Tenant Rights Eviction and Rent Affordability lowast

N Edward Coulson University of California Irvine dagger

Thao Le Georgia State University Dagger

Lily Shen Clemson University sect

July 9 2020

Abstract

We use state-level differences in the legal relationship between landlords and tenantsto estimate the impact of these differences on housing markets We construct a search-theoretic model of landlord and tenant search and matching which predicts that anincrease in the cost of eviction reduces the number of evictions but raises rents andhomeless rates and lowers housing supply and vacancy rates To test these predictionswe construct an annual index to measure the level of the legal protection of tenant rightsin each state Our instrumental variable results indicate while a one-unit increase in theTenant-Right Index reduces eviction rate by 89 percent rental housing is 61 percentmore expensive in areas where tenants have more protections against landlords Ahigher Tenant-Right Index is also associated with a decrease in housing supply andan increase in the homeless rate Taken together our findings highlight a significanttrade-off between tenant protections and rent affordability Thus the welfare effectsof tenant rights depend on the presumably large benefits for those who avoid evictionversus a loss of consumer surplus for other housing consumers

Key Words Tenant Rights Eviction Rent Affordability Landlord-Tenant LawsJEL codes I38 K25 R13 R28 R31

lowastFor helpful comments and discussions we would like to thank Ingrid Ellen Raven Molloy Carlos GarrigaPedro Gete Chris Cunningham and conferences participants at the 2019 Rental Housing Conference hostedby the Federal Reserve Bank of St Louis and the 2020 AREUEAAllied Social Science Association MeetingAll errors and omissions are ours alone

daggerPaul Merage School of Business University of California Irvine Irvine CADaggerJ Mack Robinson College of Business Georgia State University Atlanta GAsectDepartment of Finance Clemson University Clemson SC

1 Introduction

ldquoEviction isnrsquot just a condition of poverty itrsquos a cause of povertyrdquo

ndashMatthew Desmond Evicted

Every year approximately 23 million evictions are filed in the US Every minute four

renters in the US are forced out of their homes1 Research has established considerable ev-

idence that eviction-related residential mobility leads to many negative social and economic

consequences including adolescent violence (Sharkey and Sampson 2010) poor school per-

formance (Pribesh and Downey 1999) and damages on the physical and psychological well-

being (Dong et al 2005 and Oishi 2010) Moreover these eviction-induced consequences are

especially severe for the poor minorities women and children (Desmond 2012 Desmond

2016 South and Crowder 1998 Sampson and Sharkey 2008) As eviction becomes an in-

creasingly pressing issue across the nation growing tenant movements have been pushing for

stronger tenant protections and restrictions on no-cause evictions as part of the fight against

the housing affordability crisis (Desmond 2012 Desmond 2016) Borsch-Supan (1986) and

Bennett (2016) demonstrate that appropriate housing policy and statutory regulations are

vital in ensuring security of tenure for tenants

However overly strict regulations may force landlords to either find unfavorable ways to

circumvent the laws or leave the market completely As an anecdotal example in response

to a new bill passed in New York that bans landlords from requiring renters with bad credits

or low incomes to pay multiple months of rent upfront a property management company

claims ldquoWe are going to require guarantors or just reject the tenants outright So those

that have lower incomes are going to miss out because landlords are not going to take the

riskrdquo2 Studying the effect of rent control laws in San Francisco Diamond et al (2019)

document that affected landlords responded by converting their properties to condominiums

1Gross Terry 2018 ldquoFirst-Ever Evictions Database Shows Wersquore In the Mid-dle Of A Housing Crisisrdquo NPR April 18 httpswwwnprorg20180412601783346

first-ever-evictions-database-shows-were-in-the-middle-of-a-housing-crisis2Myers Emily 2019 ldquoLandlords canrsquot ask for rsquolast monthrsquos rentrsquo plus security deposit thanks to new

rent lawsrdquo Brick Underground December 19

1

1 Introduction

ldquoEviction isnrsquot just a condition of poverty itrsquos a cause of povertyrdquo

ndashMatthew Desmond Evicted

Every year approximately 23 million evictions are filed in the US Every minute four

renters in the US are forced out of their homes1 Research has established considerable ev-

idence that eviction-related residential mobility leads to many negative social and economic

consequences including adolescent violence (Sharkey and Sampson 2010) poor school per-

formance (Pribesh and Downey 1999) and damages on the physical and psychological well-

being (Dong et al 2005 and Oishi 2010) Moreover these eviction-induced consequences are

especially severe for the poor minorities women and children (Desmond 2012 Desmond

2016 South and Crowder 1998 Sampson and Sharkey 2008) As eviction becomes an in-

creasingly pressing issue across the nation growing tenant movements have been pushing for

stronger tenant protections and restrictions on no-cause evictions as part of the fight against

the housing affordability crisis (Desmond 2012 Desmond 2016) Borsch-Supan (1986) and

Bennett (2016) demonstrate that appropriate housing policy and statutory regulations are

vital in ensuring security of tenure for tenants

However overly strict regulations may force landlords to either find unfavorable ways to

circumvent the laws or leave the market completely As an anecdotal example in response

to a new bill passed in New York that bans landlords from requiring renters with bad credits

or low incomes to pay multiple months of rent upfront a property management company

claims ldquoWe are going to require guarantors or just reject the tenants outright So those

that have lower incomes are going to miss out because landlords are not going to take the

riskrdquo2 Studying the effect of rent control laws in San Francisco Diamond et al (2019)

document that affected landlords responded by converting their properties to condominiums

1Gross Terry 2018 ldquoFirst-Ever Evictions Database Shows Wersquore In the Mid-dle Of A Housing Crisisrdquo NPR April 18 httpswwwnprorg20180412601783346

first-ever-evictions-database-shows-were-in-the-middle-of-a-housing-crisis2Myers Emily 2019 ldquoLandlords canrsquot ask for rsquolast monthrsquos rentrsquo plus security deposit thanks to new

rent lawsrdquo Brick Underground December 19

1

or redeveloping buildings In the long run such substitution not only lowered the supply of

rental housing but also shifted the rental stock toward newer types of housing that cater to

higher-income renters and ultimately drove up rents in the city for the uncontrolled sector

Since excessive regulatory intervention can impose unintended spillover effects that impair

rent affordability (Been et al 2019 Molloy et al 2020) it is essential for policymakers to

understand the delicate balance between the strictness of landlord regulations evictions and

rent affordability to achieve their goal of increasing tenantsrsquo welfare The extant economics

literature has been silent on the direct impact of evictions on rent affordability due to the

absence of a comprehensive database on landlord regulations and eviction outcomes

In this study we bridge the literature gap by studying the relationship between tenant

rights evictions and other housing market outcomes We develop a simple search model in

which it is costly to evict tenants We show that an increase in eviction cost will lead to

higher rent lower supply of rental housing lower vacancy rate higher homeless rate and a

potentially higher eviction rate in equilibrium We then empirically test these predictions

using two new novel databases an eviction database released by the Eviction Lab in May

2018 and a hand-collected Tenant-Right Index database as a proxy for state-level landlord

regulations

To proxy for the cost of eviction we survey more than 15 years of landlord-tenant laws in

50 states across the US and the District of Columbia and create a yearly index to measure

the level of legal protection of tenant rights in each state Following the classic legal studies

literature on tenant eviction protections (Bennett 2016 Manheim 1989) we identify the top

twelve legal provisions that are most important in landlord-tenant relationships For each

provision we assign a score of 1 if a state is more friendly towards tenants than the average

state We then take the sum of all twelve provisions to construct our Tenant-Right Index

The index ranges from 0 to 12 with a higher index value indicating more legal protection

for tenants than for landlords which means a higher cost of eviction for landlords

We first use this index to test the effect of tenant rights on rent affordability Consistent

2

with our theoretical prediction we find that rental houses are less affordable where tenants

have better legal protection However one may argue that this observation is spurious lo-

cal authorities in areas with overwhelmingly high rental costs may have more incentives to

increase tenant protections against landlords To address this concern we employ an instru-

mental variable approach utilizing voting outcomes in presidential elections In particular

we use the percentage of votes for the Democratic candidates at the state level as the instru-

ment for our Tenant-Right Index We argue that states supporting the Democratic Party are

more likely to have pro-tenant laws as the party is often known for promoting social welfare

benefits The identifying assumption is that a statersquos affiliation with a particular political

party should have no direct bearing on rental prices except through its effect on landlord-

tenant laws Using the instrumental variable analysis our estimation results indicate that

a one-unit increase in the Tenant-Right Index is associated with a 61 percent increase in

the median rent price It also reduces the supply of rental housing by 21 percent and the

vacancy rate by 89 percent Interestingly we observe that places with more substantial

tenant rights experience more homelessness which appears counter-intuitive but is simply a

result of reduced supply

Finally we ask whether enacting more tenant-favorable laws is effective in reducing evic-

tion rates We find that landlords are less likely to file for eviction judgments and such

lawsuits are also less likely to be supported by judges in tenant-friendly regions A one-unit

increase in the Tenant-Right Index is correlated with an 89 percent decrease in the evic-

tion rate Together with the findings on housing market outcomes our results suggest that

stricter landlord regulations may help protect tenants from evictions and many eviction-

induced hardships but at the cost of higher rents lower rental units supply and higher

homeless rates

Understanding the tension between these opposite effects is essential to guide policymak-

ing in this area Our study provides support for the use of stringent landlord regulations in

tackling the eviction crisis a proposition often put forward by advocates of tenant rights

3

if the main objective is preventing eviction and the substantial social and economic costs it

entails At the same time policymakers need to balance this goal against the negative im-

pacts on rent and supply to avoid exacerbating the housing affordability problem that many

cities are already struggling with To the extent that they can even increase homelessness

overly burdensome regulations only counteract the stated welfare-enhancing goal of the laws

For a complete policy evaluation it is necessary to compare the benefits enjoyed by tenants

avoiding eviction to the loss of consumer surplus for others Still such analysis is beyond

the scope of this paper

Our theoretical and empirical findings are parallel to prior research conclusions in the

closely related topic of rent control On the one hand general consensus agrees that rent

control policies keep rents affordable increase residential stability and reduce eviction risks

for the tenant population they target (Autor et al 2014 Diamond et al 2019 Early 2000

Glaeser and Luttmer 2003 Sims 2007) On the other hand these benefits may be offset

by the negative effects they have on housing quality because landlords are discouraged from

maintaining their properties beyond the minimum requirement (Arnott and Shevyakhova

2014 Gyourko and Linneman 1989) More importantly such laws reduce the supply of

rental housing in the long run (Autor et al 2014 Sims 2007) by inducing the conversion of

existing units to owner-occupancy (Diamond et al 2019) as well as making it less profitable

to build new development (Early 2000 Glaeser and Luttmer 2003) As such rent control

laws in the US generally exempt new construction to avoid this problem

The rest of the paper is organized as follows We introduce the theoretical motivation in

