---
title: Rotating Landlord Oligopoly Dynamics
url: https://www.emergentmind.com/topics/rotating-landlord-oligopoly
type: topic
---

# Rotating Landlord Oligopoly Dynamics

Searching arXiv for the specified papers and closely related work on algorithmic pricing and landlord concentration.
“Rotating landlord oligopoly” is best understood as an **Editor’s term** for a rental-market structure in which a concentrated set of landlords, often operating through common or similarly specified pricing algorithms, attains supra-competitive rent outcomes through coordinated adjustment that may be simultaneous or, in principle, sequenced across firms. In the current arXiv record, the concept is a synthesis rather than a directly estimated object. One paper provides tract-level evidence that higher corporate landlord concentration is associated with higher rent growth, with a larger association in majority-minority neighborhoods; another shows that simple misspecified explore-then-exploit pricing systems can converge to supra-competitive prices without explicit coordination, punishment strategies, or communication. Taken together, these results make algorithmically mediated landlord oligopoly empirically and theoretically salient, while leaving “rotation” in the sense of alternating price leadership unmeasured [2606.27525] [2605.16064].

## 1. Conceptual scope and relation to algorithmic coordination

In this usage, landlord oligopoly refers to a market in which a small number of large landlords control a materially significant share of rental units in overlapping metropolitan submarkets, so that strategic interdependence matters for rent setting. The “rotating” qualifier does not denote an observed fact in the tract-level evidence. Rather, it marks a narrower hypothesis about dynamic conduct: landlords might, in principle, alternate in initiating rent increases, stagger adjustments to avoid detection, or otherwise engage in turn-taking within a repeated-game environment. The available empirical evidence does not identify such sequencing. The tract-level study explicitly states that it has “no direct measurement of dynamic price leadership, staggering, or ‘turn-taking’ among landlords,” and the pricing-theory paper reports that it does “not see rotating or cyclical price leadership,” but rather “rapid convergence to high-price fixed points whenever exploration is clustered on one side of Nash” [2606.27525] [2605.16064].

The distinction matters analytically. A synchronized oligopoly and a rotating oligopoly can both generate elevated rents, but they imply different observables. In repeated oligopoly with differentiated apartments, algorithms acting as a hub can facilitate tacit collusion by aligning expectations; rotation could then arise as a detection-avoidance strategy or as a response to occupancy constraints. By contrast, the misspecified explore-then-exploit model shows that supra-competitive prices can emerge even when no firm conditions on explicit rival identities or alternates leadership. This suggests that “rotation” is not a necessary mechanism for oligopolistic rental harm, even if it remains theoretically plausible in richer dynamic environments [2606.27525] [2605.16064].

## 2. Institutional setting: RealPage, REIT concentration, and tract-level exposure

The institutional backdrop is the 2024 Department of Justice civil antitrust complaint against RealPage, which alleges that RealPage’s revenue-management platform, identified as AIRM/YieldStar, facilitates coordination among competing landlords and enables joint-profit maximization rather than competitive price setting across hundreds of thousands of units. The complaint explicitly names five major residential REITs: AvalonBay Communities (AVB), Equity Residential (EQR), Essex Property Trust (ESS), Mid-America Apartment Communities (MAA), and UDR, Inc. (UDR). In the ten-metro sample used in the tract-level study, these five REITs collectively own 348,121 units across 980 properties, representing 18.6 percent of the total housing stock across 665 tracts. The paper characterizes this arrangement as a hub-and-spoke structure in which RealPage acts as the information/pricing hub and large landlords act as spokes [2606.27525].

Corporate Landlord Concentration (CLC) is constructed from SEC EDGAR Schedule III (10-K) filings for AVB, EQR, ESS, MAA, and UDR for FY2016–2022. Property addresses are geocoded via the Google Maps Geocoding API and crosswalked to 2020 census tract GEOIDs using Census geocoder. The analysis covers 665 tracts in ten metros: Atlanta, Charlotte, Dallas–Fort Worth, Los Angeles, New York, San Diego, San Francisco, San Jose, Seattle, and Washington DC. The 2019 pre-period treatment is defined as
$$
\text{CLC}_{it}=\frac{\sum_{j\in i}\text{REIT units}_{jt}}{\text{Renter-occupied units}_{it}}.
$$
Because CLC is highly right-skewed, it is winsorized at the 99th percentile, and the primary specification uses
$$
\log(1+\text{CLC}).
$$

Rent outcomes are measured using the Zillow Observed Rent Index (ZORI), monthly, smoothed and seasonally adjusted, all homes plus multifamily, at ZIP level from January 2014 to December 2023. Rent growth for ZIP $z$ is defined as
$$
\text{RentGrowth}_{z}=\frac{\overline{\text{ZORI}_{z,2023}}-\overline{\text{ZORI}_{z,2019}}}{\overline{\text{ZORI}_{z,2019}}},
$$
with the pre-period equal to the average of 12 months in 2019 and the post-period equal to the average of 12 months in 2023. Using the HUD–USPS ZIP-to-tract crosswalk for Q4 2019, tract-level rent growth is the weighted average of overlapping ZIPs. The ZORI-matched regression sample includes 583 tracts, or 84.6% of the full set.

