Geographic Investing: Stock Return Indexes Based on Company Operations

Abstract

Portfolio allocations to firms of various geographic areas should be guided by underlying risks of operations. In most statistical studies of international stock returns, a firm is included in a country’s index if its headquarters is located in that country, a classification scheme that ignores the operations of the firm taking place in multiple geographic areas. In prior work, we have proposed a model of country factors that is based on the business activities of all firms operating in a country, be they domestic firms or multinationals. In the present paper, we compare the resulting indexes with the domestic revenue exposure indexes already available in the industry. We conclude that our new indexes allow a portfolio manager to track geographic risk much more accurately.

Keywords:

• country risks
• factor model
• geographic segments
• portfolio focus
• risk management

PL Credits: 2.0:

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Acknowledgments

We are grateful to Humberto Gomez, a master of science student at the University of Lausanne, who first set up the MatLab programs for an earlier draft of this paper. Further research assistance was provided by Fiodor Gorokhovik and Pierre Poulain. We thank Philippe O. Piette of WVB, who generously provided the data. Marco Del Negro provided us with an expectation-maximization computer code; we are grateful to him. Dumas’s and Gabuniya’s work received the support of a grant from the INSEAD Research Fund. Dumas and Marston received funding on this project from the INSEAD-Wharton Alliance (Project # 2399-232).

Notes

1 All the same, we are not denying that there may independently exist risk factors arising from being listed in one country. See Froot and Dabora (1999) and Chaieb, Langlois, and Scaillet (2021).

2 Some or all of the database is also available from Bureau van Dijk, under the name “Osiris.”

3 The selection and filtering of firms, based on geographic-segment data from World Vest Base and stock-return data from Datastream, is explained in detail in Dumas, Gabuniya, and Marston (2022). The most important filter applied to the original dataset deleted all microcaps. That filter alone reduced the number of securities in 2014 from 17,678 to 6,690. The filtering process gave us a set of firms that grows from 1,797 in 1999 to 6,335 in 2014.

4 We adopt a permanent list: France, Germany, Great Britain, Brazil, United States, Canada, Australia, Malaysia, Singapore, China, Japan, India, and Rest of World. In 2014, the stock markets of the 12 countries specifically identified represented 74.8% of the total capitalization of the world’s stock markets.

5 See https://www.isin.org/isin/: “A two-letter country code, drawn from a list (ISO 6166) prepared by the International Organization for Standardization (ISO). This code is assigned according to the location of a company's head office.”

6 Our factor model is a significant elaboration over Brooks and Del Negro (2004), who considered a model with country factors (and a separate world factor, which we do not need here because our country factors are not assumed to be independent), but with restrictions on the loadings (contained in the matrix B below) that differ from ours. They fix the loadings on foreign factors at 0 and they force the country factors to be independent of each other so that all common movements in countries take place through the world factor. Similarly, Heston and Rouwenhorst (1994) fix the loading of a firm on its country to be equal to 1.

7 The exact statement of assumption 1 is more complicated than stated here. A summation is used for our 13th “country” that we call Rest of the World, which obviously includes several actual countries. In some cases, the equality constraint is replaced by an inequality because the information in the annual reports often refers to one of several actual countries.

8 We take care of financial leverage, albeit imperfectly, by deleveraging the stock returns in a simple way: stock returns × Equity/(Equity + Debt).

9 See Dempster, Laird, and Rubin (1977) and Rubin and Thayer (1982).

10 More precisely, the algorithm delivers the expected value of C, which is unobserved, given R, which is observed.

Additional information

Funding

Dumas’s and Gabuniya’s work received the support of a grant from the INSEAD Research Fund. Dumas and Marston received funding on this project from the INSEAD-Wharton Alliance.

Notes on contributors

Bernard Dumas

Bernard Dumas is at INSEAD, National Bureau of Economic Research, and Centre for Economic Policy Research, Fontainebleau, France.

Tymur Gabuniya

Tymur Gabuniya is an MSc student at University College London, London, UK.

Richard C. Marston

Richard C. Marston is in the Wharton School of the University of Pennsylvania, Philadelphia, PA.