Section 2 We detail the construction method of the Tenant-Right Index in Section 3 We

describe the empirical methodology and present the descriptive statistics of the supporting

data in Section 4 We report our empirical findings in Section 5 Section 6 concludes the

paper

4

2 Theoretical Motivation

In this section we construct a search model in the manner of Pissarides (2000) applied to

the rental housing market Landlords with vacant units and renters seeking housing engage

in a search process to match and bargain over rents Landlords and tenants then observe

whether the tenants are good or bad in the sense that bad tenants create extra costs for the

landlords through bothersome acts poor care of the unit or similar acts3 We extend the

standard search model by allowing landlords to evict bad tenants but they incur a cost in the

attempt Although we do not model the source of this cost explicitly in our empirical tests

later we posit that eviction cost depends on the degree of statutory regulations imposed on

landlords by the state

21 Landlords

Landlords own a single unit which is characterized by one of two states occupancy (O)

or vacancy (V ) If the unit is currently vacant the landlord pays carrying costs c each

period of vacancy They match with tenants at rate λ to be determined endogenously and

bargain over nominal rent R As standard in search models the matching rate λ is the

probability of the landlord making contact with a tenant which depends on the relative

number of potential renters to vacant units With r as the discount rate and πi where

iεO V denoting the present value of future net income the flow value function of the

landlord in the vacancy state is given by the Bellman equation

rπV = minusc+ λ(EπO minus πV ) (1)

Once the unit becomes occupied the tenantrsquos quality is revealed as either good or bad

If the tenant is bad the landlord must pay y in extra maintenance costs each period while

3We assume that tenants do not gain from this act and they pay their rent regardless This eliminatesstrategic behavior on the part of the tenants We discuss the implication of relaxing this assumption inSection 25

5

a good tenant imposes no such costs We assume that no eviction is attempted in the case

of good tenant and the flow value of utility of the landlord is then

rπO(y good) = R + δ(πV minus πO(y good)) (2)

where δ is the exogenous probability that the tenant and landlord become mismatched and

separate at any point in time Note that Equation (2) becomes

πO(y good) =R + δπVr + δ

(3)

If instead the tenant is revealed as bad the landlord may choose to evict them or not

The period utility when no eviction is chosen equals

rπO(y keep) = (Rminus y) + δ(πV minus πO(y keep)) (4)

Alternatively the landlord may try to evict a bad tenant by paying a per period fee of d

thereby raising the separation rate by an exogenous amount ε

rπO(y evict) = (Rminus y minus d) + (δ + ε)(πV minus πO(y evict)) (5)

Rewriting Equation (4) and Equation (5) respectively yields

πO(y keep) =Rminus y + δπV

r + δ (6)

and

πO(y evict) =Rminus y minus d+ (δ + ε)πV

r + δ + ε (7)

Eviction will occur when the present value of profits from eviction is greater than that

of keeping the bad tenant Combining Equation (6) and Equation (7) this condition can be

6

written as

Rminus y lt rπV minusd(r + δ)

ε (8)

The left hand side is net rent The right hand side is a parameter cluster that represents

the net flow benefit of eviction (not including the lost rent on the left hand side) This net

benefit is higher when the probability of successful eviction is higher or when the value of a

vacancy is higher and is lower when the cost of eviction is higher or when the detachment

rate is higher In the latter case a higher probability that the (bad) tenant will leave anyway

lowers the value of deliberately evicting them

Note that the eviction rate ε when it is not zero is exogenous so that the comparative

static response of evictions to changes in d is limited to the case where d rises by enough to

flip the inequality of (8) Nevertheless the model delivers the expected result that a rise in

the cost of eviction can lower the number of evictions In order to obtain other comparative

static impacts of eviction cost d it is now necessary to assume that Equation (8) holds so

that landlords always choose to evict bad tenants Using p as the share of bad tenants in

the market the expected gains for landlord from matching with a tenant then becomes

EπO minus πV = (1minus p) R + δπVr + δ

+ pRminus y minus d+ (δ + ε) πV

r + δ + εminus πV (9)

=(Rminus rπV ) (r + δ + (1minus p) ε)minus p(d+ y)(r + δ)

(r + δ)(r + δ + ε) (10)

We can now replace rπV with its value from Equation (1) and rearrange Equation (10)

to get

EπO minus πV =(R + c) (r + δ + ε (1minus p))minus p(d+ y)(r + δ)

(r + δ) (r + δ + ε) + λ((r + δ + ε (1minus p))(11)

7

=(R + c) θ2 minus θ3

θ1 + θ2λ (12)

where θ1 = (r + δ) (r + δ + ε) θ2 = r + δ + ε (1minus p) and θ3(d) = p(d+ y)(r + δ)

22 Tenants

We now turn to the derivation of the tenantrsquos utility functions in their respective states

housed (H) and unhoused (U) Recalling that their draw of y is unknown to them before

they are housed we let J described their lifetime utility from any given state and write the

flow value of being unhoused

rJU = micro(EJH minus JU) (13)

where cash flow is normalized to zero and micro is the (endogenous) matching rate of tenants

to landlords It is the rate at which tenants make contact with a landlord which is again

a function of the ratio of unhoused tenants to vacant units in the market Once they are

housed their draw of y is revealed and they observe whether the landlord is trying to evict

them With Z notating the benefit of being housed we have

rJH(y stay) = (Z minusR) + δ(JU minus JH(Y stay)) (14)

rJH(y evict) = (Z minusR) + (δ + ε)(JU minus JH(Y evict)) (15)

As a simplification we assume that tenants pay full rent and does not receive any benefit

if they are revealed to be bad This prevents tenants from strategically acting like bad

tenants in no-eviction markets An assumption to justify this would be that y represents

lack of care but that the tenant receives no (leisure) benefit from this lack We will return

to this assumption in Section 25 to discuss the impact of its relaxation on our model

8

To determine the expectation of lifetime utility from being housed first write using

Equation (14) and Equation (15) respectively

JH(y stay) =(Z minusR) + δJU

r + δ (16)

and

JH(y evict) =(Z minusR) + (δ + ε)JU

r + δ + ε (17)

so that as long as eviction is undertaken by landlords and remembering that the tenant

does not know their quality ahead of the match we have

EJH minus JU = p(Z minusR) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (18)

=(r + δ + (1minus p)ε)((Z minusR)minus rJU)

(r + δ)(r + δ + ε) (19)

Substitute the expression for the flow value of the unhoused to get from inserting Equation

(13) into Equation (18) and Equation (19) respectively

(EJH minus JU) =θ2(Z minusR)

θ1 + θ2micro (20)

23 Rents

Rents are determined in a Nash Bargain which (by assumption of equal bargaining power)

equates the gains from agreement obtained by landlord and tenant

(R + c) θ2 minus θ3θ1 + θ2λ

=θ2(Z minusR)

θ1 + θ2micro (21)

9

which yields rent as a function of the two endogenous contact rates

R =(θ3 minus cθ2)(θ1 + θ2micro) + Z(θ1 + θ2λ)θ2

θ2(2θ1 + θ2 (λ+ micro))= R(λ micro) (22)

where the derivative of R with respect to λ is positive and with respect to micro is negative

under the condition that (θ3 minus cθ2) lt 0 A sufficient condition for this is p (d+ y) lt c

which from Equation (12) is easily seen to be a sufficient condition for landlord entry into

the market These are standard results from search and bargaining models since an increase

in onersquos contact rate raises onersquos bargaining power and tilts the rent in a favorable direction

We assume free entry such that the profits from holding a vacancy equals the fixed costs

of entering the market

πV = ϕ (23)

From (1) (9) (22) and (23) we get

rϕ+ c

λ=

(R(λ micro) + c) θ2 minus θ3(d)

θ1 + θ2λ(24)

which defines an equilibrium entry zero-profit condition (ZP ) in the two contact rates In

the (micro λ)-space it is straightforward to show that the implicit function derived from Equation

(24) is upward sloping An increase in the tenant contact rate necessitates an increase in the

landlord contact rate (via landlord exit as discussed below) in order to return to zero profit

There are L landlords whose units are vacant or occupied and T tenants housed or

unhoused measured on a continuum

L = LV + LO (25)

T = TH + TU (26)

10

Note that T is assumed to be exogenous in order to anchor the model but L is determined

endogenously as the number of available units must satisfy a zero profit condition The

number of matches in any given period is generated by a standard matching function

λLV = microTU = M (LV TU) (27)

The matching function M is assumed to be constant returns to scale so that

λ = M(1 λmicro) (28)

where λmicro is ldquomarket tightnessrdquo It is straightforward to see that Equation (28) implicitly

defines a downward sloping function in the (micro λ) space which we refer to as the matching

condition

Given certain mild conditions on the matching function (Coulson et al 2001) there exists

a unique solution to the zero-profit condition and the matching condition that establishes

the equilibrium values of the two contact rates The comparative statics of the model are

intuitive to establish (Figure 1) A change in any of the modelrsquos parameters implies a shift

in the zero-profit function that re-equilibrates micro and λ

Our main interest is in the comparative static responses to changes in d the cost of

eviction For ease of presentation we eschew comparative static analysis of changes in ε We

can therefore interpret d as the cost of obtaining the given eviction rate Examination of

Equation (24) indicates that an increase in d will shift up the zero-profit line For any given

micro a higher d requires an increase in λ to maintain zero profits As Figure 1 therefore shows

a rise in d implies both higher λ and lower micro This in turn implies from Equation (22) a

higher equilibrium rent

11

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

with our theoretical prediction we find that rental houses are less affordable where tenants

have better legal protection However one may argue that this observation is spurious lo-

cal authorities in areas with overwhelmingly high rental costs may have more incentives to

increase tenant protections against landlords To address this concern we employ an instru-

mental variable approach utilizing voting outcomes in presidential elections In particular

we use the percentage of votes for the Democratic candidates at the state level as the instru-

ment for our Tenant-Right Index We argue that states supporting the Democratic Party are

more likely to have pro-tenant laws as the party is often known for promoting social welfare

benefits The identifying assumption is that a statersquos affiliation with a particular political

party should have no direct bearing on rental prices except through its effect on landlord-

tenant laws Using the instrumental variable analysis our estimation results indicate that

a one-unit increase in the Tenant-Right Index is associated with a 61 percent increase in

the median rent price It also reduces the supply of rental housing by 21 percent and the

vacancy rate by 89 percent Interestingly we observe that places with more substantial

tenant rights experience more homelessness which appears counter-intuitive but is simply a

result of reduced supply

Finally we ask whether enacting more tenant-favorable laws is effective in reducing evic-

tion rates We find that landlords are less likely to file for eviction judgments and such

lawsuits are also less likely to be supported by judges in tenant-friendly regions A one-unit

increase in the Tenant-Right Index is correlated with an 89 percent decrease in the evic-

tion rate Together with the findings on housing market outcomes our results suggest that

stricter landlord regulations may help protect tenants from evictions and many eviction-

induced hardships but at the cost of higher rents lower rental units supply and higher

homeless rates

Understanding the tension between these opposite effects is essential to guide policymak-

ing in this area Our study provides support for the use of stringent landlord regulations in

tackling the eviction crisis a proposition often put forward by advocates of tenant rights

3

if the main objective is preventing eviction and the substantial social and economic costs it

entails At the same time policymakers need to balance this goal against the negative im-

pacts on rent and supply to avoid exacerbating the housing affordability problem that many

cities are already struggling with To the extent that they can even increase homelessness

overly burdensome regulations only counteract the stated welfare-enhancing goal of the laws

For a complete policy evaluation it is necessary to compare the benefits enjoyed by tenants

avoiding eviction to the loss of consumer surplus for others Still such analysis is beyond

the scope of this paper

Our theoretical and empirical findings are parallel to prior research conclusions in the

closely related topic of rent control On the one hand general consensus agrees that rent

control policies keep rents affordable increase residential stability and reduce eviction risks

for the tenant population they target (Autor et al 2014 Diamond et al 2019 Early 2000

