Using data on loan loss provisions, we analyze the expectations of US banks and find evidence of departure from rational expectations. In particular, we find that banks overreact to actual losses incurred in the recent past. In good times, the presence of overreaction leads to neglect of risks, resulting in a rise in credit growth. This subsequently results in higher non-performing loans and lower return on assets for banks when the risks get realized in the future. Additionally, shareholders fail to adequately recognize the risky lending behavior of such banks and earn predictably lower returns in subsequent years.

We analyze banks' expectations embedded in loan loss provisions and document evidence of overreaction for the US banking sector. Using diagnostic-expectations general equilibrium setting, we establish that overreaction in provisions accentuates the procyclicality of lending rate and credit growth. Crucially, our results suggest that for the cyclically-adjusted provisioning rules (IFRS 9/CECL) to have the desired effect on financial stability, the extent of cyclical adjustment may need to be strengthened in proportion to the degree of overreaction in expectations.

We model the credibility problems financial regulators often face in disclosing the stress test result. The regulator has motives to lie about the result. The motives come from the desire to guide agents' actions by influencing their beliefs. We show that the regulator can disclose some information credibly by making an imprecise announcement, revealing only the range (or interval) in which the result lies. In the interest of credibility, the regulator must reveal information less precisely when the stress test result is either too good or too bad. As the result moves away from both extremes, information can be revealed credibly with more precision.

Assessing Bank Default Risk in an Irrational World 

with Partha Ray and Vipul Mathur

We measure default probability of banks by improving structural models to account for the presence of bias in expectations. The usual approach to infer the underlying credit risk from observed variables and solve for implied probability of default can be erroneous when the variables are not representative of credit risk, rather distorted by bias. Adapting structural models to allow for provisions, we estimate a default risk measure, more apt for banks, and show that default predictions are well explained by it. Use of provisions allows us to analyze banks' expectations and then refine the estimation of bank default probability after controlling for the presence of bias in expectations.