Expected Loan Loss Provisioning: An Empirical Model
54 Pages Posted: 1 Mar 2019 Last revised: 7 May 2020
Date Written: April 30, 2020
The new accounting standard requires that _nancial institutions provision for life-time expected losses on their loan portfolios. We develop a model for estimating long-term expected loan losses that incorporates a wide range of bank- and aggregate-level predictors of future losses. The model combines cross-sectional predictions with a high-dimensional dynamic factor model that tracks sector-wide losses over the business cycle. We show that our model predicts long-term losses out-of-sample with much greater accuracy than does the Harris et al. (2018) model, which is more effective over the short-term. As an application of the model, we construct a proxy for expected lifetime losses and measure expected loss overhang (Bushman and Williams (2015)) at the bank-quarter level. We find that the estimated lifetime losses subsume information about long-term losses contained in the reported allowances (prior to the regime change) and has an order-of-magnitude higher predictive ability. The lifetime losses are more procyclical than the reported allowances. Our evidence is also consistent with expected loss overhang distorting banks' real decisions. The model serves as a useful benchmark to evaluate the timeliness of provisioning under the new accounting rules.
Keywords: Expected loan losses, long-term losses, CECL, expected loss overhang
JEL Classification: G21, M40, M41
Suggested Citation: Suggested Citation