To address the possibility that corporate landlords preferentially locate in neighborhoods already experiencing appreciation pressure, the paper introduces the Algorithmic Housing Burden Index (AHBI), a composite of pre-existing rent burden and housing market tightness derived from ACS data. Pre-existing rent burden is the share of renter households spending at least 30% of income on rent; housing market tightness is the complement of the vacancy rate. Each component is standardized to z-scores relative to the 2022 cross-tract distribution and equally weighted. PCA was examined, and CLC loads orthogonally, mostly on PC3 with about 15% variance, supporting AHBI as an independent control for baseline stress rather than REIT presence [2606.27525].

## 3. Empirical evidence on rent growth and racial heterogeneity

The tract-level empirical strategy uses 2019 tract characteristics to predict rent growth from 2019 to 2023. The first hypothesis pools tracts across metros without metro fixed effects:
$$
\text{RentGrowth}_{i}=\alpha+\beta_1\log(1+\text{CLC}_{i,2019})+\beta_2\text{AHBI}_{i,2019}+\mathbf{X}_i'\boldsymbol{\gamma}+\varepsilon_i,
$$
where $\mathbf{X}_i$ includes median household income, renter share, percent Black, percent Hispanic, and total housing units. In this specification, the coefficient on $\log(1+\text{CLC})$ is $\beta_1=+0.0281$. The paper reports $p=0.086$ with metro-clustered standard errors and $p=0.030$ with HC1 robust standard errors. Its economic interpretation is that a proportional increase in REIT concentration is associated with about 2.8 percentage points higher rent growth over 2019–2023, after controlling for AHBI and tract demographics. Renter share is strongly negative at $-0.255$ with $p<0.001$, and AHBI is positive and marginally significant at $+0.079$ with $p=0.046$ under HC1. A diagnostic specification with metro fixed effects flips $\beta_1$ to $-0.032$ with $p<0.001$, indicating that most identification in this pooled model comes from between-metro differences [2606.27525].

The second hypothesis estimates within-metro heterogeneity by interacting CLC with a majority-minority indicator and adding metro fixed effects:
$$
\begin{aligned}
\text{RentGrowth}_{i}=\;&\alpha+\beta_1\log(1+\text{CLC}_{i,2019})+\beta_2\text{Minority}_{i}\\
&+\beta_3\big[\log(1+\text{CLC}_{i,2019})\times \text{Minority}_{i}\big]\\
&+\beta_4\text{AHBI}_{i,2019}+\mathbf{X}_i'\boldsymbol{\gamma}+\delta_m+\varepsilon_i,
\end{aligned}
$$
where $\text{Minority}_{i}=1$ if the combined Black plus Hispanic share exceeds 50%. The interaction coefficient is $\beta_3=+0.0587$ with standard error $0.0285$ and $p=0.039$, implying that within the same metro the CLC effect is 5.9 percentage points larger in majority-minority tracts. The baseline effect in white tracts is $\beta_1=-0.034$ with $p<0.001$, while the total CLC effect in minority tracts is $\beta_1+\beta_3=+0.024$. A continuous minority-share interaction yields $\beta_3=+0.148$ with $p=0.002$, and threshold sensitivity at 40%, 50%, and 60% remains similar in direction, with lower precision at 60% because of smaller sample sizes. AHBI remains positive and highly significant in this within-metro model at $+0.175$ with $p<0.001$.

The machine-learning complement is an XGBoost regression using eight features aligned with the OLS controls: $\log(1+\text{CLC})$, AHBI, percent Black, percent Hispanic, percent White NH, renter share, median household income, and total housing units. With an 80%/20% train/test split, 5-fold CV, and squared-error objective, the selected hyperparameters are learning rate $0.05$, max depth $4$, subsample $0.6$, colsample $0.8$, min child weight $1$, L2 regularization $\lambda=1$, and estimators $=94$, with CV RMSE $=0.0913$. The model achieves train $R^2=0.745$ and test $R^2=0.439$, so it explains about 44% of out-of-sample variance in tract rent growth from 2019 features alone. SHAP analysis ranks renter share highest, followed by median income, percent White NH, and then log CLC, which is fifth of eight with mean absolute SHAP approximately $0.015$. Directionally, mean SHAP for log CLC is positive in majority-minority tracts at $+0.0057$ and negative in majority-white tracts at $-0.0011$, corroborating the interaction result non-parametrically. The paper therefore frames its contribution as the first tract-level evidence consistent with corporate landlord concentration being associated with disproportionately higher rent growth in communities of color [2606.27525].

