Screening Soft Information: Evidence From Loan Officers

I evaluate how loan officers screen uncodified, soft information using data from China. After documenting substantial differences in loan decisions and outcomes across loan officers, I develop and estimate a screening model incorporating screening ability and beliefs regarding ability. Estimates imply that the typical loan officer is risk-averse, has heterogeneous screening ability, and behaves overconfidently-behaving as if he or she observes more from soft information than what the data would indicate. However, I still find that loan officers offer value over benchmarks that ignore soft information. Counterfactuals on compensation, loan assignment, and training further explore the limits of screening.