THE REDISTRIBUTION OF HOUSING WEALTH CAUSED BY RENT CONTROL
KENNETH R. AHERN † AND MARCO GIACOLETTI ‡
Abstract This paper studies the effects of rent control on the housing wealth of renters, landlords, and homeowners. Over the nine months following the passage of rent control in St. Paul, Minnesota in 2021, average property values fell by 4.4% to 5.8%. Leveraging administrative parcel-level data, we show that upper-income renters gained more than lower-income renters. In contrast, small landlords lost the same as large landlords. Owner-occupants’ wealth also fell significantly, due to both direct capitalization effects and negative externalities. These results provide the first evidence on the heterogeneous wealth effects of a wave of new rent control laws. (JEL D61, D62, G51, H23, R23, R31, R38)
September 2023
Keywords : Rent control, real estate valuation, wealth transfers, externalities
We thank Tom Chang, Katie Galioto, Andra Ghent, Richard Green, Arpit Gupta, Adam Guren, Isaac Hacamo, Erica Xuewei Jiang, Matthijs Korevaar, Lu Liu, Song Ma, Tim McQuade, Stijn Van Nieuwerburgh, Emily Nix, Chris Palmer, Chris Parsons, Kate Pennington, Jacob Sagi, Selale Tuzel, and participants at USC, Yale University, 2022 Innovations in Housing Affordability Summit, 2022 Real Estate Research Symposium at UNC, 2022 Northern Finance Association Annual Meeting, 2022 ESCP-TAU-UCLA Conference on Housing Affordability, 2023 American Real Estate and Urban Economics Association Meetings, 2023 AEA Annual Meeting, and the 2023 WFA Annual Meeting. † University of Southern California, Marshall School of Business and NBER, 701 Exposition Blvd., Ste. 231, Los Angeles, CA 90089-1422. E-mail: kenneth.ahern@marshall.usc.edu. ‡ University of Southern California, Marshall School of Business, 701 Exposition Blvd., Ste. 231, Los Angeles, CA 90089-1422. E-mail: mgiacole@marshall.usc.edu.
Electronic copy available at: https://ssrn.com/abstract=4061315
I. Background: Saint Paul and the Rent Control Ballot Measure
A. Historical Context of Rent Control
The so-called first generation of rent control laws were enacted by the federal government during World War II as a temporary method to stabilize rental markets during a period of mass relocation (Pastor, Carter and Abood, 2018). During the post-War housing boom, rents declined and the temporary rent control laws were not renewed, except in New York City (Arnott, 1995). The second generation of rent control laws were enacted in the 1970s in response to growing inflation and as part of a general regulatory practice of price controls. New laws were passed in Massachusetts, Washington DC, and California. These second generation laws were less restrictive than the first generation of rent control laws. They allowed landlords to pass some costs on to tenants; rents to be set to market rates upon vacancies; exemptions for new construction and small landlords; and rent increases to be tied to the rate of inflation. Following the second wave, a regulatory backlash led many states to pass laws that banned or limited rent control at the local level, including Massachusetts (1989), California (1995), and Illinois (1997). This trend continued in recent years in a wide range of states, including Colorado (2010), Mississippi (2013), Indiana (2017), Iowa (2017), and Florida (2018). By 2019, 37 states had passed laws that preempted rent control at the local level. Recently, as housing costs increase, the pendulum appears to have swung back in favor of rent control. As shown in Table 1, many states are revisiting their laws that preempt rent control or have enacted state-level rent control. Cities have also been exploring options for
enacting rent control, including Minneapolis and St. Paul, Minnesota. Though the Minnesota state legislature preempted rent control at the local level in 1984, the state statute had a provision that allowed local governments to enact rent control if approved in a general election. On November 2, 2021, Minneapolis and St. Paul residents voted on two separate rent control measures. St. Paul’s ballot measure was a vote for a specific rent control law that capped rental increases at 3% per year, with few exemptions. The law passed with a 53% to 47% split. Minneapolis’s ballot measure was an amendment to the city charter allowing for the possibility of introducing a new, unspecified, rent control law in the future. This provision was also approved with a 53% to 47% split. 2 In contrast to St. Paul’s stringent rent control, Minneapolis’s ballot measure did not create any new laws. Because no law was actually enacted, we cannot know what market participants anticipate about future provisions. Though Minneapolis and St. Paul tend to enact similar laws (e.g., minimum wages, COVID masking policies, and paid employee leave), the mayor of Minneapolis, who was re-elected in November, has been a vocal opponent of rent control. In addition, as discussed below, Minneapolis had confounding measures on the ballot when it passed its limited rent control law. For these reasons, this paper focuses on St. Paul’s rent control law.
