How a 24-hour change depends on its start time

Define the interval before calculating

For this study, simple percentage change is (end price / start price − 1) × 100. Both timestamps are UTC and each interval spans exactly 24 hours. A calendar-day return is a different convention unless its start and end happen to match.

The arithmetic does not establish how any provider samples or interpolates its historical values. A source’s “24h” label needs its own definition. Separate the conceptual example from claims about a live feed.

Use four explicitly fictional observations

Take invented prices at 12:00 and 18:00 UTC on January 1 and January 2, 2026. The first pair of same-hour observations defines Window A; the later pair defines Window B.

These are fabricated teaching values, not downloaded quotes, forecasts or reconstructed trading data. Sources & method links the four invented observations and the resulting two-row CSV so the calculation can be reproduced.

Evidence: Download the four invented observations (JSON)

Compare the two full 24-hour windows

Window A runs from $100 to $105 and returns +5%. Window B runs from $110 to $104 and returns about −5.45%. Both spans are 24 hours; their start and end clocks differ by six hours.

Meanwhile, the endpoint price only moves from $105 to $104, about −0.95%. That six-hour move is not Window B’s 24-hour return. The changing baseline explains why a rolling statistic can move much more than the latest endpoint alone suggests.

Do not treat a changed sign as a forecast

Neither sign predicts the next price. A large past percentage may reflect the selected baseline rather than a recent acceleration. This example assigns no likelihood to a continuation, reversal or profitable trade.

When comparing two displayed changes, align observation times, quote currency, interval definition and price source. Otherwise a difference can combine several unrelated causes rather than isolate a single market event.

Read retrieval time separately

The CryptoTrades market table distinguishes quote and retrieval times and labels stale or unavailable observations. Reloading a page is not evidence that the underlying quote has just changed.

The current snapshot is not a historical series for recreating these windows. An empirical study would need an authorized dataset, matching timestamps and a documented sampling method. No such live-provider study is claimed here.

Evidence: CryptoTrades data methodology · CryptoTrades guide to reading market data

Reproduce the synthetic windows

Check that both generated intervals are 86,400 seconds and that the start price is the denominator. Retain more precision for intermediate arithmetic than for the displayed percentage.

Download the observations and calculated windows from Sources & method. Pair each January 1 observation with the same UTC hour on January 2. Calculate (end price / start price − 1) × 100 and round to six decimal places for the CSV. Any excerpt should keep the synthetic label and the exact interval definitions.

Evidence: Download the two calculated 24-hour windows (CSV) · Download the four invented observations (JSON)

Sources & method

1. CryptoTrades data methodology

   Quote/retrieval times and cache behavior checked against the current site adapter. Checked 2026-10-29.

2. CryptoTrades guide to reading market data

   Existing educational context. The invented windows below are not a reconstruction of provider historical data. Checked 2026-10-29.

3. Download the four invented observations (JSON)

   Synthetic teaching values with explicit UTC timestamps, not provider or asset history. Checked 2026-10-29.

4. Download the two calculated 24-hour windows (CSV)

   Both intervals are 86,400 seconds. Generated from the four invented observations using simple percentage change; no market data was fetched. Checked 2026-10-29.

Continue learning

• Read a market table without forecasting
• Data source and cache methodology