Glaeser and Luttmer 2003 Sims 2007) On the other hand these benefits may be offset

by the negative effects they have on housing quality because landlords are discouraged from

maintaining their properties beyond the minimum requirement (Arnott and Shevyakhova

2014 Gyourko and Linneman 1989) More importantly such laws reduce the supply of

rental housing in the long run (Autor et al 2014 Sims 2007) by inducing the conversion of

existing units to owner-occupancy (Diamond et al 2019) as well as making it less profitable

to build new development (Early 2000 Glaeser and Luttmer 2003) As such rent control

laws in the US generally exempt new construction to avoid this problem

The rest of the paper is organized as follows We introduce the theoretical motivation in

Section 2 We detail the construction method of the Tenant-Right Index in Section 3 We

describe the empirical methodology and present the descriptive statistics of the supporting

data in Section 4 We report our empirical findings in Section 5 Section 6 concludes the

paper

4

2 Theoretical Motivation

In this section we construct a search model in the manner of Pissarides (2000) applied to

the rental housing market Landlords with vacant units and renters seeking housing engage

in a search process to match and bargain over rents Landlords and tenants then observe

whether the tenants are good or bad in the sense that bad tenants create extra costs for the

landlords through bothersome acts poor care of the unit or similar acts3 We extend the

standard search model by allowing landlords to evict bad tenants but they incur a cost in the

attempt Although we do not model the source of this cost explicitly in our empirical tests

later we posit that eviction cost depends on the degree of statutory regulations imposed on

landlords by the state

21 Landlords

Landlords own a single unit which is characterized by one of two states occupancy (O)

or vacancy (V ) If the unit is currently vacant the landlord pays carrying costs c each

period of vacancy They match with tenants at rate λ to be determined endogenously and

bargain over nominal rent R As standard in search models the matching rate λ is the

probability of the landlord making contact with a tenant which depends on the relative

number of potential renters to vacant units With r as the discount rate and πi where

iεO V denoting the present value of future net income the flow value function of the

landlord in the vacancy state is given by the Bellman equation

rπV = minusc+ λ(EπO minus πV ) (1)

Once the unit becomes occupied the tenantrsquos quality is revealed as either good or bad

If the tenant is bad the landlord must pay y in extra maintenance costs each period while

3We assume that tenants do not gain from this act and they pay their rent regardless This eliminatesstrategic behavior on the part of the tenants We discuss the implication of relaxing this assumption inSection 25

5

a good tenant imposes no such costs We assume that no eviction is attempted in the case

of good tenant and the flow value of utility of the landlord is then

rπO(y good) = R + δ(πV minus πO(y good)) (2)

where δ is the exogenous probability that the tenant and landlord become mismatched and

separate at any point in time Note that Equation (2) becomes

πO(y good) =R + δπVr + δ

(3)

If instead the tenant is revealed as bad the landlord may choose to evict them or not

The period utility when no eviction is chosen equals

rπO(y keep) = (Rminus y) + δ(πV minus πO(y keep)) (4)

Alternatively the landlord may try to evict a bad tenant by paying a per period fee of d

thereby raising the separation rate by an exogenous amount ε

rπO(y evict) = (Rminus y minus d) + (δ + ε)(πV minus πO(y evict)) (5)

Rewriting Equation (4) and Equation (5) respectively yields

πO(y keep) =Rminus y + δπV

r + δ (6)

and

πO(y evict) =Rminus y minus d+ (δ + ε)πV

r + δ + ε (7)

Eviction will occur when the present value of profits from eviction is greater than that

of keeping the bad tenant Combining Equation (6) and Equation (7) this condition can be

6

written as

Rminus y lt rπV minusd(r + δ)

ε (8)

The left hand side is net rent The right hand side is a parameter cluster that represents

the net flow benefit of eviction (not including the lost rent on the left hand side) This net

benefit is higher when the probability of successful eviction is higher or when the value of a

vacancy is higher and is lower when the cost of eviction is higher or when the detachment

rate is higher In the latter case a higher probability that the (bad) tenant will leave anyway

lowers the value of deliberately evicting them

Note that the eviction rate ε when it is not zero is exogenous so that the comparative

static response of evictions to changes in d is limited to the case where d rises by enough to

flip the inequality of (8) Nevertheless the model delivers the expected result that a rise in

the cost of eviction can lower the number of evictions In order to obtain other comparative

static impacts of eviction cost d it is now necessary to assume that Equation (8) holds so

that landlords always choose to evict bad tenants Using p as the share of bad tenants in

the market the expected gains for landlord from matching with a tenant then becomes

EπO minus πV = (1minus p) R + δπVr + δ

+ pRminus y minus d+ (δ + ε) πV

r + δ + εminus πV (9)

=(Rminus rπV ) (r + δ + (1minus p) ε)minus p(d+ y)(r + δ)

(r + δ)(r + δ + ε) (10)

We can now replace rπV with its value from Equation (1) and rearrange Equation (10)

to get

EπO minus πV =(R + c) (r + δ + ε (1minus p))minus p(d+ y)(r + δ)

(r + δ) (r + δ + ε) + λ((r + δ + ε (1minus p))(11)

7

=(R + c) θ2 minus θ3

θ1 + θ2λ (12)

where θ1 = (r + δ) (r + δ + ε) θ2 = r + δ + ε (1minus p) and θ3(d) = p(d+ y)(r + δ)

22 Tenants

We now turn to the derivation of the tenantrsquos utility functions in their respective states

housed (H) and unhoused (U) Recalling that their draw of y is unknown to them before

they are housed we let J described their lifetime utility from any given state and write the

flow value of being unhoused

rJU = micro(EJH minus JU) (13)

where cash flow is normalized to zero and micro is the (endogenous) matching rate of tenants

to landlords It is the rate at which tenants make contact with a landlord which is again

a function of the ratio of unhoused tenants to vacant units in the market Once they are

housed their draw of y is revealed and they observe whether the landlord is trying to evict

them With Z notating the benefit of being housed we have

rJH(y stay) = (Z minusR) + δ(JU minus JH(Y stay)) (14)

rJH(y evict) = (Z minusR) + (δ + ε)(JU minus JH(Y evict)) (15)

As a simplification we assume that tenants pay full rent and does not receive any benefit

if they are revealed to be bad This prevents tenants from strategically acting like bad

tenants in no-eviction markets An assumption to justify this would be that y represents

lack of care but that the tenant receives no (leisure) benefit from this lack We will return

to this assumption in Section 25 to discuss the impact of its relaxation on our model

8

To determine the expectation of lifetime utility from being housed first write using

Equation (14) and Equation (15) respectively

JH(y stay) =(Z minusR) + δJU

r + δ (16)

and

JH(y evict) =(Z minusR) + (δ + ε)JU

r + δ + ε (17)

so that as long as eviction is undertaken by landlords and remembering that the tenant

does not know their quality ahead of the match we have

EJH minus JU = p(Z minusR) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (18)

=(r + δ + (1minus p)ε)((Z minusR)minus rJU)

(r + δ)(r + δ + ε) (19)

Substitute the expression for the flow value of the unhoused to get from inserting Equation

(13) into Equation (18) and Equation (19) respectively

(EJH minus JU) =θ2(Z minusR)

θ1 + θ2micro (20)

23 Rents

Rents are determined in a Nash Bargain which (by assumption of equal bargaining power)

equates the gains from agreement obtained by landlord and tenant

(R + c) θ2 minus θ3θ1 + θ2λ

=θ2(Z minusR)

θ1 + θ2micro (21)

9

which yields rent as a function of the two endogenous contact rates

R =(θ3 minus cθ2)(θ1 + θ2micro) + Z(θ1 + θ2λ)θ2

θ2(2θ1 + θ2 (λ+ micro))= R(λ micro) (22)

where the derivative of R with respect to λ is positive and with respect to micro is negative

under the condition that (θ3 minus cθ2) lt 0 A sufficient condition for this is p (d+ y) lt c

which from Equation (12) is easily seen to be a sufficient condition for landlord entry into

the market These are standard results from search and bargaining models since an increase

in onersquos contact rate raises onersquos bargaining power and tilts the rent in a favorable direction

We assume free entry such that the profits from holding a vacancy equals the fixed costs

of entering the market

πV = ϕ (23)

From (1) (9) (22) and (23) we get

rϕ+ c

λ=

(R(λ micro) + c) θ2 minus θ3(d)

θ1 + θ2λ(24)

which defines an equilibrium entry zero-profit condition (ZP ) in the two contact rates In

the (micro λ)-space it is straightforward to show that the implicit function derived from Equation

(24) is upward sloping An increase in the tenant contact rate necessitates an increase in the

landlord contact rate (via landlord exit as discussed below) in order to return to zero profit

There are L landlords whose units are vacant or occupied and T tenants housed or

unhoused measured on a continuum

L = LV + LO (25)

T = TH + TU (26)

10

Note that T is assumed to be exogenous in order to anchor the model but L is determined

endogenously as the number of available units must satisfy a zero profit condition The

number of matches in any given period is generated by a standard matching function

λLV = microTU = M (LV TU) (27)

The matching function M is assumed to be constant returns to scale so that

λ = M(1 λmicro) (28)

where λmicro is ldquomarket tightnessrdquo It is straightforward to see that Equation (28) implicitly

defines a downward sloping function in the (micro λ) space which we refer to as the matching

condition

Given certain mild conditions on the matching function (Coulson et al 2001) there exists

a unique solution to the zero-profit condition and the matching condition that establishes

the equilibrium values of the two contact rates The comparative statics of the model are

intuitive to establish (Figure 1) A change in any of the modelrsquos parameters implies a shift

in the zero-profit function that re-equilibrates micro and λ

Our main interest is in the comparative static responses to changes in d the cost of

eviction For ease of presentation we eschew comparative static analysis of changes in ε We

can therefore interpret d as the cost of obtaining the given eviction rate Examination of

Equation (24) indicates that an increase in d will shift up the zero-profit line For any given

micro a higher d requires an increase in λ to maintain zero profits As Figure 1 therefore shows

a rise in d implies both higher λ and lower micro This in turn implies from Equation (22) a

higher equilibrium rent

11

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

if the main objective is preventing eviction and the substantial social and economic costs it

entails At the same time policymakers need to balance this goal against the negative im-

pacts on rent and supply to avoid exacerbating the housing affordability problem that many

cities are already struggling with To the extent that they can even increase homelessness

overly burdensome regulations only counteract the stated welfare-enhancing goal of the laws

For a complete policy evaluation it is necessary to compare the benefits enjoyed by tenants

avoiding eviction to the loss of consumer surplus for others Still such analysis is beyond

the scope of this paper

Our theoretical and empirical findings are parallel to prior research conclusions in the

closely related topic of rent control On the one hand general consensus agrees that rent

control policies keep rents affordable increase residential stability and reduce eviction risks

for the tenant population they target (Autor et al 2014 Diamond et al 2019 Early 2000

Glaeser and Luttmer 2003 Sims 2007) On the other hand these benefits may be offset

by the negative effects they have on housing quality because landlords are discouraged from

maintaining their properties beyond the minimum requirement (Arnott and Shevyakhova

2014 Gyourko and Linneman 1989) More importantly such laws reduce the supply of

rental housing in the long run (Autor et al 2014 Sims 2007) by inducing the conversion of

existing units to owner-occupancy (Diamond et al 2019) as well as making it less profitable

to build new development (Early 2000 Glaeser and Luttmer 2003) As such rent control

laws in the US generally exempt new construction to avoid this problem

The rest of the paper is organized as follows We introduce the theoretical motivation in