## 4. Misspecified explore-then-exploit pricing as an oligopoly mechanism

The pricing-theory paper studies a simple pipeline in which firms first randomize prices during an exploration phase and then estimate demand from their own historical data and set prices myopically in exploitation. In the theoretical model, there are $N$ symmetric firms selling homogeneous products under linear demand:
$$
Q_{i,t}=a-bP_{i,t}+\frac{c}{N-1}\sum_{j\neq i}P_{j,t}+\varepsilon_{i,t},
$$
with $a,b,c>0$ and $b>c$. The correct best response is
$$
BR(\bar p_{-i})=\frac{a+c\bar p_{-i}}{2b},
$$
the competitive Nash price is
$$
p^{NE}=\frac{a}{2b-c},
$$
and the monopoly price is
$$
p^{M}=\frac{a}{2(b-c)},
$$
possibly capped at $P_{\max}$. The misspecification enters because each firm estimates a monopoly-style demand function $q(p)=\alpha+\beta p$ from its own history, omitting competitors’ prices. OLS estimates are then plugged into a myopic pricing rule:
$$
P_{i,t+1}=\arg\max_{p\in[P_{\min},P_{\max}]}p(\hat\alpha_{i,t}+\hat\beta_{i,t}p).
$$
If $\hat\beta_{i,t}<0$, the unconstrained maximizer is $-\hat\alpha_{i,t}/(2\hat\beta_{i,t})$; if $\hat\beta_{i,t}\ge 0$, the firm sets $P_{\max}$ [2605.16064].

The central mechanism is omitted-variable bias induced by clustered exploration. During exploitation, firms treat variation due to competitors’ prices as noise. If exploration is concentrated in similar price ranges on the same side of Nash, subsequent price updates move with a common sign relative to trailing means, creating positive cross-firm price covariance. That covariance enters the pricing formula in a way that raises prices above the true best response. The fluid-limit analysis tracks running price means $U(t)$ and accumulated centered second moments $V(t)$:
$$
\dot U(t)=\frac{P(t)-U(t)}{t},\qquad
\dot V(t)=(P(t)-U(t))(P(t)-U(t))^\top.
$$
When average cross-firm covariance $\bar V_{i,-i}$ is zero, the algebraic OLS price collapses to the true best response. When $\bar V_{i,-i}>0$, omitted-variable bias pushes the implied price strictly above $BR(\bar U_{-i})$, and the bias grows with $\bar V_{i,-i}$ [2605.16064].

The paper formalizes “similar price ranges on the same side of Nash” through best-response cones. For exploration mean $\mu$, the upper cone is
$$
C^+=\{\mu:\mu_i>BR(\bar\mu_{-i})\ \forall i\},
$$
and the lower cone is
$$
C^-=\{\mu:\mu_i<BR(\bar\mu_{-i})\ \forall i\}.
$$
If $\mu\in C^+$, then for every scaled horizon $\alpha\ge 1$, ODE prices are componentwise strictly supra-competitive. If $\mu\in C^-$, then there exists a finite $\alpha_0$ such that for all $\alpha>\alpha_0$, prices are again componentwise strictly supra-competitive. Under symmetric exploration $\mu=s\cdot\mathbf{1}$ with vanishing exploration noise, the limit is especially sharp: if $s<p^{NE}$ or $s>\bar p^M$, prices converge to the capped monopoly $\bar p^M$; if $s\in[p^{NE},\bar p^M]$, prices lock in at $s$, which is often still above Nash.

The conduct interpretation makes the result legible in standard industrial-organization terms. At symmetric histories with common pairwise price correlation $\rho$, the misspecified OLS price coincides with the symmetric price in a game where each firm internalizes rivals’ profits with conduct parameter $\rho$:
$$
p_\rho=\min\left\{P_{\max},\frac{a}{2b-c(1+\rho)}\right\}.
$$
Thus $\rho=0$ yields Nash and $\rho=1$ yields monopoly. Positive historical correlation behaves like tacit conduct even though the mechanism is purely algorithmic misspecification rather than explicit collusion [2605.16064].

## 5. Rotation, cycles, and what existing evidence does not show

The most important negative result in the literature is that neither paper establishes rotating price leadership. The tract-level study is cross-sectional at the census-tract level and uses a two-period rent-growth outcome, so it lacks landlord-level, time-stamped price-change sequences. The pricing-theory paper, despite analyzing high-frequency dynamics, finds no cycles or alternating leaders; the ODE produces monotone movement relative to trailing means and convergence to fixed points. On $C^-$, prices exceed trailing means, $U(t)$ increases, cross-covariances remain positive, and prices converge above Nash. On $C^+$, prices remain below trailing means but stay above Nash, $U(t)$ decreases, and cross-covariances also remain positive. The analysis rules out convergence to exactly Nash except on knife-edge boundaries and does not generate oscillations [2606.27525] [2605.16064].

