17. Migration matrices

Transition matrices for modelling the rating migration process should be based on historical migration data using either external sources (i.e. rating agencies) or internal sources. Matrices from external sources should be preferred in cases where internal historical data is sparse. Institutions should ensure that the amount of historical data is sufficient to derive robust, accurate and statistically consistent estimates. Institutions should validate the robustness of transition matrices particularly in relation to higher rating categories, where a few severe downgrades or defaults can affect the migration frequency significantly. In accordance with the requirement for a „standard of soundness comparable to IRB‟ which the CRD establishes, a minimum historical observation period of 5 years is required.

Separate transition matrices may be applied for specific groups of issuers and specific geographical areas. Depending on (i) the composition of the institutions‟ portfolio, (ii) the availability of accurate transition matrices and (iii) possible differences in migration characteristics across products, issuers and/or geographical areas, a balanced decision should be made on the set of transition matrices used. Such a decision should consider (i) the choice of the (internal or external) source in combination with an analysis of the overlap and/or possible mismatch between the institutions‟ portfolio and the assets underlying the transition matrix; (ii) the motivation for any weighting scheme (also for the use of equal weights); and (iii) the size of the historical window. Institutions should develop one (or more when relevant data is available) transition matrix that is specific to sovereign obligors.

When default is modelled as an absorbing state, transition matrices should be adjusted to ensure that this absorbing state does not conflict with internal PD estimates. Similarly, transition matrices where „NR‟ („Not rated‟) or another column is an absorbing state for withdrawn ratings or non-rated exposures can be adjusted. Any such adjustments should be documented, and the impact of the specified adjustment should be included as part of the documentation.

Transition matrices, in general, relate to a one-year horizon. Where shorter horizon matrices are required, which is the case when liquidity horizons shorter than one year are used, the corresponding transition matrices cannot always be computed directly and approximations are required. Both the approximations and the motivation for specific assumptions applied in this process should be documented (e.g. when generator matrices are used). These assumptions should also be assessed in order to verify that they remain valid over time. As part of the validation process a matrix based on a generator could be calculated for a horizon identical to the originating matrix‟s horizon in order to assess the difference resulting from the process of developing the generator matrix.