We relied on Bureau van Dijk’s Orbis Database for MNE- and affiliate-level data. The sample contains the 2,000 largest global publicly traded firms as of December 2020. In order to select the 2,000 largest firms, we applied the screening criteria in Orbis for: 1) active firms, 2) publicly listed, and 3) with revenues over USD $4.2 billion. These screens resulted in a total of 2,000 MNEs from a broad range of industries, headquartered in 63 countries, and with a median total revenue of USD $9.2 billion and median total number of employees of 24,500.

Orbis data sometimes contains subsidiaries that do not have ownership identifying information in the shareholder data and therefore a mass downloading of the 2000 MNE subsidiary data yields many subsidiaries with no ownership identifiers. Pulling each MNE’s subsidiaries separately allows for assigning the correct MNE identifier to each subsidiary without loss of observations. Therefore, we screened for the subsidiaries of each MNE individually, selecting subsidiaries that meet one of the following criteria: 1) Ultimately owned by a company of the selected group which is a Global Ultimate Owner; 2) Ultimately owned by a company of the selected group which is a Domestic Ultimate Owners; 3) Directly or totally owned by a company of the selected group with a minimum of five percent to a maximum of 100 percent ownership; or 4) Owned by a company of the selected group with an unknown percentage.

The database contained 545,664 subsidiaries belonging to the 2,000 MNEs, of which 470,867 were currently active and only 229,632 had recent operational accounts over our sample window. The proportion of Orbis entities with revenue and employee data over our sample window is 37.7 percent and 33.8 percent, respectively.

A major part of the task was cleaning the affiliate data. Using data from the Orbis Ownership database, we found that the 545,664 entities had over 8.5 million entity-shareholder-year observations. Orbis contains separate identifiers and ownership percentages for the global ultimate owner, direct ultimate owner, controlling shareholder, and shareholders. We relied on the shareholder ownership information to track ownership percentage changes over time. The Orbis Ownership database is updated with information when there is a change in ownership. As a consequence, there are gaps in ownership information between years of updating. For entities with gaps between years of the same owner, we filled in missing ownership data with the most recent prior percentage of ownership reported. For entities that did not have a shareholder ownership percentage associated with the MNE identifier code, we relied on the global ultimate owner, the controlling shareholder, or direct shareholder as long as the owner identification code matched the MNE identification code and the ownership percentage was not missing. The global ultimate owner ownership percentages are static in the Orbis database, equal to ownership as of 2019. Therefore, we had to check whether the subsidiary entered the corporate group during our sample window for all of these affiliates.

In some cases, Orbis provides no ownership percentage, but the affiliates are categorized into independence ratings based on ownership percentage (e.g. B, C, and D ratings). We took the median value of the ownership range for the independence ratings for these firms. When information on subsidiary ownership was not available or provided conflicting information, we validated ownership using information from public filings and reports.

A total of 11,112 subsidiaries had more than one MNE owner according the Orbis data base. In examining the data, there were a number of cases where MNEs make financial investments without exerting control over the subsidiary, particularly in the case of PE firms and institutional investors. A team of RAs cleaned these data, assigning ownership to the MNE that aligned with the GUO, DUO, or CSH Orbis Data. If the controlling ownership could not be directly matched, ownership was assigned to the MNE with the largest percentage ownership of at least 25%. The final sample was composed of 10,874 subsidiaries with multiple MNE owners and 534,790 subsidiaries with only one MNE owner.

We used information on subsidiary incorporation dates and legal changes from the Orbis database to code when subsidiaries entered or exited the multinational firm group. For example, we searched for keywords associated with ['Merger', 'ownership added: parent', 'the company has merged with', 'ownership changed from shareholder to parent company'] for acquisitions of the subsidiary and ['liquidate', 'Deregistration', 'Cessation of activity', 'Insolvency proceedings', 'Bankruptcy', 'Dissolved',] for dissolutions and spin offs.

A total of 119 of the publicly listed MNEs were part of a larger conglomerate group (e.g. Hitachi High Technologies Corporation and Hitachi Construction Machinery Co., Ltd. are publicly listed but also part of Hitachi Ltd.) or acquired during our sample period. We rolled these MNEs up into their larger corporate group based on the appropriate legal event date.

Eight MNEs did not have affiliates in the Orbis database. As a result, 1,891 MNEs compose the final sample of MNEs with economic importance information.