CAPÍTULO 1. FUNDAMENTACIÓN TEÓRICA
1.2. Calidad del Software
The results of the community detection algorithm are thus an informed guess about the latent forces represented by the two corporate networks. There is no guarantee that they truly represent the global optimum. Even if they did, there is doubt whether this optimum would correspond to an intuitive idea about business groups. To mitigate these concerns, I attempt to establish the social rel-evance of each community by measuring the presence of three potential sources of solidarity: family control, government ownership, and geography in addition to the potentially confounding influence of industry homophily.
Recalling the definitions presented above, business groups are more than random collections of firms cooperating on an ad hoc or transactional basis.
Rather, they are cohesive and enduring communities of organizations, and such solidarity can in some cases be measured. I consider measure family control by grouping individuals by shared or highly similar surnames, government by hand-coding owners as state or non-state, and geography by accounting for the appearance of the same individuals in records from different exchanges. These measures are based on all individuals and firms tied to a community, including those deleted before running the community detection algorithm.
Family measures are based on assessing the extent and exclusivity of a fam-ily’s involvement with a particular community. It includes the following mea-sures: the percent of firms with at least one family member as an owner, director, or manager; mean family ownership; the family’s share of managers, chairmen, and directors; the percent of firms in which the family collectively owns a 10%
stake4; and the average number of unique individuals associated with the major surnames in that community. For each group, I assess family control based on the combined scores for the top two families measured in terms of their overall involvement. I attempt to break ties in favor or larger families. Note that these measures cannot account for family control of private companies and are thus likely to understate family ownership through these intermediaries.
I measure government ownership by hand-coding all owners based on their names and web searches. This includes clear government entities like ministries of investment or finance as well as state-owned enterprises. Just as in the case of family ownership, I construct measures of mean government ownership and
4This cutoff represents the amount of ownership necessary for significant control over a com-pany (La Porta et al. 1999; Claessens et al. 2000)
the percentage of firms in which government entities own at least 10% or 20%.
I do not attempt to code the political affiliation of individuals.
Location is another potential source of solidarity, and I develop a series of location scores for each node based on its affiliations with companies listed on different exchanges. Using an indirect proxy for true country of origin is not likely to be accurate in every case, but on the whole it captures key patterns indicating geographic diversity at the community level. For example, a director for a single Egyptian firm would be classified as Egyptian while a director for two Kuwaiti firms, one Egyptian firm, and an Iraqi firm would be classified as Kuwaiti. This allows us to assign each node one or more top countries while also measuring its geographic homogeneity.
Using these measures for each listed firm, I generate three of geographic ho-mogeneity measures. I focus on geographic hoho-mogeneity rather than a binary foreign-domestic variable due to the difficulty of maintaining this classification for multi-country data. A geographically heterogeneous community is charac-terized by ownership and control relationships that span national boundaries, and given the otherwise strong tendency for edges to occur within close geo-graphical proximity, this type of anomaly might indicate a strong relationship.
The first measure, top exchange score, is the percentage of firms that are listed on the most common stock market within that community. The next measure includes information from each firm’s activity in multiple markets and is the mean strength of each listed firm’s affiliation with its top market, hereafter re-ferred to as mean firm domesticity. The final statistic is the mean strength of each listed firm’s affiliation with the the most common stock market from the top exchange score, i.e. that group’s “home” exchange. The difference between
these two measures is that this mean home exchange score includes the same broader set of information as mean firm domesticity.
Finally, industry is also a likely source of inter-firm connections due to ho-mophily, or the tendency of nodes to form ties with similar alters (McPherson et al. 2001). I broadly assess absolute homophily at the group level as the per-cent of public firms in the community belonging to the most common top-level SIC code in that community. Since financial firms are over-represented on the stock markets of the region, I also include the the relative degree of industry concentration in the form of the euclidean distance between a vector of the per-cent of community firms within each category and the same vector defined for all public firms in that community’s home country.