Section 2 We detail the construction method of the Tenant-Right Index in Section 3 We

describe the empirical methodology and present the descriptive statistics of the supporting

data in Section 4 We report our empirical findings in Section 5 Section 6 concludes the

paper

4

2 Theoretical Motivation

In this section we construct a search model in the manner of Pissarides (2000) applied to

the rental housing market Landlords with vacant units and renters seeking housing engage

in a search process to match and bargain over rents Landlords and tenants then observe

whether the tenants are good or bad in the sense that bad tenants create extra costs for the

landlords through bothersome acts poor care of the unit or similar acts3 We extend the

standard search model by allowing landlords to evict bad tenants but they incur a cost in the

attempt Although we do not model the source of this cost explicitly in our empirical tests

later we posit that eviction cost depends on the degree of statutory regulations imposed on

landlords by the state

21 Landlords

Landlords own a single unit which is characterized by one of two states occupancy (O)

or vacancy (V ) If the unit is currently vacant the landlord pays carrying costs c each

period of vacancy They match with tenants at rate λ to be determined endogenously and

bargain over nominal rent R As standard in search models the matching rate λ is the

probability of the landlord making contact with a tenant which depends on the relative

number of potential renters to vacant units With r as the discount rate and πi where

iεO V denoting the present value of future net income the flow value function of the

landlord in the vacancy state is given by the Bellman equation

rπV = minusc+ λ(EπO minus πV ) (1)

Once the unit becomes occupied the tenantrsquos quality is revealed as either good or bad

If the tenant is bad the landlord must pay y in extra maintenance costs each period while

3We assume that tenants do not gain from this act and they pay their rent regardless This eliminatesstrategic behavior on the part of the tenants We discuss the implication of relaxing this assumption inSection 25

5

a good tenant imposes no such costs We assume that no eviction is attempted in the case

of good tenant and the flow value of utility of the landlord is then

rπO(y good) = R + δ(πV minus πO(y good)) (2)

where δ is the exogenous probability that the tenant and landlord become mismatched and

separate at any point in time Note that Equation (2) becomes

πO(y good) =R + δπVr + δ

(3)

If instead the tenant is revealed as bad the landlord may choose to evict them or not

The period utility when no eviction is chosen equals

rπO(y keep) = (Rminus y) + δ(πV minus πO(y keep)) (4)

Alternatively the landlord may try to evict a bad tenant by paying a per period fee of d

thereby raising the separation rate by an exogenous amount ε

rπO(y evict) = (Rminus y minus d) + (δ + ε)(πV minus πO(y evict)) (5)

Rewriting Equation (4) and Equation (5) respectively yields

πO(y keep) =Rminus y + δπV

r + δ (6)

and

πO(y evict) =Rminus y minus d+ (δ + ε)πV

r + δ + ε (7)

Eviction will occur when the present value of profits from eviction is greater than that

of keeping the bad tenant Combining Equation (6) and Equation (7) this condition can be

6

written as

Rminus y lt rπV minusd(r + δ)

ε (8)

The left hand side is net rent The right hand side is a parameter cluster that represents

the net flow benefit of eviction (not including the lost rent on the left hand side) This net

benefit is higher when the probability of successful eviction is higher or when the value of a

vacancy is higher and is lower when the cost of eviction is higher or when the detachment

rate is higher In the latter case a higher probability that the (bad) tenant will leave anyway

lowers the value of deliberately evicting them

Note that the eviction rate ε when it is not zero is exogenous so that the comparative

static response of evictions to changes in d is limited to the case where d rises by enough to

flip the inequality of (8) Nevertheless the model delivers the expected result that a rise in

the cost of eviction can lower the number of evictions In order to obtain other comparative

static impacts of eviction cost d it is now necessary to assume that Equation (8) holds so

that landlords always choose to evict bad tenants Using p as the share of bad tenants in

the market the expected gains for landlord from matching with a tenant then becomes

EπO minus πV = (1minus p) R + δπVr + δ

+ pRminus y minus d+ (δ + ε) πV

r + δ + εminus πV (9)

=(Rminus rπV ) (r + δ + (1minus p) ε)minus p(d+ y)(r + δ)

(r + δ)(r + δ + ε) (10)

We can now replace rπV with its value from Equation (1) and rearrange Equation (10)

to get

EπO minus πV =(R + c) (r + δ + ε (1minus p))minus p(d+ y)(r + δ)

(r + δ) (r + δ + ε) + λ((r + δ + ε (1minus p))(11)

7

=(R + c) θ2 minus θ3

θ1 + θ2λ (12)

where θ1 = (r + δ) (r + δ + ε) θ2 = r + δ + ε (1minus p) and θ3(d) = p(d+ y)(r + δ)

22 Tenants

We now turn to the derivation of the tenantrsquos utility functions in their respective states

housed (H) and unhoused (U) Recalling that their draw of y is unknown to them before

they are housed we let J described their lifetime utility from any given state and write the

flow value of being unhoused

rJU = micro(EJH minus JU) (13)

where cash flow is normalized to zero and micro is the (endogenous) matching rate of tenants

to landlords It is the rate at which tenants make contact with a landlord which is again

a function of the ratio of unhoused tenants to vacant units in the market Once they are

housed their draw of y is revealed and they observe whether the landlord is trying to evict

them With Z notating the benefit of being housed we have

rJH(y stay) = (Z minusR) + δ(JU minus JH(Y stay)) (14)

rJH(y evict) = (Z minusR) + (δ + ε)(JU minus JH(Y evict)) (15)

As a simplification we assume that tenants pay full rent and does not receive any benefit

if they are revealed to be bad This prevents tenants from strategically acting like bad

tenants in no-eviction markets An assumption to justify this would be that y represents

lack of care but that the tenant receives no (leisure) benefit from this lack We will return

to this assumption in Section 25 to discuss the impact of its relaxation on our model

8

To determine the expectation of lifetime utility from being housed first write using

Equation (14) and Equation (15) respectively

JH(y stay) =(Z minusR) + δJU

r + δ (16)

and

JH(y evict) =(Z minusR) + (δ + ε)JU

r + δ + ε (17)

so that as long as eviction is undertaken by landlords and remembering that the tenant

does not know their quality ahead of the match we have

EJH minus JU = p(Z minusR) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (18)

=(r + δ + (1minus p)ε)((Z minusR)minus rJU)

(r + δ)(r + δ + ε) (19)

Substitute the expression for the flow value of the unhoused to get from inserting Equation

(13) into Equation (18) and Equation (19) respectively

(EJH minus JU) =θ2(Z minusR)

θ1 + θ2micro (20)

23 Rents

Rents are determined in a Nash Bargain which (by assumption of equal bargaining power)

equates the gains from agreement obtained by landlord and tenant

(R + c) θ2 minus θ3θ1 + θ2λ

=θ2(Z minusR)

θ1 + θ2micro (21)

9

which yields rent as a function of the two endogenous contact rates

R =(θ3 minus cθ2)(θ1 + θ2micro) + Z(θ1 + θ2λ)θ2

θ2(2θ1 + θ2 (λ+ micro))= R(λ micro) (22)

where the derivative of R with respect to λ is positive and with respect to micro is negative

under the condition that (θ3 minus cθ2) lt 0 A sufficient condition for this is p (d+ y) lt c

which from Equation (12) is easily seen to be a sufficient condition for landlord entry into

the market These are standard results from search and bargaining models since an increase

in onersquos contact rate raises onersquos bargaining power and tilts the rent in a favorable direction

We assume free entry such that the profits from holding a vacancy equals the fixed costs

of entering the market

πV = ϕ (23)

From (1) (9) (22) and (23) we get

rϕ+ c

λ=

(R(λ micro) + c) θ2 minus θ3(d)

θ1 + θ2λ(24)

which defines an equilibrium entry zero-profit condition (ZP ) in the two contact rates In

the (micro λ)-space it is straightforward to show that the implicit function derived from Equation

(24) is upward sloping An increase in the tenant contact rate necessitates an increase in the

landlord contact rate (via landlord exit as discussed below) in order to return to zero profit

There are L landlords whose units are vacant or occupied and T tenants housed or

unhoused measured on a continuum

L = LV + LO (25)

T = TH + TU (26)

10

Note that T is assumed to be exogenous in order to anchor the model but L is determined

endogenously as the number of available units must satisfy a zero profit condition The

number of matches in any given period is generated by a standard matching function

λLV = microTU = M (LV TU) (27)

The matching function M is assumed to be constant returns to scale so that

λ = M(1 λmicro) (28)

where λmicro is ldquomarket tightnessrdquo It is straightforward to see that Equation (28) implicitly

defines a downward sloping function in the (micro λ) space which we refer to as the matching

condition

Given certain mild conditions on the matching function (Coulson et al 2001) there exists

a unique solution to the zero-profit condition and the matching condition that establishes

the equilibrium values of the two contact rates The comparative statics of the model are

intuitive to establish (Figure 1) A change in any of the modelrsquos parameters implies a shift

in the zero-profit function that re-equilibrates micro and λ

Our main interest is in the comparative static responses to changes in d the cost of

eviction For ease of presentation we eschew comparative static analysis of changes in ε We

can therefore interpret d as the cost of obtaining the given eviction rate Examination of

Equation (24) indicates that an increase in d will shift up the zero-profit line For any given

micro a higher d requires an increase in λ to maintain zero profits As Figure 1 therefore shows

a rise in d implies both higher λ and lower micro This in turn implies from Equation (22) a

higher equilibrium rent

11

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

2 Theoretical Motivation

In this section we construct a search model in the manner of Pissarides (2000) applied to

the rental housing market Landlords with vacant units and renters seeking housing engage

in a search process to match and bargain over rents Landlords and tenants then observe

whether the tenants are good or bad in the sense that bad tenants create extra costs for the

landlords through bothersome acts poor care of the unit or similar acts3 We extend the

standard search model by allowing landlords to evict bad tenants but they incur a cost in the

attempt Although we do not model the source of this cost explicitly in our empirical tests

later we posit that eviction cost depends on the degree of statutory regulations imposed on

landlords by the state

21 Landlords

Landlords own a single unit which is characterized by one of two states occupancy (O)

or vacancy (V ) If the unit is currently vacant the landlord pays carrying costs c each

period of vacancy They match with tenants at rate λ to be determined endogenously and

bargain over nominal rent R As standard in search models the matching rate λ is the

probability of the landlord making contact with a tenant which depends on the relative

number of potential renters to vacant units With r as the discount rate and πi where

iεO V denoting the present value of future net income the flow value function of the

landlord in the vacancy state is given by the Bellman equation

rπV = minusc+ λ(EπO minus πV ) (1)

Once the unit becomes occupied the tenantrsquos quality is revealed as either good or bad

If the tenant is bad the landlord must pay y in extra maintenance costs each period while

3We assume that tenants do not gain from this act and they pay their rent regardless This eliminatesstrategic behavior on the part of the tenants We discuss the implication of relaxing this assumption inSection 25

5

a good tenant imposes no such costs We assume that no eviction is attempted in the case

of good tenant and the flow value of utility of the landlord is then

rπO(y good) = R + δ(πV minus πO(y good)) (2)

where δ is the exogenous probability that the tenant and landlord become mismatched and

separate at any point in time Note that Equation (2) becomes

πO(y good) =R + δπVr + δ

(3)

