Credit Risk and Machine Learning Concepts -4
Rule-based matrix determination of Credit Limits and trading terms
The previous installments introduced the concepts and what would be the foundations of an approach to rate a financial stability and credit-worthiness score on an entity using purely arithmetic key ratios, applying appropriate weightings to this and some other information to determine a credit limit and score band. Here I will briefly touch on some stochastic approaches to contrast against the qualitive liquidity and performance metrics, and show how these all can be incorporated into a complex rule-based instruction set and solution as I have recently built out for an international supply-chain organization.
A classic stochastic method to evaluate the financial stability of a publicly traded enterprise is the Merton Model, and similar models such as Moody’s KVM. In simple terms, we are calculating the probability of the entity defaulting on meeting all debt obligations at a future date.
In this model the value of the firms assets are assumed to obey a lognormal diffusion process with constant volatility. An approximation of non-securitized debt is the modification that is for commercial credit risk rating, which differs from the securities equity and debt, treating the debt as a pure discount bond where a payment of D is promised at time T. If the firm’s asset value exceeds the promised payments D at time T, the debt will most likely be paid. If the asset value is less than the promised payment the firm is likely to default a payment.
The firm’s Equity is defined as E. The value of the firm’s tangible assets are A. If E0 and A0 represent the values for Equity and tangible assets today, and ET and AT are the values at time T.
The payment is given by : ET = max[AT — D,0]. Further standard calculations would determine equity and by setting a constant value of the assumed interest rate and asset volatility, additional steps can be taken to determine the risk-neutral Probability that a company will default by time T.
An alternative implementation of the Merton model uses two implied volatilities, considering the effect
of an expression of Leverage (L). If D* = De -rT is defined as the present value of the promised
debt payment, then the leverage ratio is derived through : L = D* / A0
The Rating and Analyst organizations such as Moody’s, Fitch, Standard and Poor apply additional factors that are proprietary to arrive at their ratings for Publicly traded companies, which include :
· Macroeconomic conditions
· Competitiveness
· Profitability Strategy
· Capital structure Corporate governance
· Quantitative financial analysis Value chain
· Financial ratios Technology and R&D
· Liquidity Other operational aspects
· Risk
· Group impact and Industry Segment trends
A few of the matrix that consider risk evaluation are shown here:
Four (4) Rating axes are commonly used in support of banking Basle II initiative requirements.
The Financial models, which address 6 factors that indicate risk from prior periods:
The 6 factors that are now more generally accepted as good indicators for credit risk, are :
Profitability; Capital Structure; Liquidity; Activity; Growth and Size. The weighting of these factors respectively is 26%. 24%. 14%. 13%. 13%, 10%. The ‘freshness’ of the financials are also very important.
2. Internal Behavioral Model — The operational rating and behavior with this customer and the organization
3. External Behavioral Model — the behavior with other organizations, reference metrics and general conditions.
4. Qualitative Mode — Expert judgement of the relationship and non-numeric indicators.
Upward or Downward Credit Risk Factors would then be applied.
Operational performance relates to segmentation of customers, and apply continuous monitoring to ensure that the rating positions in the grid below are retained, and focus alerting and monitoring effots t the high risk and low operational ratings, and determine why the high operational rating customers with low risk rating (therefore considered more unstable financially) are so graded.
The reasoning for accepting low operationally rated customers that also are low risk volatility are also to be understood to ensure that they are not impacting negatively the overall cash flow or cash availability to the organization offering credit terms.
The rating framework I have put together has taken into account many of these drivers and applied traditional rating (such as commercial credit and financial stability) scores with operational ratings, so that we may concentrate on the High score but low-performing payment history clients and the high performing but low rated clients. It is shown on analysis of the data that the 4 boxes are appropriate clustering of customer behaviors as clients with accounts receivables and credit terms extended to them.
Observation and clustering has enabled a weighting and an ‘adder and subtractor’ or ‘nudges’ approach to consider the CRA ratings as a starting point, additional financial data provided by the customer and trend information both from the CRA reporting source and the trading history to provide a calculated value that maps into the organizations bands for Credit Limit and Credit term setting with appropriate delegation of approval authority on credit limit range values.
The previous 3 installments may be found here :
The next installment addresse the perspective of a Credit Risk Analyst and may be found here:
