vangap-meliora1 0.0.1rc12.post1

Credit risk development and validation tools

About the package

meliora is a Python package that provides a set of statistical tests and tools to assess the performance of the credit risk models. All tests are covered with unit tests and algorithms have been replicated in other tools like R, MATLAB and SAS to avoid errors. Whenever possible, the definition of the test was retrieved from the authoritive source like the EBA, the ECB or the Basel Committee.

The main contributors started building their first statistical credit models back in 2003. Over the years, we have impemented similar set of tests in several different financial institutions.

This package is standing on the shoulders of giants as it makes heavy use of the Python ecosystem and especially Scikit-learn, Scipy and Statsmodels. Several functions are straightforward wrappers using these resources and are provided to the user for convenience purposes. The authors have taken great care to ensure that no part of this package contains proprietary code.

Main aim

The aim of the package is to provide all common tests used by today's modellers when developing, maintaining and validating their PD, LGD, EAD and prepayment models. The aim of this package is to provide credit risk practioners with the tools to develop their credit risk models without reinventing the wheel.

Main Features

• tests cover both IFRS 9 and IRB models as well as non-regulatory models
• the tool contains more than 30 tests
• all test have been covered with unit tests
• the tests have been documented in detail
• commonly accepted tresholds have been provided for convenience purposes

For the list of all tests, see Overview > List of tests

Tests that are currently included in the package

#	Name	Area	Estimate
1	Binomial test	Calibration	PD
2	Chi-Square test (Hoshmer-Lemeshow test)	Calibration	PD
3	Normal test	Calibration	PD
4	Traffic lights approach	Calibration	PD
5	Spiegehalter test	Calibration	PD
6	Redelmeier test	Calibration	PD
7	Herfhindahl index / Concentration of rating grades	Concentration	PD
8	Brier score	Discrimination	PD
9	Receiver Operating Characteristic	Discrimination	PD
10	Accuracy Ratio	Discrimination	PD
11	Kendall’s τ	Discrimination	PD
12	Somers’ D	Discrimination	PD
13	The Pietra Index	Discrimination	PD
14	Conditional Information Entropy Ratio	Discrimination	PD
15	Kullback-Leibler distance	Discrimination	PD
16	Information value	Discrimination	PD
17	Bayesian error rate	Discrimination	PD
18	Cumulative LGD accuracy ratio	Discrimination	LGD
19	Loss Capture Ratio	Discrimination	LGD
20	Kolmogorov-Smirnov test	Discrimination	PD
21	Spearman’s rank correlation	Discrimination	LGD
22	Jeffrey's test	Discrimination	PD
23	ELBE back-test using a t-test	Discrimination	LGD
24	Migration matrices test	Discrimination	PD
25	Loss Shortfall	Predictive power	LGD
26	Mean Absolute Deviation	Predictive power	LGD
27	Bucket test	Predictive power	LGD
28	Transition matrix test	Predictive power	LGD
29	Population Stability Index	Stability	PD
30	Stability of transition matrices	Stability	PD

Full list of dependencies

• NumPy (https://www.numpy.org)
• Pandas (https://pandas.pydata.org/)
• Statsmodels (https://www.statsmodels.org/)
• Scikit-learn (https://scikit-learn.org/)
• Scipy (https://scipy.org/)

Getting Help

For usage questions, send an email to [e-mail]

License

MIT License