B. St. Paul’s Rent Control Ordinance
At the time of its passage in November 2021, St. Paul’s rent control ordinance was unique in its stringency. First, unlike most rent control laws, St. Paul’s law did not include vacancy decontrol provisions. This meant that rent increases in St. Paul were originally limited to 3%, independent of inflation and regardless of whether a property became vacant and was re-rented to new tenants. Thus, there was no mechanism for rents to be adjusted to market prices and rent growth could be capped below inflation rates for an indefinite number of years. Second, unlike most rent control laws that exempt new construction to encourage increases in supply, there was no exemption for new construction in St. Paul. All residential rental properties were under the jurisdiction of the law. Similarly, there were no exemptions for small landlords or for properties with few units and no provisions for owner-occupants, as are common in other rent control laws. While St. Paul’s law did not include vacancy decontrol, it also did not include ‘just cause’ eviction controls. Eviction controls give more rights to tenants by restricting landlords from evicting tenants except for a limited number of predefined reasons, such as non-payment of rent, violation of other terms of the lease, and in some cases, the personal use of the For comprehensive information on Election Day results, see https://electionresults.sos.state.mn.us/20211102
landlord. This means that under just cause eviction controls, tenants in good standing have the right to remain in a rental unit indefinitely. Taken together, St. Paul’s absence of vacancy decontrol meant that landlords could not increase rents by evicting current tenants, but the absence of eviction controls meant that landlords could evict tenants for other reasons. This institutional detail will be important when we consider whether the price effects of rent control include a benefit to renters by preventing forced evictions.
To much controversy, on April 29, 2022, the city government issued a set of implementation procedures that substantially weakened the terms of the law as passed by the voters in November 2021. In particular, the new rules would allow landlords to increase rent in order to maintain an inflation-adjusted constant net operating income based on the property’s operating income in 2019. Any rent increase below 8% per year could be self-certified by the landlord, with the possibility of an audit. Increases between 8% and 15% would need to be approved by the city. The maximum allowable rent increase in one year would be 15%, but increases in excess of 15% could be deferred to future years.
After the end of our sample period in July 2022, the City Council made further amendments in September that took effect in January 2023. Most notably, the amended law exempts new construction for 20 years retroactively and allows for partial vacancy decontrol following a just cause vacancy. Beginning with the implementation of the law starting in May 2022, these amendments have made the law more comparable to existing laws in other jurisdictions. It is possible that real estate participants anticipated the weakening of the law before May 2022. On the other hand, it is possible that real estate prices respond slowly to new information (Cellini, Ferreira and Rothstein, 2010). To the degree that market prices impound future expectations, if investors anticipated the weakening of the law, then we can consider our estimates as a lower bound for the effects of the original terms of the law and as accurate estimates for the effects of more typical rent control laws.
Though St. Paul’s initial rent control law was especially severe, it is a good representation of the national trend towards stricter rent control laws. Older laws typically allow rent to increase at the rate of the local consumer price index (CPI). However, more recent laws limit rent increases to a fraction of CPI or a fixed percentage, as in St. Paul. For example, Los Angeles’s 1978 law limits increases to 100% of CPI. In contrast, Santa Ana’s 2021 law limits rent increase to 80% of CPI or 3%, whichever is lower and Pasadena’s 2022 law limits rent increase to 75% of CPI. Similarly, there is a trend towards providing fewer exemptions from rent control. For example, in Los Angeles, single-family residences are exempt from rent caps, as are properties built after 1995. In contrast, Portland, Maine’s 2022 law has no age exemption and only exempts small multi-unit buildings if they are owner-occupied. Given this trend toward stricter rent control laws, evidence from St. Paul can help shed light on the effect of the next generation of rent control laws.
