Who appears more than once

Of 4,219 distinct names in the defendant lists, 669 appear in two or more records, 129 in two or more matters, and 46 in separate enforcement episodes after a hand check: 31 firms and 15 individuals. 23 of the firms are banks, securities firms, exchanges or a transfer agent. Anonymised respondents (X, A) in French decisions are not matched. A repeat name is not evidence of repeat misconduct.

Published 2026-10-04 · 9 min read

Most names in the library appear once. Of 4,219 distinct names in the defendant lists of its 2,637 case records, 669 appear in two or more records, 129 in two or more matters, and 46 in two or more separate enforcement episodes. The gap between the first two numbers and the third is the point of this post: most repeats are one case recorded more than once, at its filing and again at its judgment, or the same conduct pursued by two regulators. The 46 that remain are 31 firms and 15 individuals, and 23 of the firms are banks, securities firms, stock exchanges or a transfer agent.

What a repeat appearance is, and is not

It is the same name in different records. Nothing more. It is not a count of offences, and it is not evidence of recidivism. The records behind these counts include a settled order against a bank’s securities arm, a complaint that has not been decided, and a tribunal case in which none of the allegations were found proven. One name is a respondent in one record and appears again as a respondent in another; the library cannot tell from that how much conduct there was, whether it was related, or whether anyone at the firm was the same person. A large bank that trades in dozens of markets can appear in four unrelated orders. A small firm can appear in two because a regulator brought an order and then, years later, a follow-up. Both are repeats in the count.

How the names were counted

The analysis reads the defendants field of every case record: 5,512 entries, of which 485 are excluded as anonymised placeholders (below); the rest give 1,421 distinct entity names and 2,798 distinct individual names after normalising. A name is lowercased, accents and punctuation are removed, “et al.” and a leading “the” are dropped, and trailing legal-form words (Inc, LLC, Ltd, plc, N.A., and the French SA, SAS and SARL, and similar) are taken off. The legal form stays part of the identity, so “Canaccord Genuity Inc.” and “Canaccord Genuity LLC” are different names, and a name with no suffix does not merge with a suffixed one unless a person has checked. One such merge was checked and made: “JPMorgan Chase Bank” and “JPMorgan Chase Bank, N.A.” in releases that name the national bank.

Anonymised respondents are excluded. Many decisions of the AMF (the French markets authority), mostly before about 2017, publish respondents only as placeholders such as “X”, “Y”, “A”, “U and V”, “Company X” or “M. A”, which mean a different person in every decision. 485 defendant entries (46 distinct strings, all in AMF records) are of this kind. They are never matched across records and take no part in any count here, so repeat appearances among those respondents cannot be seen; named AMF respondents, mostly companies, are matched like any other name.

Names were then counted at three levels.

Records. A name in two or more case records.

Matters. A name in two or more matters, using the library’s own grouping rule from src/lib/matters.ts, which links records of one scheme (a filing release, a settlement, a judgment, sometimes from different agencies) on strong evidence only. The script copies the rule unchanged. It gives 2,260 matters for 2,637 records.

Episodes. The matter rule misses many pairs, so every one of the 129 names in two or more matters was read by hand: the records, the release numbers, and for the doubtful ones the cached regulator document. Where the records are a filing and its judgment, a follow-on proceeding, or parallel orders from two regulators on the same conduct, the name was counted as one episode. Where they show a different order, scheme, period or court, it was counted as two or more. The calls are listed, with a note for each, in scripts/analysis/repeat-actors.verified.json, and the script reads that file, so the numbers can be re-run.

Two kinds of name check were also made by hand. 28 groups of spellings were confirmed as one person or firm by reading the records (a middle name in one record and not another, a nickname, a legal form given in one record and not another). 20 pairs that look alike were deliberately left apart, for example the same name with a different legal form, a common name with no link between the records, and “Marc Judah Bistricer” in an Ontario case and “Marc Bistricer” in an SEC order, which may be the same person but could not be confirmed from the cached tribunal page.

