Online lending has grown explosively in the past decade but remains a small market compared to other financial assets. As an emerging asset class, whether online loans offer sufficiently attractive returns to warrant inclusion in an asset allocation decision remains an open question. We introduce General Characteristics-based Portfolio Policies (GCPP) — a novel portfolio optimization framework that incorporates a neural network structure and overcomes the unique challenges of building a portfolio of loans. We show that loan portfolios based on GCPP yield competitive rates of return with limited comovement, compared with investments in stocks, bonds, and real estate. Moreover, by solving an inverse problem of the portfolio optimization problem, the GCPP framework becomes an interest rate determination tool. It can effectively narrow the gap in interest rates between borrowers of different credit scores by 38.7% and remove the existing bias of interest rates against African American borrowers while maintaining lenders’ profitability. The results demonstrate that online loans are an attractive novel asset class, and platforms have the potential to improve social welfare and promote fair lending practices via interest rates.