II. Conceptual Framework of Rent Control and Property Values
Basic economic theory predicts that rent control causes both transfers of wealth and deadweight losses (DWL) for property owners. These losses can be divided into a direct capitalization loss and an indirect negative externality loss. The sum of these effects is observable as a decline in the market value of real estate, as follows:
[ \text{Value Loss}t = E_t \sum\tau \frac{I_{\text{Rental}, t+\tau} \times \text{Rent Saving}{t+\tau} + \text{DWL}{t+\tau}}{(1 + r_{t+\tau})^\tau} + \text{Expected Negative Externality}_t ]
where Value Loss_t is the property value loss realized after the passage of the law at time t, I_Rental is an indicator for rental properties at time t, and r_t is the discount rate. The direct effect of rent control on existing property values includes two different components. The first component of the direct effect is a transfer of housing wealth from owners to renters, which is equal to the expected rent savings caused by rents that are constrained to be lower than free-market rents. The second component of the direct effect is a deadweight loss caused by a reduction in the level of housing quality, relative to the free-market level. In particular, landlords have an incentive to reduce maintenance expenses and let their properties deteriorate if rents are kept artificially low by rent control. Gyourko and Linneman (1990) show that rent control led owners to reduce maintenance expenditures, though Olsen (1988) argues that tenants of rent controlled units are likely to endogenously increase maintenance in response. Both components of the direct effect represent a housing wealth loss to owners. However, the transfer component represents a housing wealth gain to renters. The direct effect only occurs if a property is rented (I_Rental = 1). If the property is owner-occupied, the owner enjoys the full value of the property, even under rent control, and there is no loss. Therefore, the expected direct effect of rent control on the present value of the property is moderated by the probability that the property is rented now or in the future. As we show below, there is a positive transition probability from owner-occupied to rental housing which means that in expectation the direct capitalization effect also impacts properties that are currently owner-occupied. Owners may endogenously exit from the rental market in response to rent control by selling rental properties to owner-occupants. In our framework, this lowers the probability of being a rental, which reduces the exposure to rent control. Moreover, landlords in St. Paul have an incentive to increase current rents immediately before the passage of the law. These increases are difficult to observe if rental contracts are privately renegotiated outside of new listings. By studying forward-looking transaction prices, our results capture these effects. In contrast to the direct effect, the indirect effect of rent control on existing property values is caused by negative externalities in the city. Numerous studies report that lower valued properties cause negative spillover effects on other properties (Rossi-Hansberg, Sarte and Owens III, 2010; Autor et al., 2014). These effects could be driven by changes in such attributes as crime or school quality (Autor, Palmer and Pathak, 2019; Cellini et al., 2010). Because these externalities make the property less desirable, both for renters and owner-occupants, they represent a deadweight loss without any transfers.
III. Identification Strategy
The first step of our analysis is to identify the causal relationship between rent control and property values. Though the passage of rent control in St. Paul presents a setting that has similarities to an ideal experiment, there are important deviations.
A. Cross-Sectional Variation
First, rent control is not randomly assigned to a sample of properties. Instead, all properties within the city of St. Paul are subject to rent control. Therefore, we use properties located in the five counties surrounding St. Paul as a control sample. The advantage of this approach is that we do not need to be concerned that an omitted variable, like building age, could determine both the assignment to the treatment group and also a change in market value. Likewise, because there are no exemptions, owners cannot easily remove their properties from rent control, which would bias our treatment sample. Moreover, because the city boundaries of St. Paul are not driven by geographic boundaries that could influence property values, areas adjacent to St. Paul represent contiguous and integrated real estate markets. Rent control also creates deadweight losses by reducing the incentive to supply new housing. Though we focus on value changes of existing properties, we also provide additional evidence on changes in the supply of new housing.
