Credit Risk Assessment Methods
Five key factors in our credit risk model, and why machine-learning outperforms traditional approaches.
Our credit risk model
The core of credit risk assessment
Our credit risk assessment identifies five main factors that influence a company's credit risk profile. These are assessed using machine-learning algorithms trained on historical financial data and bankruptcy outcomes — dynamically weighting each factor based on the individual company's circumstances.
Solidity
Equity Ratio
The ratio between equity and total balance sheet. Measures the proportion of assets financed by shareholder equity — a key indicator of a company's financial buffer against losses.
Profitability
Return on Assets (ROA %)
EBIT divided by total assets. Measures how efficiently the company generates profit from its asset base. Higher ROA indicates stronger earning power relative to the balance sheet.
Liquidity
Quick Ratio
Liquid assets (cash etc.) divided by short-term liabilities. Measures the company's ability to meet short-term obligations with immediately available resources.
Industry
Sector risk level
Different sectors experience varying macroeconomic sensitivity and failure frequencies. Industry risk raises the bar — companies in riskier sectors need stronger financials to achieve equivalent ratings.
Company Size
Balance sheet & revenue scale
Larger organizations typically demonstrate greater resilience to disruptions due to scale and diversification. Size is factored into the risk model as one of several variables.
Machine learning approach
Why and how we use machine learning
Our model employs machine-learning algorithms (XGBoost) trained on historical financial data and bankruptcy outcomes. The key advantage over traditional fixed-weight models is dynamic weighting: the model adjusts the importance of each variable based on the individual company's other characteristics.
For example: liquidity is weighted more heavily for unprofitable companies, while solvency matters most for companies with high business volumes relative to equity. This individual-level adaptation is impossible with traditional methods.
Important limitation
The model currently lacks payment behavior data (invoice payment delays), which would serve as an additional early-warning signal. This limitation is noted in our assessments. Model performance could be further improved when this data becomes available in Denmark.