Registered investment advisers, methodology

This module uses the SEC Form ADV FOIA bulk archive's IA_ADV_Base_A file, staged by scripts/acquire/adv_ncen_fetch.py and built into data/parquet/adv_advisers.parquet by scripts/build_adv_advisers.py. The source-vintage line on this page derives the raw filing dates, selected adviser-year, and selected snapshot end date from that parquet at runtime.

The manually acquired SEC FOIA archive has a published boundary of 2024-12-31, while the latest date_submitted actually observed in adv_advisers.parquet is 2024-12-26. Form ADV filings from 2025 onward live on the separate adviserinfo.sec.gov surface, and this repository has no implemented acquisition path for them. The manual tier in the data catalog describes FinObservatory's acquisition state, not the SEC filing frequency. The selected 2024 adviser-year panel is therefore a historical snapshot, not a current 2026 industry cross-section.

The central trap: filings are not advisers

The source rows are filings, so summing them directly would repeatedly count advisers that filed more than once. Every page query deduplicates before aggregating.

The deduplication rule is:

1. Build an adviser key as CRD:<crd>.
2. If crd is missing, fall back to CIK:<cik>.
3. Within each adviser key, keep the latest filing by date_submitted, breaking ties by filing_id.
4. For time series, apply the same rule within each (adviser_key, filing_year) pair so every calendar year keeps one filing per adviser.

The cross-section shown uses the latest filing per adviser in the latest calendar year whose adviser count is at least 90% of the preceding year's count. That computed guard prevents a newly arriving partial year from replacing the latest full snapshot and avoids mixing the selected panel with advisers whose last observed filing is older.

Coverage and reports_raum

reports_raum is a VARCHAR, not a boolean. Before any adviser query runs, the data layer asserts that every non-null value is Y or N and throws if another token appears. The page includes only Y filings in RAUM charts, rankings, concentration measures, and mandate totals. An N or null means that the adviser does not report RAUM in that filing; it is not converted to zero.

Filed values and anomalies

RAUM queries publish the deduplicated filings as reported and apply no value-based or filing-key exclusions. The computed anomaly note below documents the unusually large filing for CRD 316306, links to its current IAPD Form ADV, and reports the annual total, top-10 share, and HHI both with and without that filing. The counterfactual is disclosure only; the charts and tables include the filing.

What the charts aggregate

The annual charts use the latest filing per adviser within each calendar year. Each history start is derived from field coverage in the parquet. A field's first public year must contain that field for every applicable adviser and have at least 90% as many advisers as the following year. For RAUM, "applicable" means filings with reports_raum = 'Y', and all three RAUM fields must be present. This omits the first partial collection year without relying on typed calendar cutoffs.

The concentration chart measures the annual RAUM shares managed by the largest reporting advisers. The RAUM chart keeps discretionary and non-discretionary mandates separate rather than treating them as interchangeable.

Comparable views of aggregate RAUM

RAUM is an as-filed stock at the latest filing in each adviser-year, not a return series. The RAUM chart offers four views of the same three annual aggregates. Level preserves the reported total, discretionary, and non-discretionary source values shown in trillion US dollars. All three series use the same adviser-year dates.

YoY is the percentage change in each aggregate stock from its date-matched prior-year observation. A missing prior annual observation produces a missing change rather than an interpolated comparison. RAUM changes combine market moves, flows, adviser entry and exit, and filing coverage. They are not investment performance or net flow, and they should not be read as investment returns.

Index sets each series' own first usable annual observation to 100. Total, discretionary, and non-discretionary RAUM therefore have three independent index bases. Discretionary and non-discretionary RAUM remain separate series in transformed views; they are not percentages of total.

Log shows log10 of each positive RAUM observation in the chart's trillion-US-dollar units. A log value is therefore not a raw-dollar level. The selected mode is stored in chart-specific URL state under advisers-raum-history, so the figure's Copy link restores the view while retaining its source and anchor.

Concentration, employees, clients, tables, and anomaly counterfactuals remain Level only. The transforms do not remove or reinterpret the disclosed anomalous filing: every RAUM view includes it as filed.

Employees and clients

Employees and clients are shown exactly as filed for every adviser with those fields, regardless of reports_raum. Neither field is imputed: a filed zero remains zero and a null remains missing.

The public client chart starts in the first year where every adviser has a client value and the panel meets the same near-full-year guard. Because later filings contain many zero client values, the chart shows median and 90th-percentile client counts only among filings with a positive count and labels that restriction.

Employee and client charts describe per-firm distributions. They do not measure aggregate industry headcount or aggregate client growth.

Units

RAUM in the parquet is raw US dollars. Overview cards and tables use dollar labels, while the RAUM chart's visible y-axis ticks use $T. Concentration ticks use %, employee ticks use employees, and client ticks use clients. Tooltips and source lines retain the same units.

Out of scope

• Form ADV brochures and narrative brochure text.
• Any adviser-to-fund linkage. That belongs with N-CEN and the separate P-70 work.
• Enforcement or disciplinary history.