Step 4: Audit the system

A practical framework for fair algorithmic pricing

Author

Fei Huang, UNSW Sydney

Does it work?

Run a pre-committed audit protocol on pricing outputs. All design choices are fixed before examining data (Huang and Hooker 2026).

End-to-end audit flow

End-to-end flow for the fairness audit protocol (Huang and Hooker 2026):

End-to-end fairness audit flow: Plan, Audit, Decide, Improve

Plan · Audit · Decide · Improve

Pre-audit setup

Select criterion: proxy discrimination (PD) or conditional demographic parity (CDP)
Specify legitimate factors X_\ell aligned with Step 1
Set tolerance bands (e.g. 5% price gap, 0.80 adverse impact ratio)
Design representative quote sample and power calculations

Statistical testing

Pricing algorithms are usually deterministic, so classical OLS/GLM standard errors are invalid. Huang and Hooker (2026) provides corrected variance for:

CDP: test \beta_A in P_i = \mu_0 + \beta_A \mathbf{1}\{A_i=a\} + \gamma^\top X_{\ell,i} + \varepsilon_i
PD: coefficient shift \Delta = \phi_j - \phi'_j when protected attribute enters model

Use equivalence testing (TOST) so models pass only when data affirm compliance within tolerance, not merely fail to reject disparity.

Three outcomes

Outcome	Action
Pass	Document; proceed with monitoring
Insufficient information	Collect more data or escalate to remediation
Fail	Remediate → re-test (return to Step 2 or Step 1)

Governance and monitoring

Role	Responsibility
Accountable actuary	Audit protocol sign-off
Independent reviewer	Replicate corrected regressions
Model owner	Remediation on fail/unclear

Re-audit on model change, regulatory update, or drift in CDP/PD statistics.

Checklist

Pre-specified protocol (criterion, tolerance, sample).
Corrected inference for deterministic outputs.
CDP and/or PD tests with TOST decision rule.
Three-outcome classification documented.
Remediation and re-audit path assigned.

References

Huang and Hooker (2026)

Xin and Huang (2024)

References

Huang, Fei, and Giles Hooker. 2026. “Fairness Testing for Algorithmic Pricing.” https://arxiv.org/abs/2605.11614.

Xin, Xi, and Fei Huang. 2024. “Antidiscrimination Insurance Pricing: Regulations, Fairness Criteria, and Models.” North American Actuarial Journal 28 (2): 285–319.