If instead the tenant is revealed as bad the landlord may choose to evict them or not

The period utility when no eviction is chosen equals

rπO(y keep) = (Rminus y) + δ(πV minus πO(y keep)) (4)

Alternatively the landlord may try to evict a bad tenant by paying a per period fee of d

thereby raising the separation rate by an exogenous amount ε

rπO(y evict) = (Rminus y minus d) + (δ + ε)(πV minus πO(y evict)) (5)

Rewriting Equation (4) and Equation (5) respectively yields

πO(y keep) =Rminus y + δπV

r + δ (6)

and

πO(y evict) =Rminus y minus d+ (δ + ε)πV

r + δ + ε (7)

Eviction will occur when the present value of profits from eviction is greater than that

of keeping the bad tenant Combining Equation (6) and Equation (7) this condition can be

6

written as

Rminus y lt rπV minusd(r + δ)

ε (8)

The left hand side is net rent The right hand side is a parameter cluster that represents

the net flow benefit of eviction (not including the lost rent on the left hand side) This net

benefit is higher when the probability of successful eviction is higher or when the value of a

vacancy is higher and is lower when the cost of eviction is higher or when the detachment

rate is higher In the latter case a higher probability that the (bad) tenant will leave anyway

lowers the value of deliberately evicting them

Note that the eviction rate ε when it is not zero is exogenous so that the comparative

static response of evictions to changes in d is limited to the case where d rises by enough to

flip the inequality of (8) Nevertheless the model delivers the expected result that a rise in

the cost of eviction can lower the number of evictions In order to obtain other comparative

static impacts of eviction cost d it is now necessary to assume that Equation (8) holds so

that landlords always choose to evict bad tenants Using p as the share of bad tenants in

the market the expected gains for landlord from matching with a tenant then becomes

EπO minus πV = (1minus p) R + δπVr + δ

+ pRminus y minus d+ (δ + ε) πV

r + δ + εminus πV (9)

=(Rminus rπV ) (r + δ + (1minus p) ε)minus p(d+ y)(r + δ)

(r + δ)(r + δ + ε) (10)

We can now replace rπV with its value from Equation (1) and rearrange Equation (10)

to get

EπO minus πV =(R + c) (r + δ + ε (1minus p))minus p(d+ y)(r + δ)

(r + δ) (r + δ + ε) + λ((r + δ + ε (1minus p))(11)

7

=(R + c) θ2 minus θ3

θ1 + θ2λ (12)

where θ1 = (r + δ) (r + δ + ε) θ2 = r + δ + ε (1minus p) and θ3(d) = p(d+ y)(r + δ)

22 Tenants

We now turn to the derivation of the tenantrsquos utility functions in their respective states

housed (H) and unhoused (U) Recalling that their draw of y is unknown to them before

they are housed we let J described their lifetime utility from any given state and write the

flow value of being unhoused

rJU = micro(EJH minus JU) (13)

where cash flow is normalized to zero and micro is the (endogenous) matching rate of tenants

to landlords It is the rate at which tenants make contact with a landlord which is again

a function of the ratio of unhoused tenants to vacant units in the market Once they are

housed their draw of y is revealed and they observe whether the landlord is trying to evict

them With Z notating the benefit of being housed we have

rJH(y stay) = (Z minusR) + δ(JU minus JH(Y stay)) (14)

rJH(y evict) = (Z minusR) + (δ + ε)(JU minus JH(Y evict)) (15)

As a simplification we assume that tenants pay full rent and does not receive any benefit

if they are revealed to be bad This prevents tenants from strategically acting like bad

tenants in no-eviction markets An assumption to justify this would be that y represents

lack of care but that the tenant receives no (leisure) benefit from this lack We will return

to this assumption in Section 25 to discuss the impact of its relaxation on our model

8

To determine the expectation of lifetime utility from being housed first write using

Equation (14) and Equation (15) respectively

JH(y stay) =(Z minusR) + δJU

r + δ (16)

and

JH(y evict) =(Z minusR) + (δ + ε)JU

r + δ + ε (17)

so that as long as eviction is undertaken by landlords and remembering that the tenant

does not know their quality ahead of the match we have

EJH minus JU = p(Z minusR) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (18)

=(r + δ + (1minus p)ε)((Z minusR)minus rJU)

(r + δ)(r + δ + ε) (19)

Substitute the expression for the flow value of the unhoused to get from inserting Equation

(13) into Equation (18) and Equation (19) respectively

(EJH minus JU) =θ2(Z minusR)

θ1 + θ2micro (20)

23 Rents

Rents are determined in a Nash Bargain which (by assumption of equal bargaining power)

equates the gains from agreement obtained by landlord and tenant

(R + c) θ2 minus θ3θ1 + θ2λ

=θ2(Z minusR)

θ1 + θ2micro (21)

9

which yields rent as a function of the two endogenous contact rates

R =(θ3 minus cθ2)(θ1 + θ2micro) + Z(θ1 + θ2λ)θ2

θ2(2θ1 + θ2 (λ+ micro))= R(λ micro) (22)

where the derivative of R with respect to λ is positive and with respect to micro is negative

under the condition that (θ3 minus cθ2) lt 0 A sufficient condition for this is p (d+ y) lt c

which from Equation (12) is easily seen to be a sufficient condition for landlord entry into

the market These are standard results from search and bargaining models since an increase

in onersquos contact rate raises onersquos bargaining power and tilts the rent in a favorable direction

We assume free entry such that the profits from holding a vacancy equals the fixed costs

of entering the market

πV = ϕ (23)

From (1) (9) (22) and (23) we get

rϕ+ c

λ=

(R(λ micro) + c) θ2 minus θ3(d)

θ1 + θ2λ(24)

which defines an equilibrium entry zero-profit condition (ZP ) in the two contact rates In

the (micro λ)-space it is straightforward to show that the implicit function derived from Equation

(24) is upward sloping An increase in the tenant contact rate necessitates an increase in the

landlord contact rate (via landlord exit as discussed below) in order to return to zero profit

There are L landlords whose units are vacant or occupied and T tenants housed or

unhoused measured on a continuum

L = LV + LO (25)

T = TH + TU (26)

10

Note that T is assumed to be exogenous in order to anchor the model but L is determined

endogenously as the number of available units must satisfy a zero profit condition The

number of matches in any given period is generated by a standard matching function

λLV = microTU = M (LV TU) (27)

The matching function M is assumed to be constant returns to scale so that

λ = M(1 λmicro) (28)

where λmicro is ldquomarket tightnessrdquo It is straightforward to see that Equation (28) implicitly

defines a downward sloping function in the (micro λ) space which we refer to as the matching

condition

Given certain mild conditions on the matching function (Coulson et al 2001) there exists

a unique solution to the zero-profit condition and the matching condition that establishes

the equilibrium values of the two contact rates The comparative statics of the model are

intuitive to establish (Figure 1) A change in any of the modelrsquos parameters implies a shift

in the zero-profit function that re-equilibrates micro and λ

Our main interest is in the comparative static responses to changes in d the cost of

eviction For ease of presentation we eschew comparative static analysis of changes in ε We

can therefore interpret d as the cost of obtaining the given eviction rate Examination of

Equation (24) indicates that an increase in d will shift up the zero-profit line For any given

micro a higher d requires an increase in λ to maintain zero profits As Figure 1 therefore shows

a rise in d implies both higher λ and lower micro This in turn implies from Equation (22) a

higher equilibrium rent

11

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

a good tenant imposes no such costs We assume that no eviction is attempted in the case

of good tenant and the flow value of utility of the landlord is then

rπO(y good) = R + δ(πV minus πO(y good)) (2)

where δ is the exogenous probability that the tenant and landlord become mismatched and

separate at any point in time Note that Equation (2) becomes

πO(y good) =R + δπVr + δ

(3)

If instead the tenant is revealed as bad the landlord may choose to evict them or not

The period utility when no eviction is chosen equals

rπO(y keep) = (Rminus y) + δ(πV minus πO(y keep)) (4)

Alternatively the landlord may try to evict a bad tenant by paying a per period fee of d

thereby raising the separation rate by an exogenous amount ε

rπO(y evict) = (Rminus y minus d) + (δ + ε)(πV minus πO(y evict)) (5)

Rewriting Equation (4) and Equation (5) respectively yields

πO(y keep) =Rminus y + δπV

r + δ (6)

and

πO(y evict) =Rminus y minus d+ (δ + ε)πV

r + δ + ε (7)

Eviction will occur when the present value of profits from eviction is greater than that

of keeping the bad tenant Combining Equation (6) and Equation (7) this condition can be

6

written as

Rminus y lt rπV minusd(r + δ)

ε (8)

The left hand side is net rent The right hand side is a parameter cluster that represents

the net flow benefit of eviction (not including the lost rent on the left hand side) This net

benefit is higher when the probability of successful eviction is higher or when the value of a

vacancy is higher and is lower when the cost of eviction is higher or when the detachment

rate is higher In the latter case a higher probability that the (bad) tenant will leave anyway

lowers the value of deliberately evicting them

Note that the eviction rate ε when it is not zero is exogenous so that the comparative

static response of evictions to changes in d is limited to the case where d rises by enough to

flip the inequality of (8) Nevertheless the model delivers the expected result that a rise in

the cost of eviction can lower the number of evictions In order to obtain other comparative

static impacts of eviction cost d it is now necessary to assume that Equation (8) holds so

that landlords always choose to evict bad tenants Using p as the share of bad tenants in

the market the expected gains for landlord from matching with a tenant then becomes

EπO minus πV = (1minus p) R + δπVr + δ

+ pRminus y minus d+ (δ + ε) πV

r + δ + εminus πV (9)

=(Rminus rπV ) (r + δ + (1minus p) ε)minus p(d+ y)(r + δ)

(r + δ)(r + δ + ε) (10)

We can now replace rπV with its value from Equation (1) and rearrange Equation (10)

to get

EπO minus πV =(R + c) (r + δ + ε (1minus p))minus p(d+ y)(r + δ)

(r + δ) (r + δ + ε) + λ((r + δ + ε (1minus p))(11)

7

=(R + c) θ2 minus θ3

θ1 + θ2λ (12)

where θ1 = (r + δ) (r + δ + ε) θ2 = r + δ + ε (1minus p) and θ3(d) = p(d+ y)(r + δ)

22 Tenants

We now turn to the derivation of the tenantrsquos utility functions in their respective states

housed (H) and unhoused (U) Recalling that their draw of y is unknown to them before

they are housed we let J described their lifetime utility from any given state and write the

flow value of being unhoused

rJU = micro(EJH minus JU) (13)

where cash flow is normalized to zero and micro is the (endogenous) matching rate of tenants

to landlords It is the rate at which tenants make contact with a landlord which is again

a function of the ratio of unhoused tenants to vacant units in the market Once they are

housed their draw of y is revealed and they observe whether the landlord is trying to evict

them With Z notating the benefit of being housed we have

rJH(y stay) = (Z minusR) + δ(JU minus JH(Y stay)) (14)

rJH(y evict) = (Z minusR) + (δ + ε)(JU minus JH(Y evict)) (15)

As a simplification we assume that tenants pay full rent and does not receive any benefit

if they are revealed to be bad This prevents tenants from strategically acting like bad

tenants in no-eviction markets An assumption to justify this would be that y represents

lack of care but that the tenant receives no (leisure) benefit from this lack We will return

to this assumption in Section 25 to discuss the impact of its relaxation on our model

