When the Benchmark Moves with the Policy

The question

When sellers prefer offers that are more likely to close, a cash buyer can win a home that a financed buyer values more. If a city wants to widen access for owner-occupiers, how would it tell which policy does that at the lowest cost?

The paper builds a simulated housing market and compares four instruments that change who wins: a cap on purchases per entity, a fair-share quota, a constant charge on institutional offers, and a duty that rises with local competition for a home.

Main findings

• 2.2% vs no crossingin the main proxy market, the moving efficiency ratio crosses zero near a 2.2% offer discount; the fixed comparison remains negative at every tested discount up to 20%
• 0.75–2.5×cohort access gains relative to full-run gains in the main-market target experiments; changing the population can also reorder the instruments
• 2rankings of the same instruments: calibrated to the same full-run access target, a constant charge keeps more value than the duty; calibrated to the same cohort target, the duty keeps slightly more
• 9.8% → 9.5%owner-occupiers who leave without buying at New York City proxy values, and with the cash discount removed: a comparison within the illustrative model

Why it matters

• Policies aimed at institutional or cash buyers are usually judged by an access number and an efficiency number. How each is measured can decide which policy looks best.
• An efficiency benchmark computed over the buyers still waiting moves with the policy itself. A two-period example in the paper shows why.
• Following an entry cohort to purchase or exit can change the measured effects and the ranking of instruments.

Run the replication check

The three-year LivSim replay is complete for both policies in all three paired replications. All 264 numerical comparisons match the archived results exactly. A separate calculation reproduces the cohort statistics. In the full-period policy-A/replication-1 control, all 24 original outcome CSVs are identical with and without audit logging.

Download the verified replication supplement (ZIP, 9.97 MB). Extract it, open a terminal in that folder and run:

python3 verify_release.py

Python 3.9 or later; no extra packages or network connection. Expected output ends with "status": "PASS". The offline check verifies the saved outputs; rerunning the model takes several hours. Read the verification note for provenance and limits.

These AI-assisted checks use the original authors’ synthetic example inputs. They do not validate actual patient outcomes or add a new policy effect. PolicySpace2’s source instrumentation was checked; its full ten-year model was not rerun. The replay adds no new estimate.

Earlier benchmark note and audit files

Auditing policy dependent benchmarks in allocation simulations is the earlier note incorporated into this paper. Its archive includes the PolicySpace2 instrumentation patch, analysis script and per-seed results. The PDF covers the housing and service-queue examples. For the external applications and current replication checks, use the main paper and the verification section above.

Cite this paper

@misc{loschi2026wedge,
  author       = {Loschi, Pablo},
  title        = {When the Benchmark Moves with the Policy: Measuring
                  Housing-Access Instruments in Allocation Simulations},
  year         = {2026},
  howpublished = {Working paper and replication package, Zenodo},
  version      = {7.2},
  doi          = {10.5281/zenodo.23760672}
}