Discussing: Nonpayment and Eviction in the Rental Housing Market

Authors: Humphries, Nelson, Nguyen, van Dijk & Waldinger

Florian Oswald
florian.oswald@unito.it

Uni Turin (ESOMAS), Collegio Carlo Alberto

April 29, 2026

What the Paper Does

  • Data: 5,809 low-income tenancies; full monthly payment histories + eviction filings (1,814 used in estimation)
  • Fact 1: 50% of tenants default at some point; landlords collect only 91% of rent due
  • Fact 2: Landlords wait 2–3 months before filing — option value from learning
  • Fact 3: 39% of first-time defaulters eventually fully repay
  • Model: Bayesian landlord, Markov tenant types, \(C_e \approx \$2{,}000\)
  • Policy: tax / delay / SRA all cut evictions 5%; differ sharply in cost and targeting

The Model in One Analogy

Like Rust (1987), but the bus changes type each period

  • Rust: Harold Zurcher decides each month whether to replace a bus engine
    • Replace → reset; keep → risk of breakdown governed by engine mileage
  • Here: landlord decides whether to evict a tenant
    • Evict → vacancy, new tenant draw; keep → rent income governed by tenant type \(\omega_t\)
  • Key difference: in Rust, the bus type is fixed — only mileage accumulates
    • Here, \(\omega_t \sim F(\cdot \mid \omega_{t-1})\): the tenant’s type itself drifts each period
    • Today’s reliable payer may not be tomorrow’s

Comment 1: What is a “Type”, \(\omega_t\)?

  • True type: innate payment ethics, personal pain threshold, financial discipline
  • Transitory shock: job loss, illness, family crisis (Li, Meghir, and Oswald 2025)

Policy implication flips depending on which predominates

  • Shock-driven: SRA timed to the shock is nearly first-best
  • Type-driven: SRA is largely wasted on persistent non-payers

A third channel: peer effects - Campaniello and Macaluso (2026): +10pp neighbor delinquency → +3pp own delinquency in Turin public housing

Comment 2: The \(\epsilon\) Shock

Eq. (8) looks like \(\sigma_\epsilon \equiv 1\) — but Table 4 estimates it

\[P(e=1 \mid \pi, b) = \frac{1}{1 + \exp(\bar{v}^{e=0} - \bar{v}^{e=1})}\]

  • Rust normalizes \(\sigma_\epsilon = 1\): no dollar scale in flow utility
  • Here \(R \cdot y_t\) is in observed dollars → \(\sigma_\epsilon\) is separately identified
  • Table 4: \(\hat\sigma_\epsilon = \$581\), i.e. 29% of \(\hat{C}_e\)
  • Identified from steepness of \(P(e=1)\) across \((\pi, b)\) states

Risk: \(\Delta\bar{v}(\pi,b)\) is computed, not observed → misspecification loads onto \(\sigma_\epsilon\)

Two Discussion Points

  1. Tax beats Delay — the mechanism challenges the case for right-to-counsel
  2. The 5% reduction in evictions result — entirely driven by estimated type persistence; how credible?

D1: Tax Beats Delay

  • Tax: raises up-front cost → deters marginal evictions → saves recoverable tenants
  • Delay: extends post-filing stay → raises threshold for filing on Type L → defers, not prevents, those evictions
Type H saved Type M saved Type L saved Net cost
Tax 30% 19% 10% $0.91/mo
Delay 6% 9% 12% $8.20/mo

Q: Delay is modeled only as \(\delta_e\) (slower exit). Right-to-counsel also reduces possession judgments and enables repayment negotiations — does the model understate its benefits?

Same issue for SRA: eligible only before filing; landlord can evict after receipt → forbearance effect, not protection effect

D2: 250$ Tax reduces evictions by 5%

Driven by high persistence: \(L \to L = 96.4\%\)

  • High \(C_e\): landlords wait → accumulate information → few marginal evictions
  • High persistence: once Type L, almost certainly stays Type L
  • Table 6: less persistent types → same tax reduces evictions 6× more

How is persistence identified?

  • Payment histories of non-evicted tenants only (evicted tenants are censored)
  • Conditional independence: eviction uncorrelated with future payments given \(h_t\)
  • Full-information model needs \(C_e = \$3{,}550\) to fit the same data

Summary

  • First paper to link payment histories to eviction decisions — opens a new empirical agenda
  • Tax beats delay: a clear, policy-relevant ranking of instruments with a precise mechanism
  • Open question: how much of the 5% reduction in evictions is a feature of this market vs. the model’s type persistence structure?
  • Natural next step: 15% of evictions are preventable — can we identify and target those tenants ex ante?

References

Campaniello, Nadia, and Mariele Macaluso. 2026. “Peer Effects in Tenant Delinquency: Evidence from a Quasi-Natural Experiment in Public Housing.” https://sites.google.com/view/marielemacaluso/research.
Li, Wenli, Costas Meghir, and Florian Oswald. 2025. “Consumer Bankruptcy, Mortgage Default, and Labor Supply.” International Economic Review 66 (3): 1019–42.
Rust, John. 1987. “Optimal Replacement of GMC Bus Engines: An Empirical Model of Harold Zurcher.” Econometrica: Journal of the Econometric Society, 999–1033.