From series “Banking and Finance Research – Short and Readable” we introduce article of Tobias Berg , Philipp Koziol , An Analysis of the Consistency of Banks’ Internal Ratings
Article describes findings, that the dispersion of PDs is large: the average standard deviation of bank’s PD estimates at the borrower/date level is 1.35%, while the mean PD is 1.70%. The standard deviation of the logarithm of the PD, a measure of the relative dispersion of PD estimates, is 77%.
This implies that the PD estimates for a firm with a median PD of 1% and three bank relationships are 1%, 0.46%, and 2.16% (i.e., 1% and 1%·exp(+/-0.77)) – clearly an economically wide range.
Using multivariate regression techniques, it was found that the dispersion of PD estimates is higher for lower credit quality borrowers and for larger borrowers, that is, exactly for those borrowers where capital requirements matter most.
Basel II introduced a homogeneous default definition as 90 DPD and a homogeneous time horizon of 12 months. Furthermore, regulators need to approve ratings used for the IRB approach under Basel II – which includes extensive checks on the accuracy and calibration of rating models.
These findings directly raise the following question: Are these large differences idiosyncratic or do some banks systematically provide lower or higher PD estimates than other banks?
If differences are idiosyncratic, then overall capital requirements are not affected. However, if some banks systematically assign lower PDs than other banks, capital requirements are no longer comparable across banks. We formally address this question by regressing the abnormal PD, i.e. the difference between a bank’s PD estimate and the mean PD estimate across all banks for the same borrower in the same quarter, on bank-quarter fixed effects.
How much regulatory capital would banks need if average risk weights across banks were applied instead of risk weights based on their own PD estimates?
To remind you:
Are these (large) dispersions of bank’s PD estimates purely idiosyncratic or is there a systematic effect? Purely idiosyncratic differences do not affect capital requirements because in the aggregate they simply cancel each other out. Idiosyncratic differences might even be welcomed by regulators, as they suggest that banks develop individual opinions about their borrowers instead of herding around a single estimate. On the other hand, systematic differences suggest that capital requirements are no longer comparable across banks.
Result is primarily driven by the cross-section of bank and should therefore be interpreted with care. In a last set of tests, it was found that bank`s reported PD estimates increase after significant capital increases. This result is consistent with the narrative that bank’s incentives play an important role for their regulatory PD estimates.
- First, the variability of PD estimates for the same borrower across banks is large.
- Second, bank fixed effects explain 5% of the variation in PD estimates across banks, while 95% of the variation is idiosyncratic.
- Third, we find some evidence that bank characteristics explain the size of bank fixed effects. In particular, weaker-capitalized banks on average report lower PD estimates.
The article tries to show, that regulatory oversight did not fully eliminated systematic differences in internal PD estimates across PDs, however the results are more than surprisingly positive.
Analysis & Data Setup:
Unique supervisory data set comprising quarterly data from Q3/2008 to Q4/2012 from 40 German banks for more than 17,000 corporate borrowers. For each bank/borrower/quarter, the data set contains the loan exposure of bank X to borrower Y in quarter t along with the internal PD estimate that is used for regulatory purposes. The internal PD only reflects credit risk and is reported for a uniform time horizon of 12 months on a numerical scale from 0.00% to 100.00%, thereby allowing a direct comparison across banks. In particular, we are able to clearly distinguish between differences in internal ratings that arise from differences in loan portfolios and those that arise from differences in internal rating models.
Based on: Tobias Berg , Philipp Koziol , An Analysis of the Consistency of Banks’ Internal Ratings, Journal of Banking and Finance (2017), doi: 10.1016/j.jbankfin.2017.01.013, online