The disadvantage of our setting is that we have to be concerned that the treated properties within St. Paul may not be comparable to the control properties outside of St. Paul. To address this concern, we use three different specifications of location fixed effects to capture time-invariant cross-sectional differences between treated and control groups: city, ZIP code, and Census block group. These fixed effects capture the large majority of potential cross-sectional time-invariant confounding differences in property values across city boundaries, such as school districts, tax rates, and urban density. Because the geographic boundaries are narrowly defined, the fixed effects also absorb more nuanced variation that may affect property values, such as commuting time, neighborhood feel, and architectural styles. We also control for individual property traits, including square footage, number of units, and building age, to absorb other sources of price variation unrelated to rent control. As an additional test to alleviate concerns that properties located in the control sample are not comparable to properties in St. Paul, we identify whether a property is a rental or owner-occupied. Following our conceptual model, we expect that rental properties will be more impacted by rent control than owner-occupied properties. The comparison between rental and owner-occupied properties allows us to compare the changes in property values of two properties within the same small geographic region within St. Paul, similar to prior research on rent control in Cambridge and San Francisco (Autor et al., 2014; Diamond et al., 2019a). To further address the concern that properties in St. Paul might be systematically different than those outside of St. Paul, we provide robustness tests that limit the properties in the control sample to those that are geographically close to the border of St. Paul. Control properties located near the border of St. Paul are likely to share many of the same qualities as the treated properties located inside St. Paul, such as commuting times, quality of construction, and local amenities, though they are not directly affected by rent control. A final threat to our identification is that real estate prices in St. Paul may reflect preferences for urban versus suburban locations. Though we control for geographic fixed effects which absorb time-invariant differences in demand for particular locations, if there was a coincidental increase in demand for suburban real estate at the time of the rent control vote, we could falsely attribute lower property values in St. Paul to rent control, when in fact it represents an unrelated shift in demand. Prior work demonstrates a surge in demand for suburban real estate by residents of large urban cities during the Covid pandemic (Gupta, Mittal, Peeters and Van Nieuwerburgh, 2022; Ramani and Bloom, 2022). It is possible that a similar shift in preferences and reallocation of housing demand occurred in November 2021 for St. Paul buyers.
To address this concern, we control for the location of real estate in city centers versus suburban areas using data from five metro areas comparable to the Twin Cities: St. Louis, Kansas City, Indianapolis, Nashville, and Denver. We choose these areas because they have roughly the same population size as the Twin Cities area and are geographically proximate. Internet Appendix Table 1 shows that the five comparable areas have similar demographic backgrounds, incomes, and housing markets as in St. Paul. In particular, though St. Paul has a higher fraction of white residents and a higher median income than the other cities, housing costs are roughly equal in the comparable cities as a fraction of income. In addition, St. Paul’s population growth, immigrant growth, and income growth is in the middle of the distribution across the comparable cities.
B. Time-Series Variation
While fixed effects and property traits account for cross-sectional confounding variables, we also need to control for confounding time-series variation in market prices unrelated to rent control. This includes general time trends, anticipation of the law, and deviations from an assumption of parallel trends between treated and control groups. First, to control for macroeconomic variation in the time-series, we include year-month fixed effects for each month from January 2018 to July 2022. These fixed effects absorb both seasonal variation and yearly variation for the average property in the sample. Thus, estimated changes to prices following the passage of rent control will reflect abnormal changes relative to seasonal norms and average yearly changes. Second, we test for anticipation of the passage of the law. As noted, the ordinance was passed with a relatively close vote of 53% to 47% with 58,546 total votes cast, out of about 210,000 voting-age citizens. In Internet Appendix Figure 2, we show that media coverage of rent control issues in the St. Paul area only increased significantly in October 2021. Given that escrow periods are about four to six weeks, media coverage might have had only limited influence on the transactions that occurred before the election. In addition, to our knowledge, there was no public polling of the law in advance of the vote which could have led to substantial anticipation and response to the passage of the law. 5 In addition, St. Paul and Minneapolis did not have excessive rent before the passage of rent control. According to Census Bureau estimates, the median gross rent as a percentage of household income in the Minneapolis-St. Paul metro area was 28.4% in 2019, which
5 See the discussion in the public press: https://minnesotareformer.com/briefs/heres-the-rent-control- question-st-paul-will-vote-on-this-fall/ and https://myvillager.com/2021/10/13/st-paul-debates-merits-of- rent-control-measure-on-ballot/
places it at the 47th percentile in a sample of over 900 metro and micro Census areas. In addition, using data from HousingLink, Internet Appendix Figure 3 shows that the median inflation-adjusted rent for a two-bedroom unit in St. Paul has remained roughly the same from January 2019 to November 2021, when rent control was approved. Finally, we need to provide evidence that the transaction prices in St. Paul would have followed a parallel trend with the controlled properties if rent control had not been passed. First, we note that the rent control law was the only initiative on the November 2 ballot in St. Paul, so its passage was not accompanied by the passage of any related laws. The only other elections in St. Paul in November 2021 were a landslide win for the incumbent mayor and contests for four school board seats. Similarly, we need to consider any one-time confounding events in control cities. Most notably, Minneapolis would be a natural control for St. Paul. However, in addition to the ballot measure on rent control, Minneapolis’s ballot also included referenda on mayoral power and policing. These confounding events mean that if property values in St. Paul changed relative to Minneapolis, we could not attribute the change to rent control. Therefore, for all of our tests, our control sample excludes real estate in Minneapolis. In the control cities, there were no ballot measures and only routine school board elections. To provide more evidence to support the assumption of parallel trends, we run an event study to identify if transaction prices in St. Paul followed a parallel trend with the control cities in the period before rent control was passed. To increase the credibility of the parallel trends assumption, we also estimate the doubly robust difference-in-differences estimator of Sant’Anna and Zhao (2020) to reduce biases caused by using time-varying covariates under the assumption of parallel trends conditional on the covariates.
