Working paper · SSRN2026

Who Bears the Burden? Heterogeneous Racial Approval Differentials in U.S. Mortgage Lending: Causal Forest DML on 42 Million HMDA Applications

Rajveer Singh Pall

Not whether an average penalty exists, but who bears it, and through which underwriting channel.

SSRN preprint ↗Code ↗

The discovery in one figure

A HEADLINE REPORTS ONE AVERAGE−9.39pp, the averagebut it spreads0 · no penalty90.7% penalisedmanual −14.79automated −6.17shape schematic · values from the paperThe underwriting channel, not the applicant, explains who lands where (CATE SD 8.47 pp).

Swipe to see the whole figure

The average hides the distribution: the underwriting channel decides who bears the penalty.

The paper in five minutes

Knowing the average approval gap is not enough, because averages hide who actually pays. Using methods from modern causal inference, the same family behind clinical-trial analysis, this paper estimates the approval penalty for each applicant profile across 42 million mortgage applications. The distribution is wide: nine in ten Black applicants face some estimated penalty, and the decisive factor is not the applicant but the process. Applications handled by human underwriters carry more than double the penalty of those decided by automated systems, which points the fairness question at something a regulator can act on.

The research question

Average disparity estimates hide distribution: which applicant profiles carry the largest conditional racial approval penalty, and is the mechanism applicant- or lender-controlled?

How it works

Partially linear double machine learning with LightGBM nuisances and 5-fold cross-fitting; causal-forest CATE estimation with SHAP attribution; placebo, Oster, and Cinelli-Hazlett sensitivity analyses; within lender-year comparisons.

  1. 01
    Engineer at scale42M HMDA applications → 2M / 1.5M estimation samples
  2. 02
    Isolate the differentialdouble machine learning, LightGBM nuisances, 5-fold cross-fitting
  3. 03
    Map who bears itcausal forest estimates the penalty per applicant profile
  4. 04
    Find the mechanismmanual vs automated underwriting · same lender, same year
  5. 05
    Attack the resultplacebo shuffles · Oster bounds · Cinelli-Hazlett sensitivity

Experimental results

Pooled conditional differential −9.39 pp with wide heterogeneity (CATE SD 8.47 pp); 90.7% of Black applicants face a negative estimated effect. The channel is decisive: manual underwriting −14.79 pp vs automated −6.17 pp, and 7.13 pp persists within the same lender and year.

Conditional approval penalty by underwriting channel
Manual underwriting
−14.79 pp
Pooled (all channels)
−9.39 pp
Same lender, same year
−7.13 pp
Automated systems
−6.17 pp

Bar length = size of the estimated Black-White differential (all negative). The channel, not the applicant, is decisive.

  • −9.39 ppconditional differential (pooled DML)
  • −14.79 vs −6.17manual vs automated underwriting (pp)
  • 90.7%of Black applicants with a negative effect

Stated honestlyThe level is treated as an upper bound (no credit scores in HMDA); the channel contrast is the robust object.

Figures from the paper

Double-machine-learning estimates of the conditional approval differential across specifications.
Fig. 1Double-machine-learning estimates of the conditional approval differential across specifications.
The distribution of individual-level estimated effects: wide heterogeneity, overwhelmingly negative.
Fig. 2The distribution of individual-level estimated effects: wide heterogeneity, overwhelmingly negative.

Figures as generated by the paper’s own analysis pipeline.

What this changes

It points the fairness question at an actionable mechanism, lender-controlled routing and handling, rather than at applicant characteristics.

Resources

SSRN preprint ↗Code ↗
  • Causal Inference
  • Fairness

Citation

Working paper. The public version is linked above. Reach me at rajveerpall04@gmail.com.

Where this leads next