8

One excellent resource is to try Kaggle and to examine some of the competitions, some of which are specifically on the application of machine learning to credit scoring. https://www.kaggle.com/c/GiveMeSomeCredit You wil see that the winning solution is made public, including source code and output. https://github.com/IdoZehori/Credit-Score/blob/master/...


7

S&P credit rating change information until 2012 (European Union only): http://www.standardandpoors.com/ratings/articles/en/us/?articleType=HTML&assetID=1245327302187


5

You cannot do it. It is an under-determined problem. That is to say, a whole multitude (subspace of $\mathbb{R}^{N\times N}$) of migration matrices will agree with any given table of default probabilities. Say you want to find a transition matrix for 2 states (IG, HY) plus default $$\left(\begin{matrix} p_{11} & p_{12} & p_{1D} \\ p_{21} &...


5

One option to do it is a heatmap. Not sure which software are you using, but in matlab it is extremely simple to do and powerful to tweak. Below an example. Let's assume there are 30 periods $t$ to $t+30$ and 21 ratings. Then you could run: rating = {'Aaa'; 'Aa1';'Aa2';'Aa3';'A1';'A2';'A3';'Baa1';'Baa2';'Baa3';'Ba1';'Ba2';'Ba3';'B1';'B2';'B3';'Caa1';'...


4

Firstly it's good to straighten out our goal. You correctly say, that IFRS9 requires analysis of expected losses. There are two components of expected losses. 1) Expected probability of a default event 2) Expected recovery rate So, not only do we need the probability but also the recovery rate. Luckily, both are approximated by the credit spread, which ...


4

U.S. Government DID save American International Group (AIG) from bankruptcy, since it was considered too big to fail, actually: a lot of financial institutions were insured by AIG. This Investopedia page is a nice summary on the topic about AIG's bailout. Here (Investopedia again) about Lehman Brothers, that became really too much leveraged and exposed to ...


4

This is an interesting question. I'll make a guess on what may be the driving factors for "ratings inflation" based on these assumptions: Rating agencies compete among themselves to conduct bond rating business with issuers, since they are paid for their services by the issuer. Bond issuers choose the agency that promises the highest rating, since the ...


4

(P) prefix : As a service to the market and typically at the request of an issuer, Moody's will assign a provisional rating when it is highly likely that the rating will become final after all documents are received, or an obligation is issued into the market. A provisional rating is denoted by placing a (P) in front of the rating. Such ratings may also be ...


4

Actually, there is a practical way to do it. You can use you PoD estimates to assign a credit rating to your securities and then use a published transition matrix for your purposes. Or you can estimate transition probabilities by linear interpolation based on the PoD values that you have. Here is a publication containing transition matrices from Moody'...


3

Most of the papers concern CDS spreads which you will need to convert to a PD. Paper using country specific fundamentals: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2517018 This paper uses leverage: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2361872 Another one that decomposes them against peer groups: http://papers.ssrn.com/sol3/papers....


3

Yes, you can have two different ratings. The issuer has one credit rating, but the individual issues, even if they are both senior unsecured/secured with the same maturity, coupon, etc. can have different ratings. The key factor is going to be the structure/provisions of the issue itself. For example, an issue with a sinking fund is going to be viewed as ...


3

Reuters uses a proprietary model defined StarMine structural/SmartRatios Credit Risk model that has been developed by themselves and provided with the Reuters data service. It does not exist a formal definition or paper about the model, in which it is explained how to get that score; Reuters simply explains roughly what is in its website without going into ...


3

You are right, the rules to time-scale a T-years transition matrix $M_T$ are: $M_{k·T} = M_T^k$ $M_{T/k} = \sqrt[k]{M_T}$ The root of a matrix M can be obtained using the spectral decomposition: $M = P·D·P^{-1} \Longrightarrow M^k = P·D^k·P^{-1}$ where $P$ and $D$ are the eigenvectors and eigenvalue matrices of $M_T$. Note: The Perron-Frobenius tells ...


2

Depending upon how much data you have, you might find Violi (2004) useful. Nickell et al. (2000), while principally considering time-dependent stability tests, refers a bit to significance testing between the matrices of different agencies and might also provide some insight.


