Method
A panel follows the same firms through successive editions of a public filing. The SEC republishes its roster of registered investment advisers every month, and each adviser in it carries a CRD number that stays the same from one file to the next. Fix a group of firms in one month’s file, find each of them again in later files, and a stack of snapshots becomes a record of what happened to particular firms: which left the SEC’s rolls, which changed, and which stayed exactly where they were.
Most of the work is in not being fooled by the files. Every piece here follows the same rules:
- Every figure traces to a line of code in the public repository that printed it.
- Columns are resolved against each file’s real header row, and their values are checked rather than assumed. A field that reads like a website address can turn out to be a yes/no flag.
- Every derived cut has a cross-check that fails loudly. Firms entered minus firms exited must equal the net change, and a diff of two rosters is checked against each firm’s own SEC status date.
- Each file is dated by the latest filing it contains, never by the date in its name.
- Every source file is checksummed and its download URL recorded. The archive is listed on the Data page.
The rest of this page is the method appendix for the one panel published so far, the Form ADV new-registrant panel, adapted from METHOD.md in the repository. Its sources and checksums are at the end of the report.
The Form ADV new-registrant panel
Run: python adv_new_registrants.py --current data/ia09012026-registered.zip --prior data/ia09022025.xlsx
The file-selection trap
The SEC publishes two extracts per month and they are not interchangeable:
ia<MMDDYYYY>-registered.zip— registered investment advisers. ~17k rows, 448 columns, includes Item 5 (AUM, employees, clients).ia<MMDDYYYY>-exempt.zip— exempt reporting advisers. ~6.6k rows, 171 columns, no Item 5 at all (ERAs file a truncated ADV: Items 1, 2B, 3, 6, 7, 10, 11, 12).
2025-and-earlier vintages are bare .xlsx with no -registered suffix, so the
naming convention changed mid-series. A bare 2026 .zip still resolves at the old
path but serves ERA content — the first download attempt for this analysis pulled
ia08032026.zip and got 6,604 ERA rows with no AUM column. The script now refuses
to run unless Firm Type is entirely Registered in both files.
The date trap
The filename date is when the SEC posted a file, not what the file contains.
ia09022025.xlsx is the July 31, 2025 roster republished under a September name:
- It holds exactly the same 16,359 CRDs as
ia08012025.xlsx. Real months churn: the adjacent months in the archive run +182/−68, +159/−76 and +207/−76 firms. August to September 2025 is +0/−0. - Its latest ADV filing date is 2025-07-31, identical to the August file.
- The only difference is one added column,
SEC Current Status.
The window is therefore August 1, 2025 to August 31, 2026, thirteen months. Cutting the validation on the filename date instead (2025-09-02) misclassifies every August 2025 registration as a re-registration. The code now derives both window ends from file contents, and archive/fetch_adv.py flags any vintage whose data does not advance past the previous one.
Field mapping
Resolved against the real header row and asserted on values, not hardcoded:
| Logical field | Column | Verification |
|---|---|---|
| CRD | Organization CRD# |
100% numeric, 17,149 unique, 0 duplicates |
| Name | Primary Business Name |
— |
| AUM | 5F(2)(c) |
>95% parseable; median $419M, total $177.0T |
| Employees | 5A |
>95% parseable, non-negative; median 8, p90 63 |
| Website | 1I |
Y/N flag, not a URL. Values ⊆ {Y, N} |
| Website count | Total Number of Website Addresses |
1,390 zeros == 1,390 1I=N — independent agreement |
| State / City | Main Office State / Main Office City |
— |
| Status date | SEC Status Effective Date |
100% parses as %m/%d/%Y |
Item 1.I is the trap. It reads like a website address field and is a yes/no
flag. Treating it as a URL string makes every N count as having a website and
inverts the headline stat. The Total Number of Website Addresses reconciliation
is what proves the reading is right.
Diff and its validation
Firms present in the current file and absent from the prior one, keyed on
normalized CRD (strips whitespace, trailing .0, non-digits).
prior (data through 2025-07-31): 16,359 registered
current (data through 2026-08-31): 17,149 registered net +790
entered: 1,693
exited: 903 1,693 − 903 = +790 ✓ (asserted)
Cross-check against each firm’s own SEC Status Effective Date:
- 1,689 of 1,693 (99.8%) have an effective date on or after 2025-08-01, the day after the prior file’s data ends. These are genuine new registrations.
- 4 (0.2%) carry an older effective date. These are re-registrations, status changes, or firms that left and returned within the window.
- 0 unparseable.
The CSV flags these as _new_registration so any cut can be run either way. The
headline figures below use all 1,693; the no-website result is unchanged on the
1,689 subset (15.9%), so it is not an artifact of the re-registrations.
Results
| Metric | Value |
|---|---|
| New registrants | 1,693 |
| No website (Item 1.I = N) | 269 (15.9%) vs 8.1% for all 17,149 registered advisers |
| ≤5 employees | 1,181 (69.8%) — median headcount 4 |
| ≤1 employee | 249 (14.7%) |
| Median reported AUM | $120.1M (vs $419.4M for all registered advisers) |
| Reporting $0 AUM | 309 (18.3%) |
No-website rate by AUM band:
| Band | n | no website |
|---|---|---|
| $0 reported | 309 | 32.4% |
| <$25M | 54 | 13.0% |
| $25–100M | 111 | 12.6% |
| $100–500M | 1,046 | 12.6% |
| $500M–1B | 83 | 7.2% |
| >$1B | 90 | 11.1% |
Top states: NY 199, CA 171, FL 135, TX 93, IL 63, MA 50, NJ 39, OH 38, PA 38, CT 38.
Limits
- The $100–500M concentration is mechanical, not a finding. $100M in regulatory AUM is the threshold at which an adviser must move from state to SEC registration. 61.8% of new SEC registrants landing in that band is the threshold doing its job. Anyone in this market knows that; presenting it as a discovery is disqualifying.
- “New to the SEC” ≠ “new firm.” A large share of these are existing state-registered advisers crossing $100M, not startups. The extract cannot distinguish the two. The 4 firms (0.2%) with older effective dates are the only part of this the file makes visible; the rest is not measurable from it.
- The no-website result is concentrated in the $0-AUM cohort. Excluding those 309 firms, the rate is 12.2% (169 of 1,384) — still above the 8.1% baseline, but the “twice as likely” headline leans on the $0 group. Say so. Note that 12.2% is the pooled remainder and has to be recomputed; it is not the ~12.6% that the middle bands individually show.
- $0 reported AUM is ambiguous. It can mean a newly approved adviser that has not yet taken on assets, an adviser with only non-discretionary or non-securities assets, or a reporting gap. Not investigated here.
- Point-in-time, not a flow. This is a two-snapshot diff. Firms that registered and deregistered between the two dates are invisible.
- Umbrella registrations mean one CRD can cover multiple relying advisers, so the firm count is not a count of legal entities.
- The panel counts new registrants only. Of the 628 firms reporting $0 on the July 31, 2025 roster, the 288 absent from the June 30, 2024 roster are the panel; the 340 already registered are a different population (they converted at 6.8%) and are never pooled in. “Deregistered” means absent from the August 31, 2026 registered roster: closing, merging and moving to state registration look the same.