Freeze what the comparison means
A benchmark specifies the opportunity set or exposure against which a result is judged. A universe specifies which instruments were eligible for the research or allocation. They are related but not interchangeable. Before evaluation, record their construction rules, weighting, currency, rebalance schedule, treatment of distributions, and applicable costs. A regime rule tested on one opportunity set cannot silently inherit a favorable comparison from another. Naming an index or asset category is not enough to reproduce the research definition.
Distinguish prospective changes from rewritten history
A prospective benchmark change applies from a declared effective point; a retrospective change replaces comparisons for earlier periods. Both can be legitimate analytical choices, but they answer different questions. The GIPS benchmark guidance distinguishes these cases and stresses disclosure of the change. For an educational research record, preserve the original comparator and show the alternative alongside it, with the reason and effective boundary. This borrows a transparency principle; it does not assert compliance, verification, or endorsement under any performance-reporting standard.
Reconstruct membership as it was knowable
Using today's surviving instruments throughout history changes the historical opportunity set. Record eligibility at each selection time, including removals, failed instruments, and instruments that later became prominent. Separate announcement time from effective membership time. A newly announced constituent is not automatically part of an earlier portfolio. When constituent history is incomplete, describe the resulting survivor-only or fixed-universe study explicitly. Do not infer that an unavailable delisting outcome was zero, and do not remove a difficult instrument merely because its final observations complicate the calculation.
Use stable instrument identifiers rather than ticker text alone. Symbol changes, reused symbols, mergers, and separate share classes can otherwise join unrelated histories or duplicate exposure. Preserve the mapping's effective interval and distinguish an issuer from a particular security. Membership checks should resolve those identities before calculating state-conditioned summaries.
Version the return definition as well as the names
A dataset can change while instrument names remain unchanged. Return construction, reinvestment timing, corporate-action treatment, and delisting conventions can all affect comparability. The original factor-data documentation and the stock-data format mapping cited below illustrate why definitions must be inspected, not inferred from a column heading. Keep the dataset version, adjustment convention, extraction timestamp, and any correction history with the results. A repaired historical return series may be useful for measurement, but it does not justify rewriting the information available to a historical selection rule.
Read a fictional benchmark substitution
In the invented one-period example, the strategy earns 4%. Against the predeclared benchmark's 6%, active return is -2 percentage points. Against an alternative benchmark's 1%, it becomes +3 percentage points. The comparison changes by 5 percentage points even though no strategy holding or outcome changed. This is a difference of simple returns, not a compounded relative-wealth calculation or a risk-adjusted alpha estimate. The example shows why an alternative benchmark should not replace the original simply because it produces a more favorable description.
Keep revisions and selection timing separate
When a universe screen depends on accounting or economic information, preserve the release and revision history used by the screen. Later-restated values may alter both membership and the regime label. Compare decision-time and revised-data reconstructions as separate experiments with a clear change log. For benchmark transitions, compound each defined return segment under its stated rule; do not splice levels without a continuity convention. Label a composite of different benchmarks as a composite, rather than implying that the current benchmark supplied the whole historical series.
Publish a comparison bridge
A useful bridge shows the original result, the revised comparator or universe, and the changes attributable to each definition. Change one dimension at a time where possible: benchmark, membership, return convention, then costs. Retain both the original research output and the corrected version if an error is discovered, explaining which is authoritative for which question. If the apparent regime advantage depends on one favorable definition, make that dependence the conclusion. This is research-accounting guidance, not investment advice or a claim about any actual benchmark's performance.
Inspect the comparison
| Invented comparison | Same strategy return | Benchmark return | Active return |
|---|---|---|---|
| Predeclared benchmark | +4% | +6% | -2 percentage points |
| Alternative benchmark | +4% | +1% | +3 percentage points |
Primary references
References support the stated methods and data conventions. Original fictional examples are not observations from the cited sources.