Banks in Taiwan, like those in many countries, are required to comply with regulations under the New Basel Capital Accord (a.k.a. Basel II). As for credit risk management under Basel II, a few advanced banks plan to adopt an internal ratings-based (IRB) approach while others will follow the standardised approach. Whichever approach is applied, most banks have devoted substantial efforts to the design and implementation of credit rating models because a robust credit rating model can be used to estimate the probability of default of a borrower, a crucial risk component for calculating a bank's capital charge. In addition, a robust credit rating model can help banks differentiate between borrowers with different levels of creditworthiness. Indeed, a consistent and meaningful credit rating model can enhance a bank's lending business and decrease its non-performing loans. In this article, several issues in the design and implementation of credit rating models associated with potential solutions are addressed on the basis of our previous experience in helping banks develop and implement credit rating models.
Data Sufficiency and Accuracy
Where there exist sufficient internal quantitative and qualitative data, developing credit rating models using statistical methods results in few problems and produces reliable ratings. For many banks in Taiwan, however, data scarcity or poor data quality seems to be a major obstacle to developing a consistent and meaningful credit rating model. Since data adequacy and accuracy are essential for the development of a robust statistically-based model, it is imperative for banks to ensure the quality of their data. If a bank simply aims to develop a credit rating model without consideration to data quality, it is likely that the rating model developed would be neither consistent nor meaningful in practice.
A bank should consider all potential factors that could influence a borrower's creditworthiness and then collect all available internal historical data. Once the data is adequate, the model builders may start to construct a basic rating model. As new data gets collected year by year, the model, subject to appropriate modification and validation, can then be fit into the bank's daily credit operations.
Statistical Methods Combined with Experts' Judgments
As pointed out in Basel II, banks can employ the experts' judgment approach, statistical methods, or a combination of both, as tools for rating borrowers' creditworthiness. Since different credit analysts possess different levels and types of experience, the traditional credit analysis processes may not follow identical standards. A credit rating model based purely on experts' judgments would need frequent modification and hence reveal severe inconsistency and instability.
A statistically based model can avoid the inconsistency and instability that comes from the subjective judgments of credit analysts. In addition, the borrower's probability of default can be derived in a fairly scientific manner. However, with the lack of historical data, credit rating models purely based on statistical methods would be unreliable. Thus, banks in Taiwan are highly encouraged to consider both statistical methods and experts' judgments when developing credit rating models.
In the best practice, a model builder would possess a solid statistics background and years of credit analysis experience. If a model builder with a statistics background has little practical credit analysis experience, it is necessary that he/she take into consideration the judgment of the bank's credit analysts and account officers to ensure that the variables to be examined include indicators commonly used to rate borrowers' creditworthiness. Also, the model builder needs to confirm with credit analysts which variables should be included in the final rating model, because a purely statistical method may overlook practically important variables with low levels of statistical significance. This may result from data scarcity or poor data quality, in which case variables may have insignificant explanatory power in terms of borrowers' creditworthiness, but it does not imply that such variables should be excluded. Credit analysts are in a position to identify variables which, though statistically insignificant at first, are nonetheless likely to be significant in the long run. Hence, cooperating closely with credit analysts improves the predictive power of the purely statistical model. Furthermore, by so doing, the model builder can lessen the resistance to using the model from account officers, who tend to argue that the credit rating model must conform to their practices and experience.
Under Basel II, banks that intend to use the IRB approach must keep at least five years of historical data for estimating the probability of default. If this is not achievable, during the transition period, banks are allowed to have a minimum of two years of historical data and collect the future subsequent three-year set of data year by year. Integration of the three-year data into the historical data yields the five-year data required to develop rating models. In addition, before banks officially adopt the IRB approach, they must have used rating models in their daily credit operations for more than three years. In other words, estimates of the probability of default should play an essential role in the credit approval, risk management, internal capital allocations, and corporate governance functions of banks.
If a bank intends to adopt the IRB approach under the transition-period rule, it will need to modify its model once new data is collected. Since the distribution of the new data set may be quite different from that of the previous data set, a newly built model could take account of different variables from the one established previously. This happens commonly in the early development stage of internal rating models. In this respect, banks must continue to collect new data, and modify and validate their models until they become consistent and meaningful. Since ratings have a great influence on daily credit operations, banks should contemplate effective solutions to prevent the friction that model modification may induce.
During the transition period, the authority will allow banks intending to adopt the IRB approach to initially apply it to part of their assets while applying the standardised approach to the remainder. However, in the second quarter of 2006, no more than four domestic banks applied to adopt the IRB approach. This may have resulted from the following two reasons: First, for banks that might adopt the IRB approach, data may have be insufficient for estimating reliably the relevant risk components, in particular the probability of default. Secondly, for banks with poor quality assets, the capital charge under the IRB approach may be much higher than under the standardised approach.
Based on the best practice from advanced banks worldwide, banks in Taiwan should be able to build consistent and meaningful rating models if the necessary data can be adequately collected and properly maintained. In addition, if banks in Taiwan could use such models in their daily credit operations, the quality of their loan portfolio would improve greatly. Therefore, adopting the IRB approach would be superior to adopting the standardised approach for those banks that can build consistent and meaningful credit rating models.