Direct answer: what to measure
Spread assumptions are measurable when you turn “expected spread” into specific, recorded quantities tied to exact quote times and clearly defined data sources. Practically, that means you measure the bid–ask spread (the difference between the ask and bid) as a time-stamped series, then use a consistent comparison method to see whether your assumed spread resembles what actually occurred under similar conditions. If you cannot specify the spread definition, the timestamp, the sampling window, and the execution context, then the assumption cannot be independently checked.
Mechanics: define the spread fields and timestamps
Start by defining the spread you mean. Common measurable choices include:
- Quote spread: ask minus bid at a given moment.
- Effective spread: the realized cost relative to a reference price, when applicable.
- Trade-cost spread proxy: an approximation that folds in other components such as commissions and execution effects.
To measure properly, you also need timestamps and sampling rules:
- Quote timestamp: when each bid/ask snapshot was observed.
- Sampling interval: how frequently you record spreads (for example, every second, every minute, or only at decision times).
- Alignment rule: how you match assumed spreads to the time your model or strategy would “act.”
A spread assumption “works” in a measurement sense only if your assumed number is compared to the same kind of spread under the same timing and sampling rules. Historical comparisons should be done with a like-for-like method.
Evidence or example: compare assumed vs measured spreads
A simple measurable approach is a scenario table:
- Choose an assumption set, such as S_assumed (a fixed value) or a S_assumed(t) rule (a time-varying rule).
- Measure S_measured(t) from time-stamped bid/ask quotes using the same spread definition.
- Evaluate differences using a clear metric, such as average spread, median spread, or the fraction of observations above a threshold.
For example, you might assume a constant spread, then measure the actual quote spread over the same market sessions you intend to model. The key is that you specify which session hours you used, which sampling interval you used, and whether you measured during high-liquidity and low-liquidity periods separately. If your assumption ignores those distinctions, the comparison will be misleading.
Limitations and risks: where spread assumptions fail
At least one material failure mode is that spreads change with liquidity and volatility. A single “typical” spread can miss spikes and widenings that matter most for execution.
Other important limitations include:
- Execution mismatch: quote spread is not the same as realized cost after fills, latency, or order-book depth changes.
- Cost omissions: commissions, fees, and other execution costs can be larger than the spread difference you measured.
- Regime mixing: averaging across different market conditions can hide heavy tails (rare but large widenings).
- Non-stationarity: relationships observed historically may not hold in future conditions.
Because of these limits, measured spreads should be treated as conditional evidence about the periods and contexts you actually sampled, not as a general guarantee.
Verification or next question: how to independently check
Independently verify spread assumptions by publishing (in your own work) the exact definition, timestamp method, sampling interval, and comparison window used to compute S_measured. Then re-run the measurement under at least one alternative, such as a different sampling interval or a different session segmentation. If the conclusions depend heavily on arbitrary choices, the spread assumption is fragile.
A useful next question to ask is: “When would the assumption break most often?” Identify the specific conditions where spreads widen or execution differs from your measurement method, then document those conditions as measurable failure modes.