*****************************************************
NOTE: Please read the technical descriptions below before using the data.  

*****************

This file accompanies the all-firms corporate ownership data set compiled as part of the ownership project "Mavericks, Universal and Common Owners - The Largest Shareholders of U.S. Public Firms", Amel-Zadeh, Kasperk, Schmalz (2022) and describes where the data comes from, the papers that should be cited when providing academic references, and some very important technical details regarding its usage.
Please read the technical details in full before using this data.  These details are critically important to ensure proper usage.

This data is a quarterly time series of all ownership stakes, in firms incorporated and publicly traded in the U.S. in the period 2000Q4 - 2020Q4, that have to be reported to the Securities Exchange Commission. In particular, it includes ownership by institutional investment managers (so called 13F filers), corporate insiders (Form 345 filers) and blockholders owning more than 5 percent of outstanding shares in an individual firm (13DG filers). It covers the whole cross section of U.S. public companies, single class and dual class alike: 58,365,351 owner-firm-quarter observations, 7,156 companies and 117,166 owners.

NOTE: The holdings in every quarter are as of the quarter end.

NOTE: This data is at the firm (company) level, not the security level. An owner's holdings in all share classes of a company are summed into one stake, so the data set carries permco but NO permno and NO cusip - a security level identifier would require splitting every row again.

The provided variables are as follows:
permco      -   permanent company identifier
coname      -   company name
ownercik    -   owner identifier (central index key)
ownername   -   owner name as reported on filings
beta        -   fraction of shares outstanding owned, over all share classes
gamma       -   fraction of voting rights outstanding owned, over all share classes
rdate       -   end of quarter when shares owned
fdate       -   actual date when ownership is reported
filed_13f   -   binary flag if owner filed a 13F report this quarter
filed_13d   -   binary flag if owner filed a 13D report this quarter
filed_13g   -   binary flag if owner filed a 13G report this quarter
filed_f345  -   binary flag if owner filed a Form 3/4/5 report this quarter
source      -   filing type on which this ownership stake was reported
is_dual     -   binary flag if the company had more than one class of stock this quarter
is_sp500    -   binary flag if the company was in the S&P 500 index this quarter
sic         -   SIC industry code of the company
outlier     -   binary flag for firm-quarters with implausible aggregate voting ownership

All binary flags are stored as the text True / False. source takes the values '13f', 'f345' and '13dg'; where the share classes of one holding were reported on different filing types, the contributing types are joined with a '*' (e.g. '13f*f345'). The filed_ flags are recorded before duplicate reports are dropped, so they still show every filing type that reported the holding, even where only the highest priority report survives in source.

***************************************************************************************Background ****************************************************************************
***************************************************************************************

For an extensive description of this data, please read the data and methodology sections of the paper noted below.  Here is a brief description.

We compile this data set from the ownership reports filed with and stored on the Edgar database of the Securities Exchange Commission (SEC). The information on holdings comes from the following reports:
Institutional investment management firms with >$100m AUM - 13F-HR (13F)
Blockholders, owning >5% of outstanding equity - SC 13D, SC 13G (13DG)
Corporate insiders, executives, directors, 10% owners - Form 3, Form 4, Form 5 (F345)

NOTE: While 13F filings are reported every quarter, 13DG and F345 are reported only upon changes in holdings, so no report does not mean no holding, but an unchanged number of shares held. We carry these holdings forward from one filing to the next, and extrapolate a last holding for at most 12 quarters (3 years) beyond the final report.

First we extract all the holdings information from 13F, 13DG and F345 filings with a parsing script for each respective file type; the 13F parser is an extension and improvement of the parsing code used by Backus et al (2021). Then we continue processing the parsed information in separate pipelines to:
- identify the securities that we include in our sample with a unique and persistent key and link each security to relevant quarterly information (shares outstanding, voting rights per class, type of security, country of incorporation of company)
- aggregate each data source to have a unique permco-class-rdate-ownercik key, i.e. to a single ownership stake per quarter for every firm-share class-owner combination.
Finally we merge the three separate data sets into one. Where several filing types report the same holding we keep the highest priority source (13F, then F345, then 13DG), remove the remaining duplicate and overlapping reports, consolidate owner families that file separately but act as one owner into a single ownercik, and sum each owner's holdings across all share classes of a company into the final firm level stake.

