Abstract
Using a randomized experiment in auto lending, we find that algorithmic underwriting outperforms the human underwriting process, resulting in 10.2% higher loan profits and 6.8% lower default rates. The human and machine underwriters show similar performance for low-risk, less complex loans. However, the performance of human underwritten loans largely declines for riskier and more complex loans, whereas the machine performance stays relatively stable across various risk dimensions and loan characteristics. The performance difference is more pronounced at underwriting thresholds with a high potential for agency conflict. These results are consistent with algorithmic underwriting mitigating agency conflicts and humans’ limited capacity for analyzing complex problems.