The new accounting standard requires that financial institutions estimate expected credit losses on their loan portfolios. The predictability of long-term losses, however, remains an open question. We develop a model that predicts long-term loan losses and incorporates adjustments for macroeconomic forecasts. The model combines cross-sectional predictions with a high-dimensional dynamic factor model that tracks aggregate losses over the business cycle. The model predicts long-term losses out-of-sample with significantly greater accuracy than the Harris et al. (2018) model and several other alternatives. It is also more effective at detecting bank failures. We use the model to estimate the present value of expected losses and the expected loss overhang for a given bank-quarter. The estimated present values subsume information in reported allowances and in fair value disclosures about long-term losses; the evidence is also consistent with loss overhang distorting banks' decisions. The model provides a useful benchmark to study loan loss provisioning.

The concept of conditional conservatism (asymmetric earnings timeliness) has provided new insight into financial reporting and stimulated considerable research since Basu (1997). Patatoukas and Thomas (2011) report bias in firm-level cross-sectional asymmetry estimates that they attribute to scale effects. We do not agree with their advice that researchers should avoid conditional conservatism estimates and inferences from research based on such estimates. Our theoretical and empirical analyses suggest the explanation is a correlated omitted variables problem that can be addressed in a straightforward fashion, including fixed-effects regression. Correlation between the expected components of earnings and returns biases estimates of how earnings incorporate the information contained in returns. Further, the correlation varies with returns, biasing asymmetric timeliness estimates. When firm-specific effects are taken into account, estimates do not exhibit the bias, are statistically and economically significant, are consistent with priors, and behave as a predictable function of book-to-market, size, and leverage. Data Availability: Data are publicly available from sources identified in the article.