How to audit portfolio data quality

1. Freeze the question and the window

Write down what the chart is meant to describe before inspecting the line: a historical return path, a risk measure, a stress window, or a modeled scenario. Record the start date, end date, timezone, date inclusivity, instruments, weights, and benchmark. A result with a different window or portfolio definition is answering a different question, even if the chart uses the same title.

2. Record provenance and freshness

Name the provider or feed and the kind of observation it supplies. Capture the latest provider observation date (as-of date) separately from the retrieval timestamp. Note the price basis, distribution and corporate-action treatment, currency, venue, and calendar. A recent retrieval does not make an old market observation current, and a display-currency label does not prove that source prices were converted.

3. Measure coverage per holding

Compare the requested window with the dates actually available for every holding. Record the common observation window, each holding's first and last usable observation, missing dates, and whether a result is complete, partial, insufficient, stale, unavailable, or unknown. Do not let portfolio-level coverage hide a short or interrupted history in one sleeve. If observations are excluded or missing, state that rule; never silently forward-fill a price to make the series look complete.

4. Reproduce the calculation assumptions

A portfolio result is a calculation applied to observations and rules. Record rebalancing cadence and date rule, fees, expense ratios, cash allocation and yield, contributions or withdrawals, inflation, taxes if included, benchmark methodology, and missing-data handling. Confirm that each assumption can affect the output when its inputs imply an effect. Keep calculated values separate from observed metadata and label modeled overlays as scenarios rather than observations.

5. Turn the audit into a bounded conclusion

A defensible conclusion says what the selected evidence supports, which limitations narrow the claim, and what it does not establish. For example, a result may describe a historical reconstruction under stated fees and partial coverage while not establishing repeatability, suitability, or future performance. Quality metadata does not make incomplete evidence complete; it tells you whether to narrow the interpretation or rerun the comparison.

Audit checklist

• Can another reader identify the provider, as-of date, retrieval time, window, and price basis?
• Is coverage reported for each holding, with missing observations and common dates visible?
• Are currency, distributions, corporate actions, venue, timezone, and calendar caveats stated where relevant?
• Are fees, rebalancing, cash, benchmark, contributions, withdrawals, and missing-data rules explicit?
• Are observed values, calculated outputs, and modeled overlays labeled as different evidence types?
• Does the conclusion state both what the result supports and what it cannot establish?

Printable data-quality worksheet

Record observed metadata, calculated summaries, modeled overlays, and the conclusion boundary in separate sections. Open the blank worksheet to print or save as a PDF.

Evidence boundary

A data-quality audit explains whether selected observations and assumptions support a stated historical or modeled result. It does not make incomplete data complete, remove market uncertainty, establish suitability, or predict future performance.