Direct answer
To assess pair spreads, gather (1) the definition and measurement rule, (2) the quote inputs used to compute spreads, (3) the provenance and timing of those quotes, and (4) quality checks that separate market mechanics from variable conditions such as liquidity, execution, and provider methodology.
Because spreads are not stable across time, you also need explicit assumptions for any example calculation and at least one limitation you are willing to accept—typically that historical relationships and quoted spreads do not guarantee future results.
Mechanism and definition
A “pair spread” is usually the difference between the buy (ask) and sell (bid) price for the same currency pair at a specific moment and under a specific quoting convention. In practice, the “spread” you see depends on what is quoted (mid, bid/ask, or trade prices), the quote currency/units, and whether you measure it in raw price terms or as a percentage.
To assess it, you need a consistent set of inputs:
- Bid and ask (or the exact formula the provider uses to derive them).
- The timestamp or time window for the quotes (to avoid mixing asynchronous data).
- The quote venue or data source (provider, exchange, or feed), since different sources can publish different bid/ask.
- The pair specification (for example, what “EUR/USD” means in that dataset: standard spot quoting versus other contract conventions).
Then decide your calculation assumptions. For example, you must state whether you will compute a raw spread (ask − bid) or a relative spread (ask − bid divided by a chosen reference price such as the mid). If you compare across time or across pairs, you must apply the same method and units throughout.
Evidence or example: what to record and how to validate
An example checklist for a self-audit (no real-time data assumed) is:
- Record quote inputs: store the bid and ask values that correspond to each observation.
- Record provenance: note the exact data source and whether it is indicative quotes or executable prices.
- Record timeliness: keep timestamps (or interval endpoints) so you can detect gaps, stale quotes, or batching artifacts.
- Perform quality checks:
- Outlier scan: spreads can spike; confirm those spikes are not due to broken timestamps, missing bid/ask, or unit mistakes.
- Consistency check: confirm ask is not lower than bid for the same observation, and confirm units match your computation.
- Liquidity sensitivity: even with identical calculation rules, spreads typically widen when liquidity is thinner. If you have order-book or liquidity proxies, use them to interpret changes.
- Separate quoted spreads from execution costs: quoted spreads omit other frictions such as commissions, slippage, and whether you can trade at the displayed bid/ask. If your goal is “what you pay,” you need data about actual executions or a documented execution model.
A material limitation or failure mode is mixing different quote sources or methods. For instance, one dataset may represent mid-derived estimates while another uses directly published bid/ask. Even if both are called “spread,” they may not be comparable.
Limitations and risks
- Non-stationarity: spreads change with market regime, liquidity conditions, and volatility. Historical patterns are not evidence of future behavior.
- Provider and methodology differences: different quote feeds, rounding rules, and “indicative vs executable” definitions can shift observed spreads.
- Timing mismatch: if bid/ask are sampled at different moments, the computed spread can be artificial.
- Costs beyond the spread: execution quality can differ from quoted bid/ask, so assessing only quoted spreads can understate real transaction costs.
To manage uncertainty, state your assumptions explicitly: calculation method, unit conversion, reference price choice, and what counts as “the spread” for your dataset.
Verification or next question
You can independently verify pair-spread claims by checking that the dataset provides bid and ask (or an unambiguous derivation), that every observation has a timestamp, and that the measurement rule (raw vs relative spread, units, and reference price) is consistent.
If you are comparing across timeframes or jurisdictions, the next question to ask is whether the quote source and execution definition stay the same. If they do not, apparent differences may reflect measurement changes rather than market behavior.