Beyond FICO: Enhancing Mortgage Default Forecasting and Inclusive Lending via Multimodal Graph Neural Networks and Urban Mobility Analytics

Abstract

Mortgage underwriting in the United States still relies heavily on FICO-style bureau scores, debt-to-income ratios, and loan-to-value cutoffs, even though default emerges from a broader system of household liquidity, neighborhood shocks, housing-market conditions, and lender relationships. This paper develops a framework for improving mortgage default forecasting and inclusive lending through multimodal graph neural networks (GNNs) enriched with urban mobility analytics. Drawing on evidence from mortgage finance, credit scoring, housing economics, network science, and mobility research, the study argues that FICO-centered pipelines understate both relational risk and place-based resilience. Borrowers with similar scores can face different default hazards when commuting stability, job accessibility, local price volatility, and origination-channel exposures differ. The paper synthesizes relevant sources and proposes an empirical design combining Freddie Mac loan-level performance data, HMDA disclosures, FHFA house-price indices, American Community Survey neighborhood variables, and Census LEHD commuting-flow measures. A heterogeneous graph architecture connects borrowers, loans, properties, tracts, lenders, and employment-access nodes, while multimodal encoders fuse tabular, spatial, temporal, and network signals. The framework is evaluated against logistic regression, gradient boosting, and non-graph deep learning baselines using discrimination, calibration, fairness, and prudent-inclusion metrics. The core contribution is a governance-aware blueprint for safer and fairer underwriting. By treating bureau scores as one modality rather than the dominant lens, lenders can identify hidden resilience among thin-file applicants and hidden fragility among superficially strong files. The paper concludes that multimodal GNN systems can outperform FICO-dominant underwriting while supporting more inclusive, auditable mortgage credit allocation in the United States.