Open Access

Francesco D'Acunto, Pulak Ghosh, Rajiv Jain, Alberto G. Rossi

• We estimate the cost of cultural biases in high-stake economic decisions by comparing agents’ peer-to-peer lending choices with those the same agents make under the assistance of an automated robo-advisor. We first confirm substantial in-group vs. out-group and stereotypical discrimination, which are stronger for lenders who reside where historical cultural biases are higher. We then exploit our unique setting to document that cultural biases are costly: agents face 8% higher default rates on favored-group borrowers when unassisted. The returns they earn on favored groups increase by 7.3 percentage points when assisted. The high riskiness of the marginal borrowers from favorite groups largely explains the bad performance of culturally-biased choices. Because varying economic incentives do not reduce agents’ biases, inaccurate statistical discrimination—unconscious biased beliefs about borrowers’ quality—can explain our results better than taste-based discrimination.