Spread widening definition and what you are actually measuring
Spread widening is the increase in the bid–ask spread (or the effective difference between buy and sell prices) observed at the time an order could be executed. For assessment, it helps to distinguish two closely related ideas:
- Quoted spread: the spread visible from the provider’s/venue’s displayed bid and ask.
- Effective spread: the cost implied by what you could actually transact, which includes order execution and may differ from the displayed quote.
Because these differ, the minimum data required depends on which you intend to evaluate. If you only track displayed quotes, you measure quoted spread widening; if you track executed trades (or simulated executions with clear rules), you can assess effective spread widening.
Data inputs you need (and the assumptions you must state)
To assess spread widening in a self-contained way, collect inputs in four groups: inputs, provenance, timeliness, and quality checks.
1) Inputs (what quantities to record)
You need time-series or matched observations for the spread measure, plus the variables that can change it.
At minimum, record:
- Instrument identifiers (e.g., the currency pair) and market/venue context (how quotes are sourced).
- Spread observations: bid and ask at each observation point, or directly the spread value if you can verify how it was computed.
- Timestamp for each observation with consistent time zone handling.
- Order/execution context if you assess effective spread: trade timestamps, fill prices, and whether fills were marketable or requested at a limit.
If you plan to compare across time or sessions, also record:
- Trading session markers (for example, major overlaps) and any calendar-based grouping you use.
- Liquidity or volume proxies only if they are measured alongside your spread data, using the same time alignment.
Assumptions to state for any calculation or example: the observation interval, how spread is computed from bid/ask, whether you use mid-price normalization, and how you match spreads to execution times (for effective spread).
2) Provenance (where the data comes from)
Provenance determines whether your observations are comparable.
Collect and document:
- Data source: your quote feed, a platform export, or an external market data provider.
- Provider/venue behavior assumptions: whether the quotes reflect streamed pricing, delayed pricing, or aggregated/derived values.
- Method of extraction: whether values are recorded directly at source, averaged, or sampled after transformation.
3) Timeliness (how current and aligned the data is)
Spread widening is time-dependent. To avoid mixing effects from different moments, you need:
- Exact timestamps for every bid, ask, or execution record.
- A clear sampling rule (e.g., “record every N milliseconds” or “record at each received quote update”).
- A defined matching window for effective spread (for example, how close the quote timestamp must be to the execution timestamp).
If timeliness is unclear, the analysis can produce misleading “widening” that is actually a measurement artifact.
4) Quality checks (whether the data is fit for analysis)
Perform checks before drawing conclusions.
Examples of necessary quality checks:
- Missing data: gaps in bid/ask or absent fills; decide whether to discard or interpolate.
- Outliers: unusually large spreads caused by stale quotes, feed errors, or corporate/technical events.
- Consistency checks: confirm that bid ≤ ask and that spread calculations always follow the same formula.
- Time alignment verification: ensure there is no systematic timestamp shift between spread quotes and any execution records.
Evidence or example approach: compare spread before vs during suspected widening
A practical, non-prescriptive way to assess spread widening is to compare spread levels around periods you label as “wider” using only your collected data.
For example, you can:
- Compute a baseline spread statistic for each instrument over a defined earlier window.
- Identify later windows where spread is higher by your chosen threshold rule.
- Quantify the change using a consistent metric (such as average spread, median spread, or percentile-based measures).
To keep the example meaningful, you must state the threshold rule and window lengths and ensure they do not overlap in a way that biases results.
If you use effective spread, you also need explicit rules for how fills are obtained and how the spread measure is paired with the relevant quote time. Without that, “widening” can reflect execution behavior rather than pricing changes.