Nonlinear Loan Loss Provisioning

Abstract

The extant banking literature often models loan loss provisions as a linear function of changes in loan portfolio quality. Large sample data indicate that this linearity assumption is invalid and that a V-shaped piecewise linear specification fits much better. Decreases in nonperforming loans are associated with increases in loan loss provisions. This anomalous asymmetric relation is partly driven by the mechanical accounting effects of loan charge-offs on nonperforming loans and allowance for loan losses. We find that, controlling for concurrent loan charge-offs, loan loss provisions move in the same direction as nonperforming loan change, but asymmetry remains. The effect of nonperforming loan increases on loan loss provisions is still twice as large as that of nonperforming loan decreases. We argue that the residual asymmetry is caused by conditional conservatism. We show that loan loss provision asymmetry is greater for banks with more high-risk construction loans and shorter-maturity loans and for public banks, and is more pronounced during economic downturns and in the fourth quarter, consistent with the predictable effects of conditional conservatism.