8

To determine the expectation of lifetime utility from being housed first write using

Equation (14) and Equation (15) respectively

JH(y stay) =(Z minusR) + δJU

r + δ (16)

and

JH(y evict) =(Z minusR) + (δ + ε)JU

r + δ + ε (17)

so that as long as eviction is undertaken by landlords and remembering that the tenant

does not know their quality ahead of the match we have

EJH minus JU = p(Z minusR) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (18)

=(r + δ + (1minus p)ε)((Z minusR)minus rJU)

(r + δ)(r + δ + ε) (19)

Substitute the expression for the flow value of the unhoused to get from inserting Equation

(13) into Equation (18) and Equation (19) respectively

(EJH minus JU) =θ2(Z minusR)

θ1 + θ2micro (20)

23 Rents

Rents are determined in a Nash Bargain which (by assumption of equal bargaining power)

equates the gains from agreement obtained by landlord and tenant

(R + c) θ2 minus θ3θ1 + θ2λ

=θ2(Z minusR)

θ1 + θ2micro (21)

9

which yields rent as a function of the two endogenous contact rates

R =(θ3 minus cθ2)(θ1 + θ2micro) + Z(θ1 + θ2λ)θ2

θ2(2θ1 + θ2 (λ+ micro))= R(λ micro) (22)

where the derivative of R with respect to λ is positive and with respect to micro is negative

under the condition that (θ3 minus cθ2) lt 0 A sufficient condition for this is p (d+ y) lt c

which from Equation (12) is easily seen to be a sufficient condition for landlord entry into

the market These are standard results from search and bargaining models since an increase

in onersquos contact rate raises onersquos bargaining power and tilts the rent in a favorable direction

We assume free entry such that the profits from holding a vacancy equals the fixed costs

of entering the market

πV = ϕ (23)

From (1) (9) (22) and (23) we get

rϕ+ c

λ=

(R(λ micro) + c) θ2 minus θ3(d)

θ1 + θ2λ(24)

which defines an equilibrium entry zero-profit condition (ZP ) in the two contact rates In

the (micro λ)-space it is straightforward to show that the implicit function derived from Equation

(24) is upward sloping An increase in the tenant contact rate necessitates an increase in the

landlord contact rate (via landlord exit as discussed below) in order to return to zero profit

There are L landlords whose units are vacant or occupied and T tenants housed or

unhoused measured on a continuum

L = LV + LO (25)

T = TH + TU (26)

10

Note that T is assumed to be exogenous in order to anchor the model but L is determined

endogenously as the number of available units must satisfy a zero profit condition The

number of matches in any given period is generated by a standard matching function

λLV = microTU = M (LV TU) (27)

The matching function M is assumed to be constant returns to scale so that

λ = M(1 λmicro) (28)

where λmicro is ldquomarket tightnessrdquo It is straightforward to see that Equation (28) implicitly

defines a downward sloping function in the (micro λ) space which we refer to as the matching

condition

Given certain mild conditions on the matching function (Coulson et al 2001) there exists

a unique solution to the zero-profit condition and the matching condition that establishes

the equilibrium values of the two contact rates The comparative statics of the model are

intuitive to establish (Figure 1) A change in any of the modelrsquos parameters implies a shift

in the zero-profit function that re-equilibrates micro and λ

Our main interest is in the comparative static responses to changes in d the cost of

eviction For ease of presentation we eschew comparative static analysis of changes in ε We

can therefore interpret d as the cost of obtaining the given eviction rate Examination of

Equation (24) indicates that an increase in d will shift up the zero-profit line For any given

micro a higher d requires an increase in λ to maintain zero profits As Figure 1 therefore shows

a rise in d implies both higher λ and lower micro This in turn implies from Equation (22) a

higher equilibrium rent

11

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

written as

Rminus y lt rπV minusd(r + δ)

ε (8)

The left hand side is net rent The right hand side is a parameter cluster that represents

the net flow benefit of eviction (not including the lost rent on the left hand side) This net

benefit is higher when the probability of successful eviction is higher or when the value of a

vacancy is higher and is lower when the cost of eviction is higher or when the detachment

rate is higher In the latter case a higher probability that the (bad) tenant will leave anyway

lowers the value of deliberately evicting them

Note that the eviction rate ε when it is not zero is exogenous so that the comparative

static response of evictions to changes in d is limited to the case where d rises by enough to

flip the inequality of (8) Nevertheless the model delivers the expected result that a rise in

the cost of eviction can lower the number of evictions In order to obtain other comparative

static impacts of eviction cost d it is now necessary to assume that Equation (8) holds so

that landlords always choose to evict bad tenants Using p as the share of bad tenants in

the market the expected gains for landlord from matching with a tenant then becomes

EπO minus πV = (1minus p) R + δπVr + δ

+ pRminus y minus d+ (δ + ε) πV

r + δ + εminus πV (9)

=(Rminus rπV ) (r + δ + (1minus p) ε)minus p(d+ y)(r + δ)

(r + δ)(r + δ + ε) (10)

We can now replace rπV with its value from Equation (1) and rearrange Equation (10)

to get

EπO minus πV =(R + c) (r + δ + ε (1minus p))minus p(d+ y)(r + δ)

(r + δ) (r + δ + ε) + λ((r + δ + ε (1minus p))(11)

7

=(R + c) θ2 minus θ3

θ1 + θ2λ (12)

where θ1 = (r + δ) (r + δ + ε) θ2 = r + δ + ε (1minus p) and θ3(d) = p(d+ y)(r + δ)

22 Tenants

We now turn to the derivation of the tenantrsquos utility functions in their respective states

housed (H) and unhoused (U) Recalling that their draw of y is unknown to them before

they are housed we let J described their lifetime utility from any given state and write the

flow value of being unhoused

rJU = micro(EJH minus JU) (13)

where cash flow is normalized to zero and micro is the (endogenous) matching rate of tenants

to landlords It is the rate at which tenants make contact with a landlord which is again

a function of the ratio of unhoused tenants to vacant units in the market Once they are

housed their draw of y is revealed and they observe whether the landlord is trying to evict

them With Z notating the benefit of being housed we have

rJH(y stay) = (Z minusR) + δ(JU minus JH(Y stay)) (14)

rJH(y evict) = (Z minusR) + (δ + ε)(JU minus JH(Y evict)) (15)

As a simplification we assume that tenants pay full rent and does not receive any benefit

if they are revealed to be bad This prevents tenants from strategically acting like bad

tenants in no-eviction markets An assumption to justify this would be that y represents

lack of care but that the tenant receives no (leisure) benefit from this lack We will return

to this assumption in Section 25 to discuss the impact of its relaxation on our model

8

To determine the expectation of lifetime utility from being housed first write using

Equation (14) and Equation (15) respectively

JH(y stay) =(Z minusR) + δJU

r + δ (16)

and

JH(y evict) =(Z minusR) + (δ + ε)JU

r + δ + ε (17)

so that as long as eviction is undertaken by landlords and remembering that the tenant

does not know their quality ahead of the match we have

EJH minus JU = p(Z minusR) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (18)

=(r + δ + (1minus p)ε)((Z minusR)minus rJU)

(r + δ)(r + δ + ε) (19)

Substitute the expression for the flow value of the unhoused to get from inserting Equation

(13) into Equation (18) and Equation (19) respectively

(EJH minus JU) =θ2(Z minusR)

θ1 + θ2micro (20)

23 Rents

Rents are determined in a Nash Bargain which (by assumption of equal bargaining power)

equates the gains from agreement obtained by landlord and tenant

(R + c) θ2 minus θ3θ1 + θ2λ

=θ2(Z minusR)

θ1 + θ2micro (21)

9

which yields rent as a function of the two endogenous contact rates

R =(θ3 minus cθ2)(θ1 + θ2micro) + Z(θ1 + θ2λ)θ2

θ2(2θ1 + θ2 (λ+ micro))= R(λ micro) (22)

where the derivative of R with respect to λ is positive and with respect to micro is negative

under the condition that (θ3 minus cθ2) lt 0 A sufficient condition for this is p (d+ y) lt c

which from Equation (12) is easily seen to be a sufficient condition for landlord entry into

the market These are standard results from search and bargaining models since an increase

in onersquos contact rate raises onersquos bargaining power and tilts the rent in a favorable direction

We assume free entry such that the profits from holding a vacancy equals the fixed costs

of entering the market

πV = ϕ (23)

From (1) (9) (22) and (23) we get

rϕ+ c

λ=

(R(λ micro) + c) θ2 minus θ3(d)

θ1 + θ2λ(24)

which defines an equilibrium entry zero-profit condition (ZP ) in the two contact rates In

the (micro λ)-space it is straightforward to show that the implicit function derived from Equation

(24) is upward sloping An increase in the tenant contact rate necessitates an increase in the

landlord contact rate (via landlord exit as discussed below) in order to return to zero profit

There are L landlords whose units are vacant or occupied and T tenants housed or

unhoused measured on a continuum

L = LV + LO (25)

T = TH + TU (26)

10

Note that T is assumed to be exogenous in order to anchor the model but L is determined

endogenously as the number of available units must satisfy a zero profit condition The

number of matches in any given period is generated by a standard matching function

λLV = microTU = M (LV TU) (27)

The matching function M is assumed to be constant returns to scale so that

λ = M(1 λmicro) (28)

where λmicro is ldquomarket tightnessrdquo It is straightforward to see that Equation (28) implicitly

defines a downward sloping function in the (micro λ) space which we refer to as the matching

condition

Given certain mild conditions on the matching function (Coulson et al 2001) there exists

a unique solution to the zero-profit condition and the matching condition that establishes

the equilibrium values of the two contact rates The comparative statics of the model are

intuitive to establish (Figure 1) A change in any of the modelrsquos parameters implies a shift

in the zero-profit function that re-equilibrates micro and λ

Our main interest is in the comparative static responses to changes in d the cost of

eviction For ease of presentation we eschew comparative static analysis of changes in ε We

can therefore interpret d as the cost of obtaining the given eviction rate Examination of

Equation (24) indicates that an increase in d will shift up the zero-profit line For any given

micro a higher d requires an increase in λ to maintain zero profits As Figure 1 therefore shows

a rise in d implies both higher λ and lower micro This in turn implies from Equation (22) a

higher equilibrium rent

11

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

=(R + c) θ2 minus θ3

θ1 + θ2λ (12)

where θ1 = (r + δ) (r + δ + ε) θ2 = r + δ + ε (1minus p) and θ3(d) = p(d+ y)(r + δ)

22 Tenants

We now turn to the derivation of the tenantrsquos utility functions in their respective states

housed (H) and unhoused (U) Recalling that their draw of y is unknown to them before

they are housed we let J described their lifetime utility from any given state and write the

flow value of being unhoused

rJU = micro(EJH minus JU) (13)

where cash flow is normalized to zero and micro is the (endogenous) matching rate of tenants

to landlords It is the rate at which tenants make contact with a landlord which is again

a function of the ratio of unhoused tenants to vacant units in the market Once they are

housed their draw of y is revealed and they observe whether the landlord is trying to evict

them With Z notating the benefit of being housed we have

rJH(y stay) = (Z minusR) + δ(JU minus JH(Y stay)) (14)

rJH(y evict) = (Z minusR) + (δ + ε)(JU minus JH(Y evict)) (15)