2

I believe that your problem can be formulated as: Find PD matrix that is as close as possible to a given PD matrix (result of some previous calibration, or the matrix computed using average hazard rate, or any other "target", or the penalty on non-smoothness) subject to the following constraints: The values that are given must be matched exactly ...


2

Yes of course, credit rates depend on interest rates (i.e. https://en.wikipedia.org/wiki/Libor), which are set by some group of banks in almost every country Going further bankers analyze the market situation and also national interest rates, which are set by central bankers in every country which has a central bank (https://en.wikipedia.org/wiki/...


2

You can do this using the optim function in R. One possible solution is as follows: base <- c(0.9190, 0.0739, 0.0072, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0113, 0.9126, 0.0709, 0.0031, 0.0021, 0.0000, 0.0000, 0.0000, 0.0010, 0.0256, 0.9119, 0.0533, 0.0062, 0.0021, 0.0000, 0.0000, 0.0000, 0.0021, 0.0536, 0.8794, 0....


2

Regarding how the rating agencies gave AAA ratings to CDOs and the like that clearly did not deserve those ratings - straightforward answer. The SEC licences all the ratings agencies as "nationally recognized statistical rating organizations" (NRSRO). It is blindingly obvious that the SEC was not actually overseeing the rating organizations that it was ...


2

I am also not aware of any papers in this area. But having developed many such models, I can list the important steps: Decide on the target variable: usual choices are historical default data, agency ratings and expert rankings Create a sample containing the possible predictors Reduce the list with the help of some expert, e.g. exclude all the predictors ...


2

The "issuer-pay" model works like this: The Rating Agency goes to the issuer and says "We heard that you are going to issue bonds. We can give you a rating if you pay us XXX dollars. It will help you a lot to have our rating". The Issuer of course is free to refuse this offer (after all this is just a rating agency, not the Cosa Nostra). In this case the ...


2

If there is a CDS on the bond, that might be a good indicator to use, esp. if you want to compare one against another.


2

Merton model will be a bit more quantitiative. Z-Score is an option, as is Ohlson. In the end you are going to want some non-defaulted->defaulted transition mapping based on factors you identify as meaningful.


2

Yes, you can. Also, do not use Altman's Z. The extreme scores are predictive, but a load of empirical research shows the intermediate values are not predictive. The best solution is a Bayesian solution because you are gambling money. Bayesian methods are coherent. Coherence is the statistical property by which fair gambles can be placed. Frequentist ...


2

Mapping ordinal data to interval data is arbitrarily. The ranking of rating agencies is ordinal data, so only comparing operators > or < can be applied. The data can be sorted and as a central tendency, you can calculate the median. The main aspect of ordinal data is that it allows for rank order but it does not allow for the relative degree of ...


2

Assume your outcome/dependant variable is the rating agencies rating category, say 10 to 20 rating categories, you can use ordinal logistic regression which is more natural for this kinda problem. So the model will predict the rating category. If your dependant variable is the internal default flag then you can have your model predict the default rate and ...


1

Take a look at the Altman Z Score, sounds like it is what you are looking for - https://en.wikipedia.org/wiki/Altman_Z-score


1

What you're looking for looks to be more in the realm of a mathematical model (specific to the company's size, available liquidity, and industry). Credit Risk Pricing Models may provide a decent overview of how to build such a model. Unfortunately duration/convexity will only help you capture the interest rate risk on your bonds, and not any of the ...


1

Such data are provided by Bloomberg Terminal. I am positive about Moody's and S&P but not sure whether Fitch rating is also available.


1

Bloomberg has a Default Risk model, which is similar to what you are querying. You can see a screenshot in this PDF. There you can also see the kind of variables they use. You can access it by typing DRSK at the CDS screen is Bloomberg. (If the screenshot in the PDF is not clear enough, let me know and I can post one with better resolution from Bbg) This ...


1

I would think it is because it can be bound between 2 points it can assume wide range shapes It fits the data empirically (as you said) On a related note Sometime back I read a paper which might give you more formal reason. It is for estimating and simulating recovery rates . I havnt used it to model credit migration probabilities . But I think one ...


Only top voted, non community-wiki answers of a minimum length are eligible