This leaves “rotation” as a hypothesis requiring substantially richer data. The tract-level study specifies what would be required: property- or unit-level panel data on asking rents and signed lease terms, time-stamped at high frequency; landlord identifiers; RealPage or other RMS usage flags and adoption timing; competitor sets within micro-markets such as submarkets, ZIPs, or neighborhood rings; and occupancy flows. It also identifies appropriate empirical strategies, including leader–follower tests such as Granger causality of price changes across landlords, event studies around algorithm updates or adoption, dynamic discrete-choice models of price-adjustment hazards, and structural estimation of repeated games with algorithmic signals. One could also examine whether recommendations generate synchronized waves or deliberately staggered adjustments to avoid detection. In that sense, the existing literature is compatible with algorithmically coordinated oligopoly but agnostic about whether coordination is simultaneous or rotating [2606.27525].

A plausible implication is that “rotating landlord oligopoly” should presently be treated as a broader conceptual umbrella rather than a validated empirical subtype. The evidence is strongest for elevated, coordinated price levels and for distributionally uneven exposure to those levels; it is not yet strongest for any particular sequencing protocol among landlords [2606.27525] [2605.16064].

## 6. Distributional significance, antitrust relevance, and open research problems

The tract-level findings give the concept a distributional dimension. Within the same metro, high-CLC majority-minority tracts are associated with 5.9 percentage points higher rent growth than comparable white tracts, with $p=0.039$, and the continuous interaction yields $p=0.002$. The paper therefore links algorithmic landlord concentration not only to aggregate rent effects but also to racial disparities in rent growth. It argues that this can inform Fair Housing Act disparate-impact assessments when facially neutral tools such as algorithms produce racially unequal outcomes. Because metro fixed effects ensure comparisons within the same regional housing market and macroeconomic context, the within-metro heterogeneity is especially important for distributional interpretation [2606.27525].

The antitrust implications operate on two levels. First, the RealPage setting raises a hub-and-spoke concern: cross-landlord occupancy and market data are ingested by a shared revenue-management platform, and recommended rent schedules may align rivals’ incentives toward higher joint profits. Second, even absent a shared platform, the pricing-theory paper shows that simple estimate-then-optimize systems can produce supra-competitive outcomes if they omit competitors’ prices and explore within common price bands. The policy responses proposed across the two papers therefore include restrictions on cross-landlord data sharing within pricing algorithms, mandated disclosure of RMS usage in SEC filings and regulatory reports, auditable logs of algorithmic recommendations versus actual prices, market deconcentration strategies in metros with high oligopolistic exposure, requirements that algorithmic pricing systems incorporate competitor prices or demonstrably neutralize omitted-variable bias, guardrails on clustered exploration, and certification that estimate-then-optimize pipelines do not systematically produce supra-competitive outcomes [2606.27525] [2605.16064].

The limitations are equally clear. The tract-level study is cross-sectional, its pooled CLC result is only marginally significant under clustered standard errors, and it does not use an instrument or panel identification strategy. CLC depends on ACS renter denominators subject to lag; the winsorization and log transform are modeling choices; the ten metros were selected to match REIT footprints; 12.3% of tracts are missing from the ZORI-matched regression sample; and ZIP-to-tract conversion induces spatial dependence, reflected in positive residual autocorrelation. The pricing-theory model, while sharp, abstracts in its core theory to symmetric firms, linear demand, and zero marginal cost, and then addresses empirical plausibility through a multifamily rental calibration in Boston Core with $N=801$ rentals and $H=920$ representative households. In that calibration, monopoly rents relative to baseline observed rents satisfy $p^M/p_0\approx 1.236$–$1.263$ at the 25th–75th percentiles, and supra-competitive terminal rents emerge quickly across wide parameter ranges, but price paths still converge rather than rotate [2606.27525] [2605.16064].

The present state of knowledge therefore supports a precise conclusion. Algorithmic landlord concentration is associated with higher rent growth at the tract level, that association is stronger in majority-minority neighborhoods, and simple misspecified pricing algorithms can theoretically and computationally converge to supra-competitive rents without explicit collusion. What remains unresolved is whether actual landlord oligopolies employ rotation in the strict sense of alternating price leadership. Existing results support the existence of algorithmically mediated oligopolistic rent elevation; they do not yet identify the temporal choreography by which that elevation is produced [2606.27525] [2605.16064].

Source: https://www.emergentmind.com/topics/rotating-landlord-oligopoly