FEDINCOME: FEDERATED LEARNING FOR INCOME ESTIMATION IN DIGITAL LENDING UNDER DATA SOVEREIGNTY CONSTRAINTS

Battle Count: 0

The paper focuses on credit risk and income estimation for lending decisions, which is distinct from quantitative trading strategies, market making, or algorithmic execution. While it involves financial data and risk management, it is not directly applicable to trading alpha generation.

Implementing a custom federated learning simulation with multiple aggregation algorithms (FedNova, SCAFFOLD, etc.) and a specific Residual MLP architecture requires significant engineering effort. Integrating downstream DTI threshold calibration adds further complexity.

The paper provides detailed implementation specifics (PyTorch 2.0, custom simulation framework, fixed random seeds, specific hyperparameters for all federated algorithms). It uses public LendingClub data. However, the codebase is not explicitly linked in the text provided, though it is described as organized into modules.