As a simplification we assume that tenants pay full rent and does not receive any benefit

if they are revealed to be bad This prevents tenants from strategically acting like bad

tenants in no-eviction markets An assumption to justify this would be that y represents

lack of care but that the tenant receives no (leisure) benefit from this lack We will return

to this assumption in Section 25 to discuss the impact of its relaxation on our model

8

To determine the expectation of lifetime utility from being housed first write using

Equation (14) and Equation (15) respectively

JH(y stay) =(Z minusR) + δJU

r + δ (16)

and

JH(y evict) =(Z minusR) + (δ + ε)JU

r + δ + ε (17)

so that as long as eviction is undertaken by landlords and remembering that the tenant

does not know their quality ahead of the match we have

EJH minus JU = p(Z minusR) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (18)

=(r + δ + (1minus p)ε)((Z minusR)minus rJU)

(r + δ)(r + δ + ε) (19)

Substitute the expression for the flow value of the unhoused to get from inserting Equation

(13) into Equation (18) and Equation (19) respectively

(EJH minus JU) =θ2(Z minusR)

θ1 + θ2micro (20)

23 Rents

Rents are determined in a Nash Bargain which (by assumption of equal bargaining power)

equates the gains from agreement obtained by landlord and tenant

(R + c) θ2 minus θ3θ1 + θ2λ

=θ2(Z minusR)

θ1 + θ2micro (21)

9

which yields rent as a function of the two endogenous contact rates

R =(θ3 minus cθ2)(θ1 + θ2micro) + Z(θ1 + θ2λ)θ2

θ2(2θ1 + θ2 (λ+ micro))= R(λ micro) (22)

where the derivative of R with respect to λ is positive and with respect to micro is negative

under the condition that (θ3 minus cθ2) lt 0 A sufficient condition for this is p (d+ y) lt c

which from Equation (12) is easily seen to be a sufficient condition for landlord entry into

the market These are standard results from search and bargaining models since an increase

in onersquos contact rate raises onersquos bargaining power and tilts the rent in a favorable direction

We assume free entry such that the profits from holding a vacancy equals the fixed costs

of entering the market

πV = ϕ (23)

From (1) (9) (22) and (23) we get

rϕ+ c

λ=

(R(λ micro) + c) θ2 minus θ3(d)

θ1 + θ2λ(24)

which defines an equilibrium entry zero-profit condition (ZP ) in the two contact rates In

the (micro λ)-space it is straightforward to show that the implicit function derived from Equation

(24) is upward sloping An increase in the tenant contact rate necessitates an increase in the

landlord contact rate (via landlord exit as discussed below) in order to return to zero profit

There are L landlords whose units are vacant or occupied and T tenants housed or

unhoused measured on a continuum

L = LV + LO (25)

T = TH + TU (26)

10

Note that T is assumed to be exogenous in order to anchor the model but L is determined

endogenously as the number of available units must satisfy a zero profit condition The

number of matches in any given period is generated by a standard matching function

λLV = microTU = M (LV TU) (27)

The matching function M is assumed to be constant returns to scale so that

λ = M(1 λmicro) (28)

where λmicro is ldquomarket tightnessrdquo It is straightforward to see that Equation (28) implicitly

defines a downward sloping function in the (micro λ) space which we refer to as the matching

condition

Given certain mild conditions on the matching function (Coulson et al 2001) there exists

a unique solution to the zero-profit condition and the matching condition that establishes

the equilibrium values of the two contact rates The comparative statics of the model are

intuitive to establish (Figure 1) A change in any of the modelrsquos parameters implies a shift

in the zero-profit function that re-equilibrates micro and λ

Our main interest is in the comparative static responses to changes in d the cost of

eviction For ease of presentation we eschew comparative static analysis of changes in ε We

can therefore interpret d as the cost of obtaining the given eviction rate Examination of

Equation (24) indicates that an increase in d will shift up the zero-profit line For any given

micro a higher d requires an increase in λ to maintain zero profits As Figure 1 therefore shows

a rise in d implies both higher λ and lower micro This in turn implies from Equation (22) a

higher equilibrium rent

11

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

To determine the expectation of lifetime utility from being housed first write using

Equation (14) and Equation (15) respectively

JH(y stay) =(Z minusR) + δJU

r + δ (16)

and

JH(y evict) =(Z minusR) + (δ + ε)JU

r + δ + ε (17)

so that as long as eviction is undertaken by landlords and remembering that the tenant

does not know their quality ahead of the match we have

EJH minus JU = p(Z minusR) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (18)

=(r + δ + (1minus p)ε)((Z minusR)minus rJU)

(r + δ)(r + δ + ε) (19)

Substitute the expression for the flow value of the unhoused to get from inserting Equation

(13) into Equation (18) and Equation (19) respectively

(EJH minus JU) =θ2(Z minusR)

θ1 + θ2micro (20)

23 Rents

Rents are determined in a Nash Bargain which (by assumption of equal bargaining power)

equates the gains from agreement obtained by landlord and tenant

(R + c) θ2 minus θ3θ1 + θ2λ

=θ2(Z minusR)

θ1 + θ2micro (21)

9

which yields rent as a function of the two endogenous contact rates

R =(θ3 minus cθ2)(θ1 + θ2micro) + Z(θ1 + θ2λ)θ2

θ2(2θ1 + θ2 (λ+ micro))= R(λ micro) (22)

where the derivative of R with respect to λ is positive and with respect to micro is negative

under the condition that (θ3 minus cθ2) lt 0 A sufficient condition for this is p (d+ y) lt c

which from Equation (12) is easily seen to be a sufficient condition for landlord entry into

the market These are standard results from search and bargaining models since an increase

in onersquos contact rate raises onersquos bargaining power and tilts the rent in a favorable direction

We assume free entry such that the profits from holding a vacancy equals the fixed costs

of entering the market

πV = ϕ (23)

From (1) (9) (22) and (23) we get

rϕ+ c

λ=

(R(λ micro) + c) θ2 minus θ3(d)

θ1 + θ2λ(24)

which defines an equilibrium entry zero-profit condition (ZP ) in the two contact rates In

the (micro λ)-space it is straightforward to show that the implicit function derived from Equation

(24) is upward sloping An increase in the tenant contact rate necessitates an increase in the

landlord contact rate (via landlord exit as discussed below) in order to return to zero profit

There are L landlords whose units are vacant or occupied and T tenants housed or

unhoused measured on a continuum

L = LV + LO (25)

T = TH + TU (26)

10

Note that T is assumed to be exogenous in order to anchor the model but L is determined

endogenously as the number of available units must satisfy a zero profit condition The

number of matches in any given period is generated by a standard matching function

λLV = microTU = M (LV TU) (27)

The matching function M is assumed to be constant returns to scale so that

λ = M(1 λmicro) (28)

where λmicro is ldquomarket tightnessrdquo It is straightforward to see that Equation (28) implicitly

defines a downward sloping function in the (micro λ) space which we refer to as the matching

condition

Given certain mild conditions on the matching function (Coulson et al 2001) there exists

a unique solution to the zero-profit condition and the matching condition that establishes

the equilibrium values of the two contact rates The comparative statics of the model are

intuitive to establish (Figure 1) A change in any of the modelrsquos parameters implies a shift

in the zero-profit function that re-equilibrates micro and λ

Our main interest is in the comparative static responses to changes in d the cost of

eviction For ease of presentation we eschew comparative static analysis of changes in ε We

can therefore interpret d as the cost of obtaining the given eviction rate Examination of

Equation (24) indicates that an increase in d will shift up the zero-profit line For any given

micro a higher d requires an increase in λ to maintain zero profits As Figure 1 therefore shows

a rise in d implies both higher λ and lower micro This in turn implies from Equation (22) a

higher equilibrium rent

11

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

which yields rent as a function of the two endogenous contact rates

R =(θ3 minus cθ2)(θ1 + θ2micro) + Z(θ1 + θ2λ)θ2

θ2(2θ1 + θ2 (λ+ micro))= R(λ micro) (22)

where the derivative of R with respect to λ is positive and with respect to micro is negative

under the condition that (θ3 minus cθ2) lt 0 A sufficient condition for this is p (d+ y) lt c

which from Equation (12) is easily seen to be a sufficient condition for landlord entry into

the market These are standard results from search and bargaining models since an increase

in onersquos contact rate raises onersquos bargaining power and tilts the rent in a favorable direction

We assume free entry such that the profits from holding a vacancy equals the fixed costs

of entering the market

πV = ϕ (23)

From (1) (9) (22) and (23) we get

rϕ+ c

λ=

(R(λ micro) + c) θ2 minus θ3(d)

θ1 + θ2λ(24)

which defines an equilibrium entry zero-profit condition (ZP ) in the two contact rates In

the (micro λ)-space it is straightforward to show that the implicit function derived from Equation

(24) is upward sloping An increase in the tenant contact rate necessitates an increase in the

landlord contact rate (via landlord exit as discussed below) in order to return to zero profit

There are L landlords whose units are vacant or occupied and T tenants housed or

unhoused measured on a continuum

L = LV + LO (25)

T = TH + TU (26)

10

Note that T is assumed to be exogenous in order to anchor the model but L is determined

endogenously as the number of available units must satisfy a zero profit condition The

number of matches in any given period is generated by a standard matching function

λLV = microTU = M (LV TU) (27)

The matching function M is assumed to be constant returns to scale so that

λ = M(1 λmicro) (28)

where λmicro is ldquomarket tightnessrdquo It is straightforward to see that Equation (28) implicitly

defines a downward sloping function in the (micro λ) space which we refer to as the matching

condition

Given certain mild conditions on the matching function (Coulson et al 2001) there exists

a unique solution to the zero-profit condition and the matching condition that establishes

the equilibrium values of the two contact rates The comparative statics of the model are

intuitive to establish (Figure 1) A change in any of the modelrsquos parameters implies a shift

in the zero-profit function that re-equilibrates micro and λ

Our main interest is in the comparative static responses to changes in d the cost of

eviction For ease of presentation we eschew comparative static analysis of changes in ε We

can therefore interpret d as the cost of obtaining the given eviction rate Examination of

Equation (24) indicates that an increase in d will shift up the zero-profit line For any given

micro a higher d requires an increase in λ to maintain zero profits As Figure 1 therefore shows

a rise in d implies both higher λ and lower micro This in turn implies from Equation (22) a

higher equilibrium rent

11

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

Note that T is assumed to be exogenous in order to anchor the model but L is determined

endogenously as the number of available units must satisfy a zero profit condition The

number of matches in any given period is generated by a standard matching function

λLV = microTU = M (LV TU) (27)

The matching function M is assumed to be constant returns to scale so that

λ = M(1 λmicro) (28)

where λmicro is ldquomarket tightnessrdquo It is straightforward to see that Equation (28) implicitly

defines a downward sloping function in the (micro λ) space which we refer to as the matching

condition

Given certain mild conditions on the matching function (Coulson et al 2001) there exists

a unique solution to the zero-profit condition and the matching condition that establishes

the equilibrium values of the two contact rates The comparative statics of the model are

intuitive to establish (Figure 1) A change in any of the modelrsquos parameters implies a shift

in the zero-profit function that re-equilibrates micro and λ

Our main interest is in the comparative static responses to changes in d the cost of

eviction For ease of presentation we eschew comparative static analysis of changes in ε We

can therefore interpret d as the cost of obtaining the given eviction rate Examination of