C. Econometric Specifications
Based on the discussion in the previous sections, we estimate the following difference-in-difference equation using only data from the St. Paul area:
[ \ln(\text{price})_{ikt} = \beta \cdot \text{StPaul}_i \times \text{Post}t + \gamma X_i + \alpha_k + \tau_t + \varepsilon{ikt} ]
in which StPaul_i is a dummy variable equal to one for properties located in St. Paul and zero for properties outside of St. Paul; Post_t is a dummy variable equal to one for transactions that closed after the passage of the law; X_i is a vector of characteristics including the log of the building age, the log size of the building in square feet, the log number of units, and dummies for different property types (apartments, townhouses, single family residences); and alpha_k and tau_t are families of geographic and year-month fixed effects. The coefficient beta reflects
a percentage change in property prices within St. Paul, relative to the change in property values in the control cities. Throughout the paper, standard errors are double-clustered by year-month and by the geographic level of the fixed effects.
To control for changes in preferences for downtown versus suburban areas, we also estimate a triple-differences model as shown in the following equation:
\begin{aligned} \ln(\text{price})_{ikmt} &= \beta \cdot \text{TwinCities}_m \times \text{Downtown}_i \times \text{Post}_t \ &+ \lambda \cdot \text{TwinCities}_m \times \text{Post}_t + \delta \cdot \text{Downtown}_i \times \text{Post}t \ &+ \gamma X_i + \alpha_k + \tau_t + \varepsilon{ikmt} \end{aligned}
where TwinCities_m is a dummy variable equal to one for properties located in the Twin Cities metro area and zero for properties located in the other five metro areas; Downtown_i is a dummy variable equal to one for properties located in the downtown area of its metro, and zero for properties located in suburban areas; and Post_t is defined as before. For the Twin Cities area, downtown is defined as St. Paul. For the control cities, the city center (downtown) is the main city area as defined by Census. The triple interaction coefficient beta reflects whether the difference-in-differences effect in Saint Paul versus the surrounding area is equal to the difference-in-differences effect in the downtown of the control cities.
IV. The Effect of Rent Control on Real Estate Values in St. Paul
A. Data
We construct a comprehensive micro-dataset of real estate prices, covering both single-family houses and multi-unit properties in the five counties surrounding St. Paul and in the counties surrounding the five comparable metro areas. For sales of houses and small multi-unit properties, we download data from Redfin, which includes property types (single-family residence, townhouse, multifamily, etc.), characteristics (square footage and age), addresses, and precise geo-location (latitude and longitude). We exclude properties with missing or nonsensical geo-locations, with missing prices, with missing number of bathrooms or bedrooms, with number of bedrooms exceeding 10, and with number of bathrooms exceeding eight, or equal to zero.
Data on transactions of larger multi-unit properties are from Electronic Certificates of Real
Estate Value (eCRV) from the Minnesota Department of Revenue. We restrict the data to residential apartment buildings with four or more units. We also only include ‘clean’ transactions with complete eCRVs as defined by the Department of Revenue. We omit duplicate copies of transactions that appear in both Redfin and the eCRV data. Our final sample includes 169,119 transactions in the Twin Cities (including 16,943 in St. Paul with 2,564 in the post-period), and 805,271 total transactions in the comparable metro areas over the period from January 2018 to July 2022. Internet Appendix Tables 2 and 3 provide a complete breakdown of observations by city and county. To our knowledge, between the Redfin data and the eCRV data, our sample includes the near-universe of all residential properties sold in the Twin Cities area.
