Choose an ADV window without smuggling in future volume

Write a volume specification

A reproducible ADV entry needs more than a number and ticker. Record the security identifier, currency, share or notional unit, trading dates, session boundaries, treatment of auctions, corrections and missing records. Also record when the underlying data became available. A backdated spreadsheet containing later corrections is not automatically the information a trader had at the decision time.

NYSE’s market-statistics discussion separates shares, notional, price groups and closing-auction activity. Those distinctions motivate a practical rule for this publication: do not substitute an attractive market-wide headline for the instrument-level denominator required by an order calculation. The guide’s numerical series below is invented for arithmetic demonstration, not drawn from that report.

An outlier can change the apparent constraint

Suppose five hypothetical daily dollar volumes are $8 million, $9 million, $10 million, $11 million and $42 million. Their arithmetic mean is $16 million; their median is $10 million. A $1 million order represents 6.25% of the mean but 10% of the median. Neither statistic alone tells you which denominator is appropriate for tomorrow.

If the $42 million session reflects an identified event, inspect whether the next trading window shares that event condition. Do not delete the observation merely because it makes a stress test inconvenient. Preserve the full sample, identify the event and present both calculations with their intended use. An explicitly chosen low-volume scenario can be more informative than silently editing the history.

Keep lookbacks causal

For a decision made before Monday’s session, use only volume observations and corrections available by that cutoff. If Monday later becomes a record-volume day, inserting it into the Monday forecast understates the uncertainty faced at the time. Store separate observation date, release timestamp and retrieval timestamp fields where the source permits.

A longer lookback reduces sensitivity to individual sessions but can lag a structural change. A shorter lookback responds quickly but can be dominated by noise. Compare these choices on the same dated snapshot, then freeze the convention used for a particular experiment. Changing the window after seeing expensive executions confuses model repair with evaluation.

Diagnose unit and corporate-action problems

A share-volume history around a stock split requires a consistent basis if it is combined with a share order. A dollar-volume history computed from raw prices and raw shares may avoid some unit discontinuities, but it still requires compatible timestamps and vendor definitions. Never multiply an adjusted price by unadjusted shares without investigating the resulting meaning.

Use simple reconciliation checks. Sum the included daily dollar volumes and divide by the number of included valid sessions. List missing days rather than treating absent data as zero. Conversely, a verified zero-trade session is meaningful and should not disappear through an automatic filter that drops all zeros.

Document the decision consequence

In the five-day example, a hypothetical 8% participation ceiling admits the order under the mean and rejects it under the median. That difference is the result worth discussing. The input choice changes the scenario decision; it does not reveal which order would actually fill.

Export the original observations when redistribution rights permit, or export the calculation recipe and a lawful reference instead. A public source link is not a redistribution licence. This site’s downloadable fixtures use hypothetical values so readers can reproduce the arithmetic without mistaking the examples for licensed market observations.

Keep in the decision record

• Freeze data availability and lookback before evaluating a decision.
• Show sensitivity to unusual sessions instead of deleting inconvenient observations.
• Distinguish missing records, verified zero volume and corporate-action adjustments.

Source notes

1. NYSE — Daily U.S. Equity Market Statistics Report
2. NYSE — Market Data Policies

Sources inform the definitions and limitations. Scenario inputs and arithmetic are publication illustrations, not extracted market observations. Access checks are dated in the source inventory.

Continue the analysis