The full dataset used in this paper brings together financial and subsidy data from the OECD MAGIC database, which are collected at the consolidated group level, and data from the Steel Committee steelmaking Capacity Database, which is collected at the plant level. Capacity data at the plant level are thus aggregated at the consolidated group level, taking into account the ever-evolving ownership structure of individual plants over time by using an Ownership database built over the previous years by the Secretariat. The merged dataset forms an unbalanced panel with observations covering 47 firms across the period 2005 to 2022, wherever applicable and available.
To control for observed and unobserved firm characteristics that are time-invariant (e.g. firms’ government ownership, their political connections, or their geographical origin), and thus avoid potential bias stemming from omitted variables, the analysis includes firm fixed effects across several specifications. The variable BMB is defined as:
where y denotes years, i individual companies, and avg_int stands for firms’ average effective rates of interest on their outstanding debt. Base rates used in the calculation of BMB will typically account for most of the yearly variation in BMB values. They are also identical across all firms having the same credit rating and borrowing in the same currency. This causes BMB to be highly correlated with year fixed effects across several specifications.
Instead, the analysis controls for the time dimension by adding a ‘market’ variable, which corresponds to the sum of the revenue of all firms in the MAGIC database sample in year y for the steel sector, or to a broad yet steel specific GDP-like variable. The total revenue variable is defined as , where n is the number of steel firms, while the steel_GDP variable is defined by:
Those two ad-hoc steel market variables serve a similar role (and hence should probably not be used in conjunction). Their aim is to control for global factors such as changes in steel prices as well as the business cycle affecting both sectors (i.e. the general level of activity or broad market conditions in either sector), either through the total revenue earned by firms in the sample, or through a more direct proxy for global steel demand conditions. In that sense, can be seen as a control variable for the “steel business cycle”.
Given that the calculation of BMB involves companies’ change in outstanding debt, it is closely related with the change of the size of firms, which in turn correlates with capacity change. This is the reason why in some of the regressions, in particular those studying the relationship between BMB and total debt of the firm, it was more appropriate to “scale” back variables of interests by dividing by asset size.