Cross-sectional data record different entities at a common point in time or over a common reference period. A table of stock returns for the same month, bond yields at a specified valuation time, or firms' annual balance-sheet measures can be a cross section. In quantitative finance, this structure lets you compare assets with one another: which stocks performed better over the interval, or which bonds have higher yields at the valuation time?
The observation index identifies the entity. It does not mean “the next date.” A time series follows observations through time; a cross section compares observations across entities.
A return comparison across assets
Suppose three hypothetical assets have the following total returns. Every return covers the same preceding interval, uses the same currency basis and is available at the common evaluation time just after the interval ends. These are illustrative measurements, not market quotes.
| Asset | Return over the common interval |
|---|---|
| A | +2% |
| B | −1% |
| C | +5% |
The equal-weight cross-sectional mean is
Here identifies the asset, identifies the common interval end, and . The ranking is C, A, B, from highest to lowest return. Reordering the table alphabetically or by return leaves the cross-sectional mean unchanged, provided each return stays attached to its asset.
Original synthetic cross section. Each bar measures a different asset over the same interval; the dashed line marks the equal-weight mean. The asset axis is not a sequence of dates.
This mean summarises performance across assets for that interval. It is also the interval return of a portfolio with equal initial values in these assets, before costs, under the stated common return conventions. It is not a time average or an estimate of what any asset will earn next. Compounding A's return with B's and C's would invent a sequential investment path that the table does not contain.
“Common time” needs an information rule
A shared date label is only a starting point. Make the comparison economically consistent:
- Measurement window: compare returns over the same beginning and ending timestamps. Different exchanges' closing times need not coincide, even when the date labels match.
- Definition and units: keep total return distinct from price return, and specify the currency basis. For accounting characteristics, check fiscal periods and definitions rather than assuming every annual figure covers the same window.
- Availability: a financial statement's period end is different from its publication time. A trading signal evaluated at a given cutoff can use only information available by that cutoff; later releases or revisions can introduce look-ahead bias.
- Universe: define which securities qualify at the evaluation date. Selecting only firms that survive to the end of a later sample can introduce survivorship bias.
“Cross-sectional” describes the observation structure. It does not guarantee independent observations: stocks can respond to the same market or industry news. Nor does a relationship between characteristics and returns establish causation. These distinctions matter when a linear regression turns a descriptive comparison into an estimated relationship with reported uncertainty.
Adding dates changes the dataset
Observing the same assets over multiple dates creates panel data, with both entity and time indices. Taking fresh samples at successive dates instead produces repeated cross sections; the entities need not be the same. Keep identifiers and dates, even if the final analysis uses a separate cross section for each date.
In asset-pricing research, a time-series regression can estimate each asset's factor exposure, while a subsequent cross-sectional regression asks how returns differ across assets with different exposures. The Time-series and cross-sectional tests lesson develops that distinction and the complications introduced by estimated exposures. It is an advanced follow-on, with a public preview; the full lesson requires membership.
Sources
- Chris Brooks, Introductory Econometrics for Finance, 2nd edition (Cambridge University Press, 2008), §1.3.2–1.3.3, p. 5: cross-sectional observations, entity indexing, arbitrary row order and panel data; §4.5.13, p. 160: dependence between cross-sectional observations.
- Singapore Department of Statistics, “How to Analyse Data”: comparisons within a common period and the distinction between cross-sectional, time-series, pooled and longitudinal data.