Ownership (beta) and control (gamma): beta is the economic, cash flow ownership share - the shares an owner holds in a company, over all of its share classes, divided by all shares outstanding across those classes. gamma is the voting share - the voting rights those same shares confer, divided by all voting rights outstanding. In a single class firm one share carries one vote, so gamma equals beta. The two diverge only for dual class firms, where classes carry different numbers of votes per share: most commonly one class with no vote and one with one vote per share, or a one vote against ten votes per share structure, in extreme cases a ratio as large as 1:10,000. Both shares and votes of traded and non-traded classes are included, so beta and gamma are the owner's full economic and voting stake in the company.

The share and vote counts per class are not available from standard databases, so we collect them by hand: candidates are taken from published dual class lists and from our own parsing of the security descriptions on insider filings, and each is verified quarter by quarter against the company's 10-K and 10-Q filings (shares outstanding per class) and its Def-14A proxy statements (voting rights per class). is_dual marks the quarters in which a company actually had more than one class outstanding, not firms that ever had one.


*************************************************************************************** Citations ****************************************************************************
***************************************************************************************


"Mavericks, Universal, and Common Owners - The Largest Shareholders of U.S. Public Firms", Amel-Zadeh, Kasperk, Schmalz (2022)


***************************************************************************************Technical Details *********************************************************************
***************************************************************************************


Please read the following to ensure proper usage of this data.

The outlier flag marks firm-quarters in which the firm's aggregate voting ownership (the sum of gamma over all its owners that quarter) lies more than 1.96 standard deviations from that same firm's own average aggregate voting ownership over the sample period. Such jumps are usually data errors, but sometimes genuine changes in the stock structure, so we mark rather than drop them. 1,907,493 of the 58,365,351 rows are flagged. We recommend re-running any result excluding them as a robustness check; note that dropping them removes 55 of the 7,156 companies entirely.

The sample is restricted to common equity of U.S. incorporated companies. We start in 2000Q4 because earlier quarters are too incompletely covered by the SEC archive. Insider coverage is thin at the beginning of the sample: electronic Form 3/4/5 filing only became mandatory in 2003, and before 2004 we capture the most significant insiders (holding more than 5% of outstanding equity) through their blockholder filings instead. 13F holdings are dated to the quarter end they report; blockholder and insider holdings are dated to the quarter of the reported ownership event and of the filing respectively.

We further exclude implausibly reported ownership: individual stakes at or above 110% of shares outstanding are removed, as are firms in which a single owner is recorded above 104% in more than one quarter, firms where the ownership of all owners together reaches 200% in any quarter, and firms where it exceeds 150% or falls below 30% in more than eight quarters or in more than a fifth of their quarters; a firm's first or last quarter is dropped if less than 30% of it is captured. S&P 500 firms are exempt from these exclusions and always retained. Coverage is in any case much denser for index members: an S&P 500 firm-quarter has 691 distinct owners on average against 156 for a non-member, which is why S&P 500 firm-quarters make up 26,995,131 of the 58,365,351 rows - about 46% - although only 911 of the 7,156 companies are ever in the index.

We do not guarantee complete accuracy of the data set. We have ascertained through extensive comparisons that our data is more comprehensive and accurate than Thomson Reuters S34 data or Backus, Conlon, Sinkinson corporate ownership data (Backus et al, "Common Ownership in America: 1980-2017", AEJ Micro, 2021) and the quality of our data is on par with Factset and Capital IQ corporate ownership data. 

Just as Factset and Capital IQ have cases where the ownership stakes across all owners in a given firm-quarter sum to more than 100% of outstanding shares, so do we have these cases. In most cases this is driven by 13-F institutional owners who report ownership in shares lent out to other 13-F institutions, who also report these shares. Such cases cannot be identified with our automated procedure, nor are they cleaned out in other sources of ownership data.

We recommend checking the original ownership filings available in the SEC archive in case the ownership stakes reported seem out of proportion, because there may still be mis-parsing or merging issues in our data set. Please get in touch if you identify any such errors and find a way to fix them, so we can correct our data and make a better version available to all researchers in the field.

Although parsing is difficult and can lead to errors on its own, there will always remain inaccuracies due to misfiled ownership reports by investors. We try to correct for any such issues using the variety of ownership reports parsed, however given the current reporting requirements of the SEC there is a limit to how accurate this data can ever be.