Figure 1 provides a map of the transactions in the Twin Cities sample. Transactions in St. Paul are indicated by black dots. Transactions in the suburbs are indicated by blue dots. The empty space next to St. Paul is Minneapolis. This figure shows that the large majority of the control transactions are located close to St. Paul and the city boundaries appear arbitrary. To provide a pre-rent control benchmark, Table 2 reports sample statistics for the period January 2018 to October 2021. Panel A shows that the average transaction price of a single family home in St. Paul over the pre-rent control period is $280,395 and the median is $240,400. This represents a price per square foot of $178 (average) and $170 (median). Multi-unit properties in St. Paul sell for $616,146 on average ($292,500 at the median). The average property has 5 units and sells for $134,139 per unit, while the median has two units and sells for $122,450 per unit. Nearly 7% of the transactions in St. Paul are rental properties, with an average rent of $1,620 per month, and $1,375 at the median. In comparison, in the suburbs of St. Paul, transaction prices of single-family properties are higher though the price per square foot is lower and the properties are larger. Multi-unit properties in the suburbs of St. Paul have more units and transact at higher prices, on average. The properties in the suburbs also have considerably newer construction. Panel C provides summary statistics for single-family and small multi-family residences in the five comparable metro areas. On average, the transaction prices, sizes, and ages of transactions in the comparable areas are nearly identical to prices of single-family properties in the suburbs of St. Paul.
B. Estimates of the effect of rent control on transaction values
Table 3 presents estimates of Equation 2 using data on all transactions, including large multi-unit properties, from the Twin Cities area, and controlling for different levels of location fixed effects. Across the three specifications, the results show that rent control caused a statistically significant decline in transaction prices over the entire nine-month post period. The estimate of the average decline varies across the three types of geographic fixed effects from −4.4% to −5.8%. Next, we control for migration from downtown areas into suburban areas. We first estimate placebo tests in Panel A of Table 4 in which the sample only includes the five comparable metro areas. Across three out of four specifications of location fixed effects, we find positive and significant changes in property values for downtown vs. suburban areas following the passage of rent control in St. Paul. In the fourth specification using city-level fixed effects, we find no statistical significance. Panel B of Table 4 estimates the triple-difference effect in Equation 3 using observations from St. Paul and the five comparable areas. The estimated effect is statistically and economically significant, ranging from −5.2% to −8.1%. These results imply that the decline in property values in St. Paul following rent control does not reflect a general trend common to the downtown areas of other Midwestern cities of the similar sizes. According to the Ramsey County Assessor’s Office, there are 73,103 private residential parcels in St. Paul, with an aggregated estimated market value of $24.2 billion. Using the most conservative estimate of a value loss of 4.4%, our estimates imply that rent control caused an aggregate loss of $1.06 billion dollars to property owners in St. Paul over the nine months since its passage. Using the upper-range of 8.1% from the triple-difference tests, the aggregate loss is $1.96 billion dollars.
C. Robustness Tests
First, to provide corroborating evidence that rent control caused values losses, we provide evidence that the law is binding. Using a difference-in-differences framework, Internet Appendix Table 4 shows that rents significantly declined by 3.6% to 8.9% relative to surrounding areas after the law was passed. Next, using US Department of Housing and Urban Development (HUD) data on monthly building permits, Internet Appendix Table 5 shows
that rent control caused a statistically significant decline in building permits in St. Paul, relative to comparable areas.
Second, we show our results are not driven by poorly matched control groups. Restricting the sample to the 54,701 transactions in cities that are directly adjacent to St. Paul or Minneapolis (as mapped in Internet Appendix Figure 4), Internet Appendix Table 6 shows that property values declined by −3.0% to −4.5%, which is slightly muted compared to the main results though still highly statistically significant. Spillover effects from directly adjacent areas in these tests will reduce the distinction between treated and control properties and bias the results towards zero (Autor et al., 2014; Campbell, Giglio and Pathak, 2011; Anenberg and Kung, 2014). Following Kline and Moretti (2014), Internet Appendix Table 7 addresses this concern by using a control sample that only includes observations from the five comparable metro areas excluding all observations from Minnesota. The estimates range from −3.9% to −6.5% and are statistically significant. Internet Appendix Table 8, uses transactions only in the downtown areas of the comparable metro areas as controls and finds effects of −12.5% to −15.3%. Finally, Internet Appendix Table 9, shows nearly identical results when we use observations that are averaged over ZIP code, city, and block group geographic levels, rather than transaction level observations.
Third, to provide evidence to support our assumption of parallel time trends, Figure 2 provides an event study graph. [Placeholder for figure description]
To provide additional credibility to the parallel trends assumption, Internet Appendix Table 10 reports estimates of the average treatment effect on the treated of −3.5% to −4.5% using the doubly robust improved estimator of Sant’Anna and Zhao (2020). Results are similar when we restrict the control sample to the adjacent cities. This estimator increases the credibility of the parallel trends assumption because it finds similar results but only requires that the parallel trends assumption holds conditional on the covariates of the model.