As of 2026-10-05. The figures are from a re-run on 2,637 records after three rounds of changes. First, the status and flag corrections of 2026-10-04: 543 of the 733 records that had status “filed” were resolved (190 remain “filed”), and name fixes were made (the “Royal Bank” record now names Royal Bank of Canada, “Standard Registrar and Transfer Co., Inc.” is one name, and three names that began with “and” were cleaned). The new narratives and timeline events let the library’s matter rule join more records: matters fell from 2,162 to 2,051 and names in two or more matters from 304 to 126, while the counts of names in two or more records (666) and in separate episodes (43) did not change. Second, on 2026-10-05, 209 records from the AMF (France) were added, which brings matters to 2,260, names in two or more records to 669, in two or more matters to 129 and in separate episodes to 46. The three added episode names are HSBC France, Natixis and 3Red Trading LLC, each an AMF decision alongside a record on different conduct (3Red Trading was sanctioned by the AMF for orders in a French index future and by the CFTC for spoofing in US futures, so the two are counted as separate episodes). Records the status pass flagged as duplicates of one matter (for example two records of one SEC complaint and its judgment) were checked: each falls in one episode, so duplicates raise the record count and cannot create a repeat.

What it shows

Most repeats are one case

Of the 129 names in two or more matters, 84 are in one episode. Agridime is the plainest example. The firm and its two co-founders appear in four records: a CFTC complaint (May 2024) and consent order (June 2025), and an SEC complaint (December 2023) and judgments (September 2025). The matter rule now joins all four into one matter, which fits what the records describe as one cattle-investment scheme; before the records were updated on 2026-10-04 it counted them as four. A different shape is the follow-on proceeding, which the rule still misses. An SEC administrative decision of 2019 imposes an industry bar on a former salesperson on the strength of an injunction entered in a civil case that is already in the library, and cites it (the 2015 case, the 2019 decision). Both are the same name in two matters and one episode. The rule also errs the other way once: it joins the two records of one individual, a 2015 subpoena-enforcement action and a 2026 judgment in a separate case, into one matter, and that name was counted as two episodes by hand. So 45 of the 46 names in separate episodes are also in two or more matters.

So 208 matters (9%) involve a name that appears in two or more matters, but that is mostly the same case seen twice. Counting only names in separate episodes, 90 of 2,260 matters (4.0%) and 102 of 2,637 records (3.9%) involve a repeat name. Firms account for 69 of those matters (3.1%) and individuals for 26 (1.2%); five matters have both. The 23 banks, securities firms, exchanges and the transfer agent appear in 53 matters (2.3%).

The names that come back are mostly banks and securities firms

Of the 46 names, 39 appear in two episodes, four in three and three in four. The three with four are JPMorgan Chase Bank, Citigroup Global Markets Inc. and Deutsche Bank Securities Inc. The firms with three or more:

These count each legal entity separately, as the records do. Banking groups hold several, and when the entities of one group are joined by name the totals rise: Citigroup and Citibank to seven episodes, JPMorgan and Deutsche Bank to six, Morgan Stanley to five. The records do not say which entities share a parent, so that grouping is the library’s own, built from names.

Two cautions on the table. First, one record can serve several names: the CFTC’s order record of November 2014 on foreign exchange benchmarks names five banks, and counts once for each. Second, the entries are not the same kind of event. Citigroup Global Markets is a respondent in four SEC and CFTC orders on, among other things, hedge-fund disclosures, surveillance gaps, failure to supervise spoofing traders and mismarking; JPMorgan Chase Bank’s four include a 2013 swaps-trading order and a 2020 order on precious metals and Treasuries. Each episode concerns different traders, products and years.

