Weighted averages: preserve the denominator across entities
A group average is only meaningful when its weights match the question. Preserve the denominator behind each company-level rate before aggregating it. Taking the simple average of percentages can make a small business count as much as a large one and produce a result that describes neither the typical company nor the combined economic exposure.
Choose the estimand first
There are several legitimate questions. What is the average issuer’s margin? What is the combined margin of this group? What margin is associated with a portfolio’s exposure? Equal company weights, revenue weights and portfolio weights answer different questions. State which quantity you are estimating before selecting a convenient spreadsheet formula.
The NIST Dataplot weighted mean reference gives the weighted mean as the sum of weighted observations divided by the sum of weights. The arithmetic is general. It does not establish that revenue, market capitalization or portfolio value is the appropriate weight for every financial ratio.
A hypothetical margin comparison
Suppose an invented group contains Company A with revenue 900 and operating profit 90, and Company B with revenue 100 and operating profit 30. Their operating margins are 10% and 30%. The equal-weight average is 20%, while the combined margin is 120 / 1,000 = 12%. Revenue weighting reproduces that combined margin because each margin’s denominator is revenue.
| Issuer | Revenue | Operating profit | Margin | Revenue weight |
|---|---|---|---|---|
| A | 900 | 90 | 10% | 90% |
| B | 100 | 30 | 30% | 10% |
| Combined | 1,000 | 120 | 12% | 100% |
Neither 20% nor 12% is inherently dishonest. Calling 20% the group’s combined margin would be wrong; calling 12% the mean company margin would obscure the weighting choice. The median margin of two companies is also 20% here, but that numerical coincidence does not make the concepts interchangeable.
Preserve the numerator and denominator
Whenever possible, keep the original numerator and denominator alongside the reported ratio. Confirm that periods, units, currencies and accounting scopes are compatible before summing. If one company reports an adjusted profit measure and another reports an accounting operating result, consistent weighting will not repair the definition mismatch.
Do not assume every ratio supports the same aggregation. A price-to-earnings ratio involves market value and earnings; an arithmetic average of issuer ratios does not automatically equal combined market value divided by combined earnings. Negative or near-zero earnings can make interpretation particularly unstable. Specify the inclusion rule and show the excluded exposure rather than hiding it.
Distinguish missing from zero
If a denominator or ratio is unavailable, a missing observation should not become a zero. Report how much of the relevant weight has valid data. For example, a portfolio-weighted metric with coverage of 70% describes the observed portion unless you explicitly model the missing 30%. Renormalizing known weights is acceptable for a stated partial-sample statistic, but it changes the question.
Also verify that weights belong to the same observation date as the intended analysis. Current market values applied to old financial ratios can answer a current-exposure question, yet they cannot recreate what a historical investor knew. Preserve both the measurement period and the weighting date in the output.
Review changes in the aggregate
When a weighted rate changes, separate issuer-level changes from weight changes. A group margin can improve even if each continuing issuer’s margin is flat, simply because revenue shifts toward the higher-margin company. Keep a constant-weight comparison as a diagnostic, label it clearly and do not present that diagnostic as the actual combined result.
The final table should include the weighting basis, numerator and denominator definitions, coverage and treatment of problematic values. An AI summary should use those labels exactly. A correct-looking percentage without its denominator is an invitation to answer the wrong question.
FAQ
Should I always use market-cap weights?
No. The weight must correspond to the quantity being estimated. Revenue weights can reproduce a combined operating margin; market-cap weights describe a different exposure.
Does a weighted mean eliminate sample bias?
No. Weighting does not restore omitted issuers or make incompatible measures comparable. The sample definition and coverage remain separate checks.