Fourth, Internet Appendix Table 11 addresses selection bias concerns using a battery of difference-in-differences regressions in which the dependent variable is an observable property characteristic, such as size and age. The estimates are economically small and statistically insignificant for all property traits. Internet Appendix Figure 5 shows that the distributions of observable traits of properties sold in the two quarters preceding the ballot are nearly identical to the distributions in the quarter following the ballot. To the extent that unobservable and observable characteristics are correlated, this finding indicates that the properties that were sold after the ballot are comparable to the ones that were sold before. Last, we use the methodology in Oster (2019) to show that our estimates are robust to even large amounts of unobservable bias in the data. We find that in order to shrink our estimates of the effect of rent control to zero, unobservables would need to have an impact on prices that is 19 times the impact of observables, which include micro-location, property size, and age.
V. Direct and Indirect Effects of Rent Control
Following our conceptual framework, the next step in our analysis is to test whether the observed decline in property values is driven by direct capitalization effects or indirect externality effects. As discussed above, the capitalization effect is amplified by the probability that a property is rented. Therefore, we test whether rental properties realize larger losses than owner-occupied properties.
Table 5 shows that rent control had a larger negative impact on rental properties than owner-occupied properties. In Panel A, we find that single-family residences that are rented experience an additional loss of 7.4% to 8.2% in value beyond single-family owner-occupied properties. This implies that single-family rental properties in St. Paul have a total loss of about 12%. In Panels B and C, we show that multi-unit properties also experience negative and significant price drops. Panel B includes all multi-unit properties and Panel C limits the sample to large multi-unit buildings with at least 8, 12, or 16 units. The effects are negative and range from −4.8% loss up to −21% loss for larger units. Given the relatively small sample size in these tests, the statistically and economic significance of the results indicates that rental properties were especially impacted by rent control. Because the results are stronger for rental properties than owner-occupied properties within St. Paul, it is less likely that the results are caused by a coincident policy change specific to St. Paul that affected all properties equally.
Negative effects for both owner-occupied and rental properties are consistent with the notion that rent control caused both a sizable, direct capitalization loss as well as an indirect loss from negative externalities. In Section VII, we provide quantitative estimates of average direct effects and externalities within the city, and of differences across neighborhoods, using a calibrated model.
VI. The Redistribution of Housing Wealth Within and Across Neighborhoods
The stated goal of St. Paul’s rent control law is to reduce the burden of housing costs, especially for “persons in low and moderate income households” (Saint Paul Legislative Code, 2021). It is clear from the results in the previous sections that the law generated a substantial decrease in housing wealth for properties across the entire city, including owner-occupied units. It is then natural to ask how the decrease in wealth was distributed across owners and renters, and whether the law met its goal of creating expected rent savings for low and moderate income households. The answers to these questions are not obvious, since the provisions of the law are uniform across the entire city. If the law binds differently in different neighborhoods, the most affected neighborhoods might not be locations in which low-income tenants live.
To answer these questions, this section of the paper presents a hedonic model of property values. We then analyze how demographic traits of landlords and renters correlate with the estimated changes in housing wealth.
1. Hedonic Model for Estimating Parcel-Level Value Losses
Of the single family residences, 90% are owner-occupied, 7% are rentals with small landlords (defined below), and 3% are rentals with large landlords. Of the two-to-three unit parcels, 74% are owned by small landlords, and the remaining 26% are owned by large landlords. Of parcels with four or more units, 39% are owned by small landlords and 61% are owned by large landlords. The majority of small landlords live in or near St. Paul. For all properties owned by small landlords, 89% of owners live in Minnesota, 63% live in the Twin-Cities area, and 41% live in St. Paul.
There are 78,221 parcels in St. Paul, including 73,103 residential parcels. Of the residential parcels with available data on the number of units, 64,960 are single-family residences, 6,093 are multi-unit parcels with 2–3 units, and 1,958 are apartments with four or more units. Due to missing fields in the administrative data, we can calculate the value loss for 64,654 single family residences, 5,926 two-to-three unit parcels, and 1,925 parcels with four or more units. Of the single family residences, 90% are owner-occupied, 7% are rentals with small landlords, and 3% are rentals with large landlords.
