Housing Affordability & Commute Trade-Off Analysis · RQ4

Did COVID reprice the commute?

Where repricing happened, it favored the periphery — long-commute, low-job-access ZIP-code areas gained rent relative to job-rich cores — and it did not reverse in the return-to-office era. Explore the per-metro evidence and the honesty rails that discipline it: pre-trend checks, coarse-cluster bootstrap inference, and identification flags.

Repository Full findings Working paper (PDF)

The gradients on the ground

Every ZCTA in the metro, colored by the pre-COVID gradient measure the regressions interact with the break. Hatched areas never clear Zillow's listing-volume threshold — they are outside the covered rental submarket and outside every estimate on this page.

View as table

Two-phase repricing of the gradient

Interaction of each pre-COVID (2019-vintage) gradient measure with the disruption phase (Post1, 2020-03…2021-12) and the return-to-office phase (Post2, 2022-01+), on log rent with ZCTA and month fixed effects. Whiskers are ±1.96 clustered SE.

View as table

Event study — was it a break, or a trend?

Gradient × event-time bins relative to 2020-03 (shaded bin = reference, 2019-03…2020-02). Flat pre-break coefficients mean the post-break movement is not the continuation of a pre-existing path. The strip beneath each panel counts identifying ZCTAs per bin — thin early bins mean wide early uncertainty.

View as table

Read this honestly

Covered submarket only. Zillow's rent index publishes only where listing volume clears a threshold (70–96% of ZCTAs per metro, far fewer pre-2020); nothing here speaks to uncovered areas.
Relative, not causal. Every ZCTA is treated by COVID and there is no control group — estimates are within-metro relative descriptions.
“Repricing” = the listing index moved — an amalgam of price and listing-composition change; no hedonic adjustment is possible at this altitude.
Joint coefficients are conditional. The three gradients are correlated; where joint and single-interaction models disagree, collinearity is doing the talking. Toggle both above.
Memphis is under-identified (12 ZCTAs on both sides of the break) and reads as uninformative, not as a zero.

Estimates recomputed from the repository's committed data products (scripts/export_rq4_explorer.py). Methods, robustness, and per-metro reports: findings §10 · built by Charles Coonce.