Different regulators, different conduct

Thirteen of the 46 names have their separate episodes opened by more than one regulator, and all thirteen are firms. Eleven have a record from a regulator outside the United States (the FCA, the Ontario Securities Commission or the French AMF): nine banks, a Canadian securities dealer and a US trading firm. In several of those, the non-US order and a US order are the same episode, as in the 2014 foreign exchange settlements, so the second regulator is not always a second episode. Among the 31 names with at least two episodes that carry a technique tag, 17 share no tag across their episodes: the library tags one episode as, say, spoofing and another as something else. 14 share at least one. Fifteen of the 46 have fewer than two tagged episodes, because some orders concern supervision or reporting failures to which the library applies no technique. The median gap between the first and last episode of a name is three years. The longest is 11, an individual.

Gatekeepers

The library’s records name few gatekeepers in the narrow sense, and fewer still twice.

• Transfer agents. Three, each in a settled SEC order (2014 and 2016, 2017, 2018). Only one, Empire Stock Transfer, appears more than once, in two separate orders: one on undisclosed control of the firm, one on unregistered distributions of four penny stocks. An officer of the firm, a vice president, is also a respondent in both.
• Audit firms. Five China-based firms, in one SEC administrative proceeding of 2014 about audit work papers. None repeats.
• Law firms. Five, in five unrelated records, two of them subpoena-enforcement actions (the record for one, against a large firm, says no violation is charged). None repeats as a firm. See also the earlier post on reverse-merger gatekeepers.
• Banks and broker-dealers. These are where the repeats concentrate: 20 of the 46 names are banks or securities firms, and with two stock-exchange entities and the transfer agent they make 23.

Individuals

Fifteen individuals appear in separate episodes: eleven in SEC cases, four in CFTC cases. Three are the owner, an officer or the principal of a firm that is itself on the list. One is the owner of a Houston introducing broker whose settled CFTC orders of 2019 and 2021 are separate; the second order describes conduct that continued after the first, and the respondents admitted its facts. Two pairs of individuals come together in the same records. Two more were counted in a subpoena-enforcement action and later in a complaint, with nothing in the records linking the two. The other six each have two or three proceedings on different conduct. This post names no individual: most of the records describe allegations or settled orders, and a count does not need names.

Named in someone else’s record

Firms also come up in the text of records where they are not respondents. Twenty-two records name one of eleven large banking groups (Citigroup, JPMorgan, Deutsche Bank, UBS, Royal Bank of Scotland, Barclays, HSBC, Goldman Sachs, Morgan Stanley, Merrill Lynch and Bank of America, BNP Paribas) in their text without naming any of its entities as a defendant. One is the CFTC’s January 2018 announcement of eight actions, which names three banks but no defendants. Fifteen of the rest are actions against people described as former employees of the bank, such as former traders sanctioned individually; two concern an insider-trading tip that traced to deal information at J.P. Morgan (a 2015 case and a 2026 case against one of its defendants); one is a firm’s order that mentions a bank’s traders; three are AMF decisions that mention a bank in passing (an employee of one, a credit line, a share traded). None of the twenty-one charges the bank. These are mentions, not respondent roles, and they are not in the counts above.

How sensitive the counts are

• Without the hand-verified spelling merges the figures are 653 names in two or more records, 106 in two or more matters, and 43 in separate episodes. The merges add three names to the 43.
• Two names could not be settled. One appears in two SEC records of 2020 and 2025, the other in a 2023 and a 2024 CFTC record, and the cached releases do not say whether each pair is one matter or two. Counted as separate, the total is 48.
• Three names were assumed to be one episode though the records do not state it: a CFTC consent order and a later CFTC judgment on a Texas managed-account fraud, an SEC follow-on after a criminal conviction paired with a CFTC judgment on a commodity pool, and a run of SEC records on a ticket-resale scheme. If any is two episodes the total rises by one.
• Four pairs of names always travel together (two stock-exchange entities, an introducing broker and its owner, a boiler room and its principal, two individuals in the same pair of cases). Counted as parties, the 46 names are 42.
• Names that run together in the data. A crude check (individual entries containing “and”, ”&”, ”/” or ”;”) flags 25 of the 4,822 entries outside the AMF records, about one in 190, as possibly holding several people in one string; entity names containing “and” were not counted. Such entries are treated as one name, which can hide a repeat but cannot create one.

