RDP 2022-05: The Real Effects of Debt Covenants: Evidence from Australia Appendix A: Construction of Debt Covenants Data

I construct a database on the prevalence and types of debt covenants used by non-financial listed Australian firms by applying text analytic techniques to their publicly available annual reports, collected from the Connect4 website. I write a Python program to first convert the files into readable text and then extract relevant information from the text as follows:

1. I search for the term ‘covenant’ and its inflections in the text. If the search query returns non-empty results, I classify the firm as having debt covenants in that year.
2. I isolate the blocks of text surrounding the mentions of covenants. Figure 1 shows an example of an extracted block of text.
3. In each block of text, I search for keywords (and their inflections) that indicate the possible types of debt covenants (e.g. interest cover, gearing ratio, leverage ratio).
4. For each type of debt covenant, I count the appearances of its indicative keywords. If the counter returns a positive value, I classify the firm as having that particular type of covenant. In the example in Figure 1, the firm mentions three types of debt covenants: equity ratio, leverage ratio and interest cover ratio.
5. Finally, I determine if firms comply with or violate their covenants from the reports by counting the appearances of keywords such as ‘breach’ and ‘violate’ (and their inflections while incorporating negation). The example in Figure 2 suggests that the firm breached its financial covenants in the period to the date of the report.