Risk prediction based on financial statements

I have a profit loss statement and balance sheet with the following fields: Example

P&L
Turnover420,363 -
Cost of sales             £118,730    £140,169    -
Gross Profit                £178,862    £280,194    -
Operating Costs           £154,889    £255,123    -
Interest paid (received)  £3,007  £4  -
Tax paid  -               £4,838  -
Profit After Tax (Loss)     £20,966 £20,229 -

Balance Sheet
Total assets                £68,090 £47,032
Current assets            £62,975 £43,847
Fixed assets              £5,115  £3,185
Total liabilities           £56,560 £55,731
Current liabilities       £56,560 £55,731
Long-term liabilities     -   -
Shareholder Funds / Net Assets £11,530  £(8,699)


I'd be interested in what models exist to predict credit risk or bankruptcy likelihood based on this information. (I have other meta data available as well, such as industry and location, but I ignore that for the moment)

After researching for a while I found this paper by Altman http://onlinelibrary.wiley.com/doi/10.1111/j.1540-6261.1968.tb00843.x/full He uses a method called Z-score that is the linear combination of five financial ratios. wikipedia Is this still widely used today? If not what are the alternatives?

• I do not see how this relates to quantitative finance, given financial statement analysis is pretty much the antidote to financial mathematics. With all the accounting gimmicks (some corporations hold more off-balance sheet assets and liabilities than on-balance) its a moot point to derive meaningful conclusions regarding risk and expected return by looking at financial statements only. Sadly, in today's time the CFO's most valued skill set is in making accountants sign off on massaged balance sheets and income statements. What more to say? – Matt Jun 4 '13 at 2:56
• Certainly your conclusion is more than a little simplistic and its origins blindly dismissive. Last I checked, AQR have managed to put together a nice business based primarily on applying relatively simply quant methods to fundamental data and I don't see many writing them off. It may not be 'haute finance', but frankly, and particularly in the face of a lot of more recent attempts to apply AI in finance that I've seen fall squarely on their face, I'm most often inclined to keep it simple unless there's a reason not to. – Chris Jul 16 '19 at 23:04
• @Chris You are missing his point. Matt meant that this kind of analysis is relatively hard to model as there are a lot of human influences. In quantitative finance we assume that the human influences is relative small and hence "random". There is no clear right or wrong which approach to pursue. But concerning the audience in quant.stackexchange, this kind of modelling might be off-topic. – quallenjäger Jul 17 '19 at 9:25
• @quallenjäger, not missing the point, and I even generally agree with his position, but his presentation was a little 'baby with the bathwater', as several of the responses below indicate. – Chris Jul 17 '19 at 16:30
• @siammii do you have access to data, other than the company above? – Dave Harris Jul 19 '19 at 15:21

You can start exploring the subject by having a look at Credit risk measurement: Developments over the last 20 years if only for the reference list. As a more modern approach an "upgraded" version of the original Z-score method was recently proposed by Altman: Z-Metrics™ Methodology For Estimating Company Credit.

Though be aware that despite numerous alternative methods being available (just search for discriminant analysis), most of the time you could describe the process as fitting a simple model to the available training/test data set. So the question you have to answer before all is about the appropriateness of the assumptions to your situation (analyzed universe, regime changes etc.). Quality of data is also an issue as already mentioned by @MattWolf.

• Z-metric might be what I'm looking for because my data concerns small private companies that are not listed on the stock market. – siamii Jun 5 '13 at 14:15

If you don’t have any market quotes, one possible way to assess the credit risk of an obligor is to use its financial statements. For instance, this paper describes the criteria that S&P use to derive the credit rating of a given obligor:

1. Indicative credit ratings might be calculated using the following ratios: FFO/Debt, Debt/EBITDA and Debt/Capital.
2. You also need to consider the business risk profile of the obligor (country & industry risk, competitive position, etc.)

Given its financial and business risk, a theoretical rating can be obtained using the matrix provided by S&P. Once you have the credit rating, to infer its credit risk you may:

• Compute the credit spread of bonds issued by companies with the same credit rating.
• Compute the historical default rates that have been observed for companies with the same rating (see S&P).

You maybe want to have a look at this paper

Are Ratings the Worst Form of Credit Assessment Apart from All the Others?

• Can you provide a summary of this paper? – chrisaycock Jun 5 '13 at 11:14