How we calculate the bankruptcy risk

The bankruptcy risk is the probability that a company goes bankrupt within the next 12 months. It is calculated with a statistical model that has learned from 141,080 Danish bankruptcies which characteristics typically precede a bankruptcy.

Data

We only use public data: company information from the Danish CVR register (addresses, management, owners, auditor, status and history) and every annual report the companies have filed with the Danish Business Authority – including the digital XBRL reports with equity, profit, assets, debt and the auditor's statement. The dataset covers 829,466 private limited, public limited and entrepreneurial companies, including every company that has ever been bankrupt.

What the model looks at

• Filing: annual reports filed late, missing reports, restated reports and dissolution warnings from the Danish Business Authority.
• Financials: equity, profit, equity ratio, current ratio, return on assets, capital loss and the change from the previous year.
• Auditor: the type of auditor's statement, qualifications and remarks about going-concern uncertainty, and auditor changes.
• Management: CEO changes, the number of CEOs and the CEO's previous bankruptcies in other companies.
• Address: the number of moves, moves between municipalities, c/o addresses and the number of companies at the same address.
• Structure: age, company type, capital, industry, region, number of employees and changes in employees.

How the model is trained

The model is a gradient boosting model. For every bankrupt company it looks at the information as it was 3–15 months before the company first ran into distress (compulsory dissolution, restructuring or bankruptcy). These are compared with active companies. This way the model learns the early warning signs – not the registrations that only appear once the bankruptcy is already a fact.

The probabilities are calibrated to match the actual frequency of bankruptcies: among active Danish companies about 1.4% go bankrupt within a year. The model achieves an AUC of 0.927 on companies it was not trained on.

Risk levels

Limitations

The risk is a statistical estimate based on public information and not a credit rating or an opinion on the individual company's finances. The model does not know about anything that is not in the CVR register and the annual reports – such as new orders, financing or payment remarks. A newly founded company without annual reports is assessed on less information. Always use several sources before making decisions.