ChangeLog

PLEASE NOTE THAT THE API OF TRANSITION MATRIX IS STILL UNSTABLE AS MORE USE CASES / FEATURES ARE ADDED REGULARLY

v0.5.2 (XX-12-2024)

Refactoring: All non-core functionality moved to separate directories/sub-packages
- credit curve stuff moved to credit ratings modules
- data generators moved to generators modules
- etc.
Documentation: Major expansion (Still incomplete)
- Expanded Data Formats
- Rating Scales, CQS etc
- Listing all datasets and examples
Testing / Training: An interesting use case raised as issue #20
- Added an end-to-end example of estimating a credit rating matrix from raw data
- Includes various data preprocessing examples
Datasets:
- rating_data.csv (cleaned up credit data)
- synthetic_data10.csv Credit Rating Migrations in Long Format / Compact Form (for testing)
- deterministic generator (replicate given trajectories)
Tests:
- test_roundtrip.py testing via roundtriping methods

Training: Monthly_from_Annual.ipynb (a Jupyter notebook illustrating how to obtain interpolate transition rates on monthly intervals)
Datasets: generic_monthly.json
Feature: print_matrix function for generic matrix pretty printing
Feature: matrix_exponent function for obtaining arbitrary integral matrices from a given generator

Documentation: Cleanup of docs following separation of threshold / portfolio models
Datasets: generic_multiperiod.json
Feature: CreditCurve class for holding credit curves

Refactoring: Significant rearrangement of code (the threshold models package moved to portfolioAnalytics for more consistent structure of the code base / functionality)

Feature: converter function in transitionMatrix.utils.converters to convert long form dataframes into canonical float form
Datasets: synthetic_data9.csv (datetime in string format)
Training: new data generator in examples/generate_synthetic_data.py to generate long format with string dates
Training: Additional example (=3) in examples/empirical_transition_matrix.py to process long format with string dates
Documentation: More detailed explanation of Long Data Formats with links to Open Risk Manual
Documentation: Enabled sphinx.ext.autosectionlabel for easy internal links / removed duplicate labels

Feature: Added functionality for conditioning multi-period transition matrices
Training: Example calculation and visualization of conditional matrices
Datasets: State space description and CGS mappings for top-6 credit rating agencies

Installation: First PyPI and wheel installation options
Feature: Added Aalen-Johansen Duration Estimator
Documentation: Major overhaul of documentation, now targeting ReadTheDocs distribution
Training: Streamlining of all examples
Datasets: Synthetic Datasets in long format

Feature: Addition of portfolio models (formerly portfolio_analytics_library) for data generation and testing
Training: Added examples in jupyter notebook format

Feature: TransitionMatrix model: new methods to merge States, fix problematic probability matrices, I/O API’s
Feature: TransitionMatrixSet mode: json and csv readers, methods for set-wise manipulations
Datasets: Additional multiperiod datasets (Standard and Poors historical corporate rating transition rates)
Feature: Enhanced matrix comparison functionality
Training: Three additional example workflows
- fixing multiperiod matrices (completing State Space)
- adjusting matrices for withdrawn entries
- generating full multi-period sets from limited observations