Crypto trades around the clock, but “Asian session,” “New York open” and “the best hour” are labels rather than complete research rules. Specify a timezone, dates, venue, market, bar boundaries and a measurable outcome before comparing intraday behavior. A clock pattern in a historical sample does not establish a current trading advantage.
What recent intraday studies do—and do not—show
A September 2026 paper in the Journal of Risk and Financial Management tests whether Bitcoin’s volatility clock moves with the U.S. cash market. It studies hourly Kraken XBT/USD data from 2016 through 2025. In its 2022–2025 subsample, the peak UTC hour shifts from 14:00 during U.S. daylight-saving time to 15:00 during standard time. On NYSE holidays, the measured U.S.-hours share of realized variance falls from 55.7% on matched weekdays to 41.9%. The reported clock-change effect is absent in 2016–2018, before the paper’s institutionalized period. The result is evidence from one dollar spot pair and one venue; the paper describes deeper drivers as associational, not as proof that institutionalization alone caused the pattern. Read the study and its methods.
An earlier study, published in the 2025 issue of Review of Quantitative Finance and Accounting and first available online in June 2024, analyzed 1,940 trading-pair and exchange combinations across 38 rated centralized exchanges on five continents. Its hourly sample ran from July 2018 through January 1, 2022; pairs needed at least 75% hourly coverage, and poorly rated exchanges were filtered out. Returns were highest between 15:00 and 16:00 UTC in that sample. Volume, transaction counts, volatility and the authors’ illiquidity measure tended to peak between 16:00 and 17:00 UTC, while their high/low-based spread estimator peaked between 15:00 and 16:00 UTC.
These studies are not a like-for-like replication: the exchange coverage, dates and outcome measures differ. Read together, they caution against treating one historical UTC average as a timeless “best session.” The earlier paper also averaged pair-level patterns without weighting by trading volume, so it describes that aggregation rather than a volume-weighted dollar of current activity. Its illiquidity measure is a Corwin–Schultz estimate built from hourly highs and lows—not a synchronized order-book spread or an executable quote for a stated order size. A volume peak, volatility peak, spread proxy and return are different measurements; more activity can arrive with worse execution conditions.
Convert a local session with its date
Keep raw observations in UTC, then convert named local windows with a geographical time-zone identifier such as America/New_York. The IANA time-zone database records location-specific clock changes; governments can change those rules. Do not replace a region with a permanent offset such as “UTC−5” when the study spans daylight-saving transitions.
For example, the UTC interval 16:00–17:00 corresponds to 11:00–12:00 in New York on 15 January 2026 (EST), but 12:00–13:00 on 15 July 2026 (EDT). The UTC window stayed fixed; the local clock label moved by an hour. Conversely, a fixed 11:00–12:00 New York session maps to different UTC intervals on those dates.
State whether intervals are half-open, for example [16:00, 17:00), and whether a bar label is its open or close time. If a provider labels bars by their close, a bar marked 16:00 may contain the preceding interval [15:00, 16:00); provider conventions differ. Repeated and missing local hours during clock changes need an explicit policy. Retain the original UTC timestamp and timezone identifier so another researcher can reproduce the conversion.
Choose one measure and a fair comparison
Write the question so the answer can be measured: “Does quoted spread differ by UTC hour on this venue for this contract?” is testable; “When is crypto most liquid?” is too broad. Define the instrument and price source, sample dates, bar duration, window boundaries, missing-data rule and comparison baseline before viewing results.
Match the measure to the claim. Candle volume measures activity, not order-book depth. Transaction count is not a price direction. A volatility estimator describes price variation, not the expected return. To study executable liquidity, collect synchronized book quotes at a specified depth and size; to study a strategy, hold entry, exit and position sizing fixed and include fees, spread, slippage and any funding relevant to the instrument. Report the sample count and uncertainty with the result.
Compare like with like across dates and market conditions. Decide beforehand whether weekends are included, how holidays and venue outages are treated, and whether pairs receive equal weight or are weighted by turnover. Record scheduled events rather than excluding inconvenient dates afterward. If you scan all 24 UTC hours, several local sessions, many coins and multiple outcome measures, preserve the search history: the best-looking window is selected evidence, not a fresh independent test. Check a later period or another venue before describing the pattern as persistent.
Before calculating, save a short study record: venue and contract; timezone identifier; exact local and UTC bounds; bar interval and timestamp convention; measure and unit; date range and weekday rule; data source and retrieval date; gaps or exclusions; aggregation weights; baseline window; and how many comparisons were tried. The day-trading plan helps freeze the strategy question, while the historical data guide covers timestamp coverage and bar integrity.
SuperQuant’s bundled hourly histories are synthetic examples. They cannot establish current venue activity, validate the published study on modern data or show a universal session edge. This is a method guide, not a recommendation for any trading hour.