What this does not show

It is a count of cases regulators chose to bring and publish. A name that never appears has not been shown to have done nothing, and a name that appears more than once has not been shown to be worse than one that appears once. The library holds records from six regulators (the SEC supplies 1,860 of the 2,637, the CFTC 338, the French AMF 209, ASIC 113, the Ontario Securities Commission 69 and the FCA 48), with different publication habits and different start years, so a firm active under a regulator that publishes more will appear more often.

Anonymised respondents cannot be matched. 485 defendant entries in AMF decisions are placeholders (X, Y, A and similar) and are excluded, so a repeat among those respondents is invisible and AMF repeats are under-counted relative to the regulators that publish names.

It has no denominator for activity. A bank that trades in more markets than a boutique has more chances to appear. The library holds no data on firm size, trading volume or employees, so it cannot rank firms by how often they appeared relative to what they do.

A repeat name is not a repeat finding. Some orders are settled without admitting or denying the findings; some matters are filed and have no recorded outcome (190 of the 2,637 records carry the status “filed” as of 2026-10-05); one repeat name, Cormark, appears in a tribunal case in which none of the allegations were found proven. The defendants field also lists relief defendants, and the library does not tag the role of each name.

The episode calls are judgements by an AI agent. The 129 names were read by a Claude AI agent working from the case records and, for some, the cached regulator documents; the 28 name merges and 20 near-misses were decided the same way. No lawyer reviewed them. The rule for “one episode” is a rule of thumb (a filing and its judgment, a follow-on, parallel orders on the same conduct), and a reasonable reader could draw a few lines differently. Names that sit in a single matter but cover two episodes cannot be seen: the matter rule only merges, so it can hide a repeat inside a matter (one name is an example). The list of settled calls, near-misses and assumptions is in the verified file.

The data has gaps. personSlug is null on every record, so people are matched by name alone and two people with the same name could be merged. Defendant lists are taken from the records as published and can still omit or mis-state a name; a repeat can hide behind an unlisted or differently spelled name. Several corporate groups are joined by name, not by ownership. Records change as corrections are made: a count here can move by a name or two between runs.

It is not legal advice or a ranking. If you find a record this analysis got wrong, use the error link on its page, or write to [email protected] with the slug and a link to the document; it will be logged on the corrections page. The earlier post What reading every record found explains how the records were checked.

Re-running it

node scripts/analysis/repeat-actors.mjs out.json reads src/content/cases/*.json and scripts/analysis/repeat-actors.verified.json and writes every number used here. Setting NO_ALIASES=1 skips the person-name merges to give the sensitivity figures above. Five of the headline figures (the 2,428 records, 4,822 defendant entries, 190 “filed” records, the record counts of five repeat names and the agency totals) were recomputed with a separate throwaway script and matched.

Techniques referenced

• Spoofing — Spoofing is placing orders a trader intends to cancel before execution, to create a false impression of supply or demand and move the price toward a smaller genuine order on the other side.
• Benchmark submission rigging — Benchmark submission rigging is skewing the figures a panel member reports into a reference rate calculation, so that the published benchmark moves in favour of positions that settle against it.
• FX fixing — FX fixing manipulation is trading during the short window used to calculate a daily currency benchmark, using advance knowledge of client orders that must be executed at that fix.
• The naked short selling debate — Naked short selling is selling short without having borrowed the security or arranged to borrow it, which is restricted by Regulation SHO and whose real-world scale and price impact remain genuinely contested.

Cases referenced

Reviewed October 4, 2026. Spotted an error? Tell us.