Paper Type
full
Description
Recently, the high penetration of mobile devices and internet access offers a new source of fine-grained user behavior data (aka “alternative data”) to improve the financial credit risk assessment. This paper conducts a comprehensive evaluation of the value of alternative data on microloan platforms with a large field experiment. Our machine-learning-based empirical analyses demonstrate that alternative data can significantly improve the prediction accuracy of borrowers’ default behavior and increase platform profits. Cellphone usage and mobility trace information perform the best among the multiple sources of alternative data. Moreover, we find that our proposed framework helps financial institutions extend their service to more lower-income and less-educated loan applicants from less-developed geographical areas – those historically disadvantaged population who have been largely neglected in the past. Our study demonstrates the tremendous potential of leveraging alternative data to alleviate such inequality in the financial service markets, while in the meantime achieving higher platform revenues.
Recommended Citation
Lu, Tian; Zhang, Yingjie; and Li, Beibei, "The Value of Alternative Data in Credit Risk Prediction: Evidence from a Large Field Experiment" (2019). ICIS 2019 Proceedings. 10.
https://aisel.aisnet.org/icis2019/data_science/data_science/10
The Value of Alternative Data in Credit Risk Prediction: Evidence from a Large Field Experiment
Recently, the high penetration of mobile devices and internet access offers a new source of fine-grained user behavior data (aka “alternative data”) to improve the financial credit risk assessment. This paper conducts a comprehensive evaluation of the value of alternative data on microloan platforms with a large field experiment. Our machine-learning-based empirical analyses demonstrate that alternative data can significantly improve the prediction accuracy of borrowers’ default behavior and increase platform profits. Cellphone usage and mobility trace information perform the best among the multiple sources of alternative data. Moreover, we find that our proposed framework helps financial institutions extend their service to more lower-income and less-educated loan applicants from less-developed geographical areas – those historically disadvantaged population who have been largely neglected in the past. Our study demonstrates the tremendous potential of leveraging alternative data to alleviate such inequality in the financial service markets, while in the meantime achieving higher platform revenues.