Coverage Windows
Coverage windows are time ranges that describe when data is available for a given symbol and data type. They are typically expressed as a start timestamp (earliest available) and an end timestamp (latest available) for trades, quotes, or order books. Coverage windows help you avoid requesting data outside of the period where the dataset actually contains events.
Coverage windows by data type
A symbol can have trades coverage but no quotes coverage, or quotes coverage that starts earlier than order book snapshots. That’s why robust pipelines track separate windows per data type rather than a single “data start/end.” When building strategies or analytics, align your required inputs (for example, quotes for spread) to the corresponding coverage window.
How to use coverage windows in analysis?
First, pick the data type(s) you need, then intersect your analysis period with each symbol’s coverage window for those types. For universe construction, you can require minimum coverage length (for example, at least 90 days of trades) or continuous coverage over the backtest window. Report coverage-driven exclusions so results are interpretable.
Are coverage windows the same as listing and delisting dates?
Not necessarily. Coverage windows describe the availability of data in a dataset, which may start after a symbol’s real listing date or end before a symbol’s actual delisting date due to collection or normalization constraints. Treat them as dataset-coverage boundaries, not authoritative exchange lifecycle dates.
How should you handle gaps inside a coverage window?
A coverage window typically describes the outer bounds, but gaps can still occur. For robust analysis, measure completeness (for example, expected vs observed quote updates) and decide how to treat missing intervals. For execution modeling, gaps may require excluding the symbol or switching to a different data type.
Why do coverage windows matter for avoiding lookahead bias?
If you filter symbols by “has data” using the full sample, you can leak future coverage into earlier periods. A point-in-time approach evaluates coverage as of each decision timestamp and uses only information available up to that time. This keeps universe selection time-consistent.
Real-world example
You compute 2020 bid-ask spreads for all spot pairs on an exchange. Many pairs have trades data in 2020, but reliable quote data only starts mid-2021. If you don’t respect quote coverage windows, you’ll end up with missing spreads and may incorrectly conclude those markets had no quotes.
Coverage Windows and CoinAPI Market Data API
CoinAPI symbol metadata exposes coverage timestamps (start/end) that you can use as coverage windows to bound your historical requests. Use these windows to (1) avoid empty queries outside coverage and (2) build point-in-time universes that reflect what data exists for each symbol and data type.