Citi Research published an adaptive non-linear regression model to estimate default probabilities for U.S. banks back in September of last year. Following up on their bank failure prediction model, Citi Research issued a research report Friday of last week discussing the performance of the model, and elaborating on how different analytical criteria offer greater predictive strength over differing time horizons.
Multiple factor bank failure model
The model Citi has developed looks at a variety of factors to assign a probability of default (PD) to each institution. The factors considered in the model are all information that can be derived from an institution’s financial statement. In calculating the PD, the model examines each banks’ fraction of non-performing loans, the ratios of assets to liabilities, return on equity, the ratio of net loans to bank equity, yields on earning assets, asset size and net interest margin.
Citi describes the predictive performance of its model versus the models used by ratings agencies in the overview of the report. “We find that the predictive power of agency ratings drops off dramatically as credit quality of the scoring sample increases, with much less deterioration in default prediction using Citi’s bank model.”
There's a gold rush coming as electric vehicle manufacturers fight for market share, proclaimed David Einhorn at this year's 2021 Sohn Investment Conference. Check out our coverage of the 2021 Sohn Investment Conference here. Q1 2021 hedge fund letters, conferences and more SORRY! This content is exclusively for paying members. SIGN UP HERE If you Read More
Bank failure: Predictive strength at a one- to two-year horizon
When you crunch the numbers, it turns out that the factors of banks’ fraction of non-performing loans, the ratios of banks’ assets to their liabilities and banks’ return on equity proved to be the most predictive over a one- to two-year horizon.
Bank failure: Predictive strength at a three- to four-year horizon
Citi’s analysis showed that the predictive factors discussed above were not as significant over longer periods of time. It turns out the ratio of net loans to bank equity and yields on earning assets were the most predictive factors over the three- to four-year horizon.
Bank failure: Predictive strength at a five-year horizon
When you extend the model out to five years, the ratio of net loans to bank equity and yields on earning assets lose their predictive strength. At five years, the yield on earning assets, asset size and net interest margin are the most predictive factors.
Bank failure: Citi’s PD model superior overall to agency models
The authors of the report highlight the overall superiority of Citi’s PD model to the models used by ratings agencies.
“Although default predictions from agency ratings fall off rapidly for banks rated at or above single-B and single-C, Citi’s bank model predictions deteriorate far less.