Credit Risk and Machine Learning Concepts -3
In my first 2 articles on this topic I introduced 3 approaches to credit risk management and discussed some details of the solutions that I had put together for an Organization in the UK before there were online systems, internet and abundance of data and computing power.
1. The Advent of Global businesses
The global supply chain is complex, and the only constant is change — whether it be owing to economic or regional instability, or domestic policy changes. For example, even with the move to on-line consumer commerce, reducing the importance of Distribution center-to-store shipments in favor of Distribution Centers to Consumers, the stability and long — term logistics and freight business is changing rapidly. As new markets open or the opportunity to create businesses grows, there are 4 business sectors that have the most failures within the first 18 months. These are : 1. Construction; 2. Retail; 3. Transportation; and 4. Manufacturing. (source — Dun and Bradstreet Global Business Failure report).
So the micro algorithms pertaining to an individual entity, such as the Altman Z-Tests and Merton Models, do not take into account the volatility of the general global business marketplace. If you are representing an organization that is doing business globally, or your customers are part of a global enterprise, this information would be an adjunct on how to consider future potential issues with a customer’s ability to pay. Dun and Bradstreet publish and calculate a Global Business Index quarterly that helps model the Global trends.
2. Liquidity and other micro-indicators
The ability of an entity to pay and meet debt obligations are based on the business sector, cashflow, borrowing rate and management direction. In the previous article I touched on the Acid Test or Standard ratio, Debt to Tangible net worth. In addition, EBITDA/Total Assets, Net Sales to Total Assets, book value of Equity to Total Liabilities, Working Capital to total assets, and retained earnings to total assets are traditional components of the Altman Z-Score, with weighting applied to these ratios depending on company structure and business segment.
The Altman Z-Score Formula
The original formula was created for publicly traded manufacturing companies.
Z-Score = 1.2(A) + 1.4(B) + 3.3(C) + 0.6(D) + 1.0(E)
Where:
A = Working Capital (Current Assets — Current Assets) / Total Assets (Measures liquidity of firm)
B= Retained Earnings / Total Assets (measures accumulated profits compared to assets)
C= Earnings Before Interest & Taxes (EBIT) / Total Assets (measures how much profit the firms assets are producing)
D= Market Value of Equity (Mkt. Cap. + Preferred Stock) / Total Liabilities (compares the company’s value versus it’s liabilities)
E= Sales / Total Assets (efficiency ratio — measures how much the company’s assets are producing in sales).
Z-Score Results:
Z-Score of < 1.81 represents a company in distress.
Z-Score between 1.81 and 2.99 represents the “caution” zone.
Z-Score of over 3.0 represents a company with a safe balance sheet.
Z1-Score — Score for Private Firms
Z1 = .72(A) + .84(B) + 3.107(C) + .42(D) + 1.0(E)
Z2 — Score For Non-Manufacturers & Emerging Markets
Z2 = 6.56(A) + 3.26(B) + 6.72(C) + 1.05(D)
These have been the standard algorithms to determine and predict the financial health of a company. There are studies that show Z-scores are fairly accurate to predict 80%-90% of bankruptcies around 18 months to a year prior to the fact. However, it does create some false positives. Wesley Grey, a PhD from Research Insights Factor investing, called out the reliance on these 5 ratios of the Z-Score and consider other more sophisticated alternatives.( https://alphaarchitect.com/2011/07/23/stop-using-altman-z-score/)
There is a model that considers 8 factors that may predict financial troubles ahead for an entity. The model is explained in the Journal of Finance in 2008 by John Campbell, Jens Hilscher and Jan Szilagyi
The Net Income over market value of total tangible assets. (NIMTA)
The Total Liabilities over market value of total tangible assets.(TLMTA)
Stock or Cash and Short-term investments over market value of total tangible assets (CASHMTA)
Excess return related to value-weighted Standard and Poor 500 index return (EXRET)
For publicly traded companies –
The Market Equity valuation of the firm over the total valuation of the S&P 500 (RSIZE)
Return on Stock standard deviation computed as a square root of the sum of squared stock returns over a rolling 3 month period, annualized. (SIGMA)
Market to Book value of the Firm (MB)
Current Share Price of the Entity.
In the table below, panel A shows observations for all firms for a month; panel B shows the the summary statistics for businesses that failed within a year.
3. Trading performance reported.
Having determined the appropriate ratios to base our score, and determined that a low risk of insolvency is indicated by appropriate weighting and consideration of the Financials, the next set of information that can be supplied from major credit rating or crowd-sourcing information is a measure of the entities promptness in making payments to debt obligations and open invoices with agreed payment terms. This is known as ‘Days Beyond Terms’ with a value of 0 indicating that the entity meets payment obligations as expected. An existing customer will also have internal payment histories to add to this determination, and trends may be indicative of the near future — if the entity starts paying later and later, that may indicate a management or cash-flow issue.
4. Other factors that are reported
During a normal lifecycle of a business, it may be open to litigation and legal judgements, and the outstanding balances on judgements may be a factor to consider. An enterprise is most vulnerable at start up, as a high number of enterprises fail within the first year of business activities. Certification of the details of the entity, such as prompt or completed regulatory filings, validation of the contact details and corporate location as a center for business activities are also addressed and can be factored into the overall score and determination of Credit Terms. Search and confirmation of a stated phone number can be made automatically. A street map or aerial view of the main trading address can be used to identify if it appears to be a legitimate place of business and shows business activity, with Visual Image analysis and evaluation. RSS streams can be collected from Business and Trade Publications and a search and sentiment analysis performed on any relevant articles mentioning the Entity or its senior management.
5. Pulling it all together
The final output would apply the weighted outputs of the chosen calculations, whether following Altman Z-Score, Merton or Campbell, Hilscher and Szilagyi or any other approach. The outcomes that have most usability to a Credit Manager would be an overall score and banding applied to indicate whether a Credit facility and limit would be extended, and an indication of the Maximum Credit and other terms. An algorithm that can determine a maximum credit limit based on the aggregation of the weighted scores from all the appropriate elements.
The previous 2 installments may be found here:
The next installment addresses some of the mechanisms involved in creating a matrix to determine credit limits and credit terms, and may be found here: https://medium.com/@geoff.leigh19/credit-risk-and-machine-learning-concepts-4-3c44b479a3d1?source=friends_link&sk=cf6fe8b0a96d01c68971f72cbc179229
