Effort-Centric Fairness in Lending Decisions

Battle Count: 50

The paper is primarily focused on credit scoring fairness and lending decisions rather than quantitative trading. However, it has tangential relevance to quantitative finance through its treatment of portfolio credit risk measures (Expected Loss, Unexpected Loss, RAROC), the Basel ASRF framework, and the general methodology of embedding fairness constraints into predictive model training. The risk bounds and financial trade-off analysis could inform credit portfolio management decisions relevant to financial institutions engaged in quantitative risk assessment.

The core logistic regression implementation with exact effort expressions is relatively straightforward. However, the full framework requires: (1) causal structure learning via PC-stable algorithm and BIC-scored greedy search, (2) soft rejection indicator implementation, (3) mixed-integer programming for discrete feature handling, (4) greedy algorithm for pathway generation, (5) risk bound computation, and (6) careful hyperparameter tuning (lambda, kappa, weight matrix W). The bi-level optimisation is collapsed to single-level via local surrogates, reducing computational burden. SVM robustness tests add additional complexity. Overall, moderate-to-high implementation effort for the complete pipeline.

The paper provides detailed algorithmic descriptions (Algorithm 1 and 2), explicit formulas for all key quantities, and specifies hyperparameters (learning rate 10^-2, batch size 128, kappa=10, threshold tau=0.5). However, no code repository is mentioned. The data sources (HMDA and Freddie Mac) are publicly available but the matching procedure requires specific implementation. Five random runs are reported with means and standard errors. The causal structure learning procedure (PC-stable + BIC greedy search) is described but implementation details for the specific software used are not fully specified.