects.
• No regional breakdown. Listings are UK-sourced but we do not yet capture location reliably enough to publish regional series.
• High within-category dispersion. A category mean summarises a very wide price distribution; use the median and interquartile-range (iqr_gbp) columns in the CSV when the spread matters to your question.
• Uneven category depth. Road has the longest, deepest history; smaller categories (folding, touring, kids) start later and publish intermittently when they clear the sample floor.
• Nominal prices. The series is not inflation-adjusted.

Changelog

How to cite

“Cyclesite UK Bike Retail Price Index”, Cyclesite, https://www.cyclesite.co.uk/bike-market-insights/price-index. The full series is available as a CSV download and is free to reuse with attribution.