Exact demographic parity in regression requires identical prediction distributions across sensitive groups, which can be too restrictive and degrade predictive accuracy. The authors propose an approximate version: instead of forcing equality, they bound every pairwise Wasserstein distance between group-specific prediction distributions by a user-chosen tolerance. This keeps fairness tunable while allowing groups to differ.

The paper establishes that on the real line, the prediction problem and a constrained transport problem have the same infimum. When the transport problem attains this value and suitable Monge maps exist, a deterministic transport-form predictor realizes it. Under squared loss with two positively weighted groups, the W1, W2, and W∞ constraints yield interpretable operations: soft-thresholding, shrinkage, and clipping, respectively.

For estimation, the authors derive convergence rates for a plug-in estimator when the cost is known, and describe an observed-loss computational proxy when it is not. Experiments on synthetic data and California 2018 ACS data illustrate the tunable fairness-accuracy trade-off and distinguish calibration feasibility from held-out fairness and accuracy.