What spread assumptions mean
“Spread assumptions” are the stated expectations about the transaction cost represented by the bid–ask spread used in a calculation, backtest, forecast, or example. Spreads are not just a single number: they depend on how the quote is defined (bid and ask), when it is measured, and how the trade is assumed to execute.
To assess whether your assumptions are reasonable, you need data that lets you: (1) define the spread concept precisely, (2) reproduce the same spread calculation you use, and (3) check whether that spread behavior is stable enough for the purpose.
Core inputs you need
You typically need four input groups: the spread definition, the price/quote data, the execution assumptions, and any costs that interact with spreads.
- Spread definition and measurement rule State which spread you mean:
- Raw spread: computed directly from bid and ask at the time of measurement.
- Effective spread: spread experienced after accounting for execution timing and order conditions.
- Modeled spread: an estimate derived from a transformation (for example, an average or a rule). Without a specific definition, later comparisons are ambiguous.
- Quote or price data used to derive the spread Collect the bid and ask series (or equivalent fields) that your spread assumption is based on. For each series, record:
- the instrument identification,
- the quote timestamps,
- whether the data is tick-level, aggregated, or sampled.
-
Timeliness and alignment with execution Spread data should match the “when” of your calculation. If you assume an entry at a specific time, the spread data should be derived from quotes that are contemporaneous with that entry time. Misaligned timestamps can create a systematic mismatch.
-
Execution and trading-cost assumptions that affect realized spread Even if you assume a spread, the realized cost can change due to execution details. Data to capture includes:
- whether trades are assumed to happen instantly at the quoted price or with delay,
- whether orders are assumed to be market or limit,
- any other transaction cost components you combine with spread (for example, commissions or fees, if you model them).
Provenance: where the data comes from
You also need provenance—evidence of origin and intended use—for every variable that enters the spread assumption.
Key provenance details include:
- Data source type: broker-provided quotes, exchange/venue data, or a data vendor feed.
- Data processing transparency: whether the feed is raw or already filtered/cleaned.
- Sampling policy: whether quotes are complete or if gaps are filled.
- Consistency across time: whether the feed format or liquidity conditions changed.
If the spread assumption is derived from one kind of feed, but your evaluation uses another, you cannot reliably validate the assumption.
Timeliness and quality checks before you trust the assumption
Because spreads can vary quickly, you should apply quality checks that confirm the assumption is not built on artifacts.
Material checks include:
- Data completeness: look for missing bid/ask points or long gaps.
- Timestamp integrity: confirm monotonic timestamps and correct time zone handling.
- Outlier handling policy: document how you treat extreme values (for example, whether you cap, exclude, or keep them).
- Granularity compatibility: ensure the spread series granularity matches how you use it (per trade, per bar, per day, etc.).
A spread assumption can fail even when the average spread looks plausible if the dataset hides periods of illiquidity.
Evidence or example: validating with comparable periods
A practical way to assess spread assumptions is to compare the assumed spread behavior against observed behavior in separate, comparable periods.
For example, if you assume a constant average spread, you need a dataset that shows:
- the distribution of observed spreads over the relevant times,
- the proportion of time the spread exceeds your assumed value,
- whether spread volatility increases during certain market regimes.
If you assume a time-varying spread rule, you need to verify that the rule reproduces observed spreads when applied to the same definitions and timestamp alignment you used originally.
Limitations and failure modes to account for
At least one material limitation is essential: spread assumptions can be violated by conditions that your data or model does not capture.
Common failure modes include:
- Liquidity-driven widening: spreads can widen abruptly when liquidity drops.