What followed crypto volume spikes across up to 90 days?

Current source-backed snapshot

A volume spike means hourly quote turnover reached at least twice the preceding 24-hour mean. Up-volume and down-volume events remain separate, while non-event hours enter only through the matched-control procedure. Forward returns, MFE and MAE cover the next 24 complete hours. Estimates require 20 complete events and carry moving-block bootstrap intervals; this is not the live alert model or proof that volume predicts direction.

Input documentation: Binance Spot candle field documentation · CoinNudge methodology

Observation window: Requested windows and actual coverage are recorded separately

Calculation cadence: Auto-refresh about every 21600 seconds

Calculation version: historical-evidence-1.2

Download current dataset: JSON · CSV · Matching candidates (7487) Raw values retain the dataset's published precision. Free fair-use limit: 60 requests per minute per IP, shared across all Research JSON and CSV endpoints.

Actual candle coverage

Requested history is not guaranteed coverage. Missing bars remain counted.

Observation evidence

First 30 of 404 observations. JSON and CSV contain all published evidence. CSV dataset_section distinguishes summary and observation rows.

Historical reconstruction is not proof of information available then or an executed return.

How to read this page

• Did unusual trading activity lead to continuation, or did the move fade?
• Six named Binance USDT spot pairs; requested 90-day hourly reconstruction.
• Read actual coverage and missing values before comparing outcomes.
• Download the evidence; distinguish historical reconstruction from locally recorded events.

What can this page tell you quickly?

Best for

Did unusual trading activity lead to continuation, or did the move fade?

Measured scope

Six named Binance USDT spot pairs; requested 90-day hourly reconstruction.

Update schedule

Recalculated every six hours; completed-window studies exclude the ongoing UTC day.

Do not infer

No guaranteed trading edge, complete point-in-time history or automatic entitlement to every data family.

Why split rising-price and falling-price volume spikes?

High turnover does not tell you who gained price control. A surge during a selloff is not the same condition as a surge during an advance. Combining them can hide opposite distributions and produce an average with little practical meaning.

The up/down split describes the event hour, while subsequent returns remain signed asset-price returns. A negative follow-up after a down-volume event is not automatically counted as a profitable short: there is no short execution model in this study.

Verify: Binance Spot candle field documentation

Why is the event excluded from its volume baseline?

Including the spike in its own reference average would dilute the anomaly and make the threshold partially self-referential. Using only the preceding complete hours makes the arithmetic auditable before looking at the future outcome.

The baseline still depends on recent activity and the hourly sampling choice. It differs from intraday seasonal normalization and from CoinNudge's live alert filters. A 2x ratio in a quiet session need not carry the same interpretation as 2x around a major scheduled event.

How should I test whether the observation adds value?

Compare each event group with its non-overlapping matched controls, then inspect the event-minus-control interval and adverse excursions instead of selecting only positive cases. Downloaded rows include pair ID, matching inputs, distance, threshold ratio, reference close and outcome status.

Fewer than 20 complete events are labeled exploratory and receive no outcome estimate. Even above that gate, the interval is descriptive rather than a validated edge; a stronger follow-up must predefine a future sample and include trading costs.

How can I audit and cite this measurement?

What is measured, and what is not?

Method and data boundary

Require continuous hourly history on six fixed Binance Spot pairs. Ratio equals current quote volume divided by the mean of the previous 24 hours, excluding the event hour; select ratio at least 2, split by price direction and space events by 24 hours. Statistics require 20 complete events. The 95% CI is a 1,000-resample circular moving-block bootstrap of the median with block size 3, published uint32 xorshift32 seeds and linear interpolation at (999 × q). Controls match symbol and UTC hour-of-week exactly, then minimize the published equally weighted normalized L1 distance over prior 24h return/5, prior volatility/2 and BTC prior return/5. Missing components cost 10; caliper is null; controls are selected without replacement in symbol/time order. Same-symbol event-window overlaps are excluded; ties use distance, absolute time gap, then earliest control time. The JSON links the complete candidate pool and a SHA-256 digest of its ordered observation IDs.

Six named Binance USDT spot pairs; requested 90-day hourly reconstruction. Historical results are descriptive, not executed profits or guaranteed future behavior.

Read the complete CoinNudge methodology

Sources and verification

• Binance Spot candle field documentationField definitions; statistical selection and outcome formulas are CoinNudge's own disclosed methods.
• CoinNudge methodologyShared boundaries, data time semantics and calculation contracts.

Frequently asked questions