How to spot outliers in crypto liquidation data
An outlier is a liquidation value far from a time-matched baseline; clean the data, set a baseline, and cross-check each flagged value across sources.
Vahe HakobyanEditor-in-chief
Updated Oct 6, 20263 min readFact-checked
On this page
A large print may come from one venue or one delayed record. This guide shows how to separate a real market event from a reporting problem.
Before you start: data sources
Liquidation data comes from exchange APIs, aggregators, and on-chain feeds. Each source has its own delay, coverage, and definition of a liquidation. Check those details before you analyze.
Steps to spot liquidation outliers
Cleaning and baseline work decide whether a flag means anything. Keep the same market and time window for every comparison.
- 1Clean the raw recordsRemove duplicate fills, fill missing timestamps, and discard delayed or mislabeled entries. Use one time zone and one market pair.
- 2Match time windowsGroup records into equal windows, such as one minute or one hour. A quiet window and a busy window are not comparable.
- 3Set a baselineCalculate typical liquidation volume and frequency for each market from the cleaned windows. Keep separate baselines for each exchange and contract type.
- 4Score each valueCompute a z-score or an IQR deviation for every window. Flag values that sit far outside the usual range.
- 5Cross-check flagsCompare each flagged outlier with at least one other source. If only one feed shows it, treat it as a possible data error.
- 6Review market contextCheck whether funding, price moves, or open interest changed at the same time. This helps separate a real liquidation cascade from a bad print.
After you spot an outlier
A flag is a lead, not proof. Record what you found so another person can repeat your work, and protect the accounts that gave you access.
Frequently asked questions
Yes. Duplicate, delayed, or mislabeled records can create a value that looks extreme but did not happen. Cross-check every flag with another source before treating it as a market event.
You can start with a spreadsheet for a small sample. Coding helps when you need to clean and match many records across sources.
Spreadsheets and statistical software can calculate z-scores or IQR deviation. A scripting language can handle larger files, and exchange APIs or aggregators provide the raw data.
Review it on a regular schedule, and after any clear change in market conditions. A new contract listing or a shift in trading volume can make an old baseline stale.
