Credit Risk and Machine Learning Concepts -2
In the first blog, I mentioned 3 approaches and a number of scenarios that concern those in the world of Credit Risk Management. There is a whole more complex set of considerations that a Bank or Financial Services organization may need to consider in terms of identifying what may be appropriate to determine risk in setting up a loan to a commercial entity or individual, and this is not what I am addressing in this series of blogs.
What I need to talk about next are some facts and figures. If a counter-party has a poor rating for credit, more than 15% will default over a 10-year time horizon.
(Fitch Ratings Global Corporate Finance 2009 Transition and Default Study)
So, what do commercial credit scores model, and what is the science behind it?
The Credit Rating agencies provide relative credit rating information of Business Entities, as well as some forecast of Systemic Risk by trying to calculate a probability of default, usually expressed as a Credit Score.
Timing is not a major consideration with the Credit Score being updated with new information so a trend of scores is seldom factored into the rating agency approach. They are inherently slow to downgrade, to avoid procyclical effects and erosion of reputation. The risks are measured and calculated at multiple time horizons, rather than a single time span.
Through these perceived issues, Credit Monitoring and Rating Agencies are getting less confidence by a Business and therefore in-house Credit Management capabilities enhance this risk assessment.
(Association for Financial Professionals Annual Conference 2010 Developing Predictive Analytics for Supply Chain Credit Risk presentation)
There are 2 mainstream and rule-based models that have been developed since the 1960’s that address credit rating calculations. They mainly address two types of information, Financial Factors and Qualitative factors. These are the Altman Z-test and the Merton Models.
A stability rating is also typically calculated from this base data, and the typical weighting is as an illustration as above.
I created and maintained a system in the UK prior to online company information and internet features being common. The UK requires all registered companies to file an annual return report with Companies House, including a snapshot of core financials. This includes all private corporations as well as publicly traded companies. At the time this was mandated, the information was made available by imaging the paper returns onto Microfiche. Microfiche were like photography negatives on plastic, and may have one or many pages depending on how the information was filed that included financial account information. An analyst had to mount this plastic squares into a reader that would illuminate the images of the paper returns (typically in tiny- less than 1 cm square panels — on the 5 inch by 4 inch plastic sheet, and magnify the image so that they could read through the content. The system I contributed to allowed them to enter data from the Microfiche into a computer system so that the Company that was to be rated as requested by one of our customers had the Financial Factors only captured, plus the timeliness of the reporting of the financials (some companies spent longer than a year after the financial year closed to file a report so either it was late or very stale information!).
The computer program was able to calculate and output to a printer for faxing to the requester, and later we added a Mainframe Dial-up solution and specified the Teletext Graphics 2 for IBM and compatible Personal Computer access through Dial-up modem to reduce the faxing and shorten the delay, especially if Company 2 requested a credit score for company A that had just recently been entered and scored from a request from Company 1.
The calculations were very similar to those published by Altman in 1968 as the Z-score formula.
Firstly, I would like to revisit why a commercial customer may not be able to pay an invoice. There are 5 main reasons:
1. Product was defective, so need to return.
2. Invoice was incorrect or not at the agreed price.
3. Receiving company did not receive an invoice in a timely manner to allocate funds to.
4. The company has a cash flow issue and is temporarily unable to meet all debt obligations.
5. The company has a fundamental liquidity and stability problem and may never be able to meet its debt obligations or refuses to pay.
An organization cannot predict the first 2 issues, an organization may have to address its internal processes to improve problems around issue 2 and 3, but the issues of 4 and 5 are what credit rating agencies and Z-Score, Merton Model or more sophisticated approaches based on these models address.
So, what are the financial fundamentals, and what may we calculate to understand a customer organization’s ability to pay invoices within the agreed credit terms and to what aggregated limit of credit should they be permitted to reach?
The first thing is to understand the Cash Flow, and also the total shareholders funds or net assets.
The point of the Cash flow factor to open Debt shows the health of a company, and the net assets indicate the long-term stability, in simple terms. A company may be very profitable, but may have factors such as its capital structure, Financial Ratios and Liquidity that would caution extending particularly favorable terms.
The key ratios and calculations that are common in Financial performance analysis, mainly for publicly traded company bond rating and investment rating, include:
· Tangible Net Worth (Net worth less any intangibles, such as the value of Property, Cash in hand, Equipment, Stock and vehicles owned by the company less any debts and discounting any leased property, premises or equipment). Common practice is to treat total tangible asset values of less than $1 Million as a negative.
· An important and consistently referenced metric is Net Debt to Earnings before Tax, Interest, Depreciation and Amortization (EBITDA). Common practice is to treat this value if above 3.5 as a negative factor
· Acid Test Ratio, or quick ratio, compares short term assets to short term liabilities, an indicator if the company has enough cash flow to meet current obligations. A value less than 1 is usually cause for caution. There are 2 ways this may be calculated. The most common formula takes the sum of the values of Cash, cash equivalents, marketable securities and current accounts receivables and divides this by total current liabilities. The second method takes the total current assets less inventory value and divides by total current liabilities from the balance sheet financial statements. This is more accurate than an alternative calculating the current ratio, as this includes the valuation on inventory but inventory cannot necessarily be rapidly converted to cash and not always at the valuation. The Acid Test ratio, if over time show a decrease in value, would indicate that the organization may be weakening in top line growth.
· Another common ratio is Total Debt to Tangible Net Worth, with a value of less than 2 being a cause for concern.
The fact that all registered companies in England are required to file the base information that would allow such calculations makes this scoring easier. In the US, there is no such annual filing requirement and the Companies being formed and registered in each of the individual 50 states plus Washington DC and dependent territories means that there is not just one source for this information. With Publicly traded companies, they are required to provide quarterly and annual financial statements by the SEC in a standard format, and these are centrally available and online.
There are a number of non-registered business entities in every country, commonly operating as sole traders, and the only information that can be leveraged is the individual’s person credit score.
I primarily rated and applied weightings to these ratios based on observation and tests of failed companies, and with year over year comparisons (as financial reports usually have the previous year’s data as well as the latest year’s data.) Another indication is years in Business, with a sweet spot of 5 to 25 years usually indicating the continuity and stability of the business.
In the next installment, I will discuss the Hybrid approach, utilizing some rule-based as well as the opportunity for Assisted Intelligence and Machine Learning to be applied.
Previous article may be read here : https://medium.com/@geoff.leigh19/credit-risk-and-machine-learning-concepts-85ef47c978c7?source=friends_link&sk=5249acc679330bd64c76bcae1dc074d1
The next installment may be read here :