Equation (24) indicates that an increase in d will shift up the zero-profit line For any given

micro a higher d requires an increase in λ to maintain zero profits As Figure 1 therefore shows

a rise in d implies both higher λ and lower micro This in turn implies from Equation (22) a

higher equilibrium rent

11

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

24 Quantities in the Rental Market

In order to say more about quantities we can invoke the assumption of a steady state A

steady state equilibrium requires that the number of tenants that enter and exit the housed

and unhoused states is equal

(δ + pε)TH = microTU (29)

Similarly the number of landlords that enter and exit the vacant and occupied states must

be equal

(δ + pε)LO = λLV (30)

Housing supply is the sum of occupied and vacant units

Housing supply = LO + LV (31)

=

(1 +

λ

δ + pε

)LV (32)

Using (27)

=

(1 +

λ

δ + pε

)(micro

λ)TU (33)

and (29)

=

(δ + pε+ λ

δ + pε+ micro

)(microλ

)T (34)

Since the model is anchored to an exogenous population T it is clear from Equation

(34) that equilibrium housing supply is lower with a higher landlord contact rate and higher

with a higher tenant contact rate Therefore a higher cost of eviction decreases the supply

12

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

of rental housing

Next we define the homeless rate as the percentage of tenants in the unhoused state

Homeless rate =TU

TU + TH (35)

and from (29) we have that this is

Homeless rate =δ + pε

δ + pε+ micro (36)

so that an increase in the cost of eviction lowers the tenant contact rate and increases

homelessness This appears paradoxical but works through the eviction costrsquos effect on

housing supply as shown above

We define the vacancy rate for landlords as the percentage of units that are vacant and

use Equation (30) to find

V acancy rate =LV

LV + LO

=δ + pε

δ + pε+ λ (37)

so that an increase in d raises the landlord contact rate and lowers the vacancy rate again

through the effect of housing supply

25 Nonpayment

As noted earlier in this section we view y as merely a cost to the landlord without these

extra costs bestowing any benefit to the tenant This eliminates any strategic behavior on

the part of tenants One of the common reasons for eviction of course is nonpayment of

rent Nonpayment does not necessarily mean that the tenant never pays they might be late

or may be able to pay some months and not others A wide variety of behaviors are available

(Kim 2018) In any case we can treat y not as extra maintenance costs but (average) rent

nonpayment over the course of the residence spell This might not entail strategic behavior

by the tenant Just as there is a distinction between ruthless default and default that is really

13

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

not under the control of the homeowners (Gerardi et al 2018 and Cunningham et al 2020)

there can be a distinction between forced nonpayment of rent due to sudden unemployment

and more strategic conduct This distinction gained considerable salience at the emergence of

the COVID-19 pandemic when widespread and sudden unemployment led to led to equally

widespread calls for rent forgiveness and subsequent legislation preventing eviction

Under this view that nonpayment is ldquoforcedrdquo almost nothing in the model changes The

cost parameter y becomes the (average) reduction in rent and the probability p is now

the probability that the tenant loses their job Tenants do see a reduction in rent paid but

under this interpretation of nonpayment it is hard to see this as an improvement in utility

for them One might view Z as income and Z minus R is non-housing consumption under full

rental payment In this case Z minus (R minus y) is under partial rental payment With sudden

unemployment we might conveniently assume that Z and Rminus y are coincidentally reduced

together dollar for dollar Then nothing in the model changes and all of our results go

through

On the other hand under strategic behavior by the tenant the tenant considers their

options under fully payment or partial payment (which can include zero payment) While

Equation (14) remains unchanged Equation (15) becomes

rJH(y evict) = (Z minus (Rminus y)) + (δ + ε)(JU minus JH(Y evict)) (38)

The tenant then considers the relative merits of fully and partially paying rent Assuming

as before that Equation (8) holds he compares lifetime utility when the landlord tries to

evict and when she does not Full payment of rent (and no eviction) is optimal when

Z minusR gt rJU +y(r + δ)

ε (39)

That is the tenant will pay full rent when net value from renting is high enough when the

value of being homeless is low when the reduction in rent is low and the probability of

14

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

eviction is high However it is easy to see at this point that the model either degenerates

or becomes intractable If Equation (30) holds then p = 0 and landlords never evict If

the reverse inequality holds then no tenant pays (full) rent and landlords evict everyone

A modeling alternative would be to introduce heterogeneity by specifying a distribution of

y the draw of which is revealed after the lease is signed If this is designed as a two-point

distribution where one of the values satisfies Equation (30) (call it ylowast) and one does not (y)

the first part of Equation (19) becomes

EJH minus JU = p(Z minus (Rminus y)) + (δ + ε)JU

r + δ + ε+ (1minus p)(Z minusR) + δJU

r + δminus JU (40)

which would induce more complex versions of equation (24) without adding much in the way

of additional insight For this reason we choose to eschew this alternative model

26 Model Predictions

To summarize our model predicts that an increase in eviction cost d has the following effects

1 An increase in rents

2 A decrease in housing supply

3 A lower vacancy rate

4 A higher homeless rate

5 A potentially lower eviction rate

We now turn to the empirical testing of these hypotheses using the Tenant-Right Index

described in the next section

3 Tenant-Right Index

Central to our theory is the idea that it is costly for landlords to evict unwanted tenants In

the most narrow sense we can measure this cost as the various monetary expenses landlords

have to incur in the process At the minimum this often includes legal fees constable fees

15

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

and lost rent Additionally since tenants at risk of eviction most likely neglect or even

maliciously damage the properties repair and cleaning bills can be the most substantial cost

items for landlords In this paper however we define eviction cost in a more general sense to

include the strictness of statutory regulation imposed on landlords We argue that when the

laws provide stronger protections for tenants it is more difficult and costly for landlords to

deal with bad tenants Thus we use state laws to construct a proxy for the degree of tenant

protection in each state in the US which we refer to as the Tenant-Right Index It is worth

noting at this point that we do not imply that safeguarding tenant rights is undesirable In

fact there is little doubt about the importance of tenant protection in maintaining a healthy

and functioning rental market However to the extent that we are interested in eviction

costs we interpret tenant protection as a cost to landlords in this paperrsquos context

31 Index Construction

Landlord-tenant laws govern the rental of residential property in the US It is composed

primarily of state statutes that are guided by the Uniform Residential Landlord and Tenant

Act (URLTA)4 We conduct a comprehensive survey of landlord-tenant laws in each of the

50 US states and the District of Columbia from 1997 to 2016 and hand collect data on

twelve statutes that we identify as important for tenant protection5

Table 1 provides a summary of their definitions and examples We then develop a simple

binary scoring system to measure whether a particular state offers more protection to tenants

than landlords in each of these twelve aspects we assign a score of 1 if a state favors tenants

compared to the average state in the sample For each state in each year we take the sum

of all twelve categories as its Tenant-Right Index for that year Our index ranges between 0

and 12 and the higher the index value the more tenant-friendly a state is The details of

4Landlord-Tenant Law Cornell Law School httpswwwlawcornelleduwexlandlord-tenant_

law and the Uniform Law Commissionrsquos Uniform Residential Landlord-Tenant Act https

d1unatz8mcf3a5cloudfrontnetuploadsUniform-Residential-Landlord-and-Tenant-Actpdf5Although the Index is constructed from 1997 to 2016 we limit our analysis to the period from 2005 to

2016 to accommodate the other data used in our empirical analysis

16

our index calculation are as follows

Maximum Deposit This is the state rule on the maximum security deposit landlords can

collect from tenants to cover potential property damages or unpaid rent6 It ranges from

one to three monthsrsquo rent in our sample with an average of 15 months Any state with a

limit lower or equal to the average receives a score of 1 If a state does not have any statute

governing this aspect we assume that it gives landlords more discretion and assigns 0

Deposit Interest This category takes the value of 1 for states that require landlords to

pay tenants the interests due on their security deposit

Deposit Return Landlords are required to return the security deposit within a certain

time after the tenants move out7 The average deadline is 30 days but it can range from

10 days to 60 days We code this provision as 1 if a state has a deadline of 30 days or less

Again states with no statute are considered to be more landlord-friendly thus receiving no

points for this provision

Regular Termination Landlords can end a month-to-month tenancy by giving tenants a

notice typically 30 days in advance The shortest notice allowed is 3 days in Connecticut

while the longest is 60 days in Georgia and Delaware States that require at least 30 days

or more receive a score of 1

Rent Increase Notice Landlords are also required to provide advance notice in order to

increase rent in a month-to-month tenancy The amount of notice varies between 7 and 60

days This provision equals 1 if a state requires 30 days or more

Rent Withholding When landlords fail to perform proper maintenance to keep the prop-

erty habitable many states allow tenants to withhold rent payment until the problem is

fixed Generally tenants are not allowed to do so if there are no statute that explicitly

permit this action We give one point for the existence of this statute

6The limit may vary depending on various factors such as the age of the tenants whether the unit isfurnished whether the tenant has pets We use the deposit limit for the most general case of an unfurnishedapartment with no pets

7Some states have different deadlines depending on whether there are deductions made In our calculationwe use the deadline applied to the case of no deductions

17

Repair and Deduct This is similar to the provision above except that instead of with-

holding rent tenants can make the repair themselves and deduct the cost from rent

Nonpayment Termination For nonpayment of rent on average landlords need to give

tenants a 7-day notice to vacate before they can file an eviction lawsuit with the court8

Legally landlords in Alabama can start the eviction proceeding as soon as rent is due while

at the other end of the spectrum those in the District of Columbia must wait 30 days9 A

notice requirement of 7 days or more equals 1 point

Lease Violation Termination Similar to nonpayment landlords must give a proper notice

if they want the tenants to vacate due to a major lease violation which can range from 0 to

30 days We assign one point to states requiring a notice more than the average of 12 days

Self-help Eviction This provision deals with the penalty for landlords engaging in illegal

self-help eviction such as locking out tenants or utility shutoff In most cases tenants can

sue for at least the actual damages they suffer but several states allow more severe penalty

up to 3 times that amount These states are assigned a score of 1 We treat states with no

statute on this issue as leaning towards landlords rather than tenants

Right to Stay We add one point if state law gives tenants the right to remain in the

property after an illegal self-help eviction

Rent Control Preemption This category takes the value of 1 if a state does not have

legislation preventing local governments from passing rent control laws

32 Descriptive and Validation Statistics

Table 2 reports the average score over 2005-2016 for each of the twelve law provisions as

well as the total Tenant-Right Index for each state We regard the 22 states with index

8In addition to the notice requirement some states also have a statutory grace period For example inMaine landlords must wait until the rent is at least 7 days late upon which they can issue a 7-day noticeto the tenants Effectively the total wait period for landlords is 14 days We therefore use the sum of graceperiod and notice requirement in our calculation

9Note that filing a lawsuit with the court is just the beginning of the eviction process Landlords mustthen wait for the court to schedule a hearing (if the tenants do not already leave voluntarily) which maytake anywhere from a few days to several weeks or months in big cities with a huge backlog Only after theyare granted a judgment can they have law enforcement remove the tenants

18

values higher than 6 as tenant-friendly states in the sense that they offer more protection

to tenants in more than half of the aspects that we consider important There are 23 states

with index values lower than 6 which are considered landlord-friendly and the remaining

are neutral states The average index for all states over our sample period is 554 implying

that on average state laws slightly favor landlords to tenants We also map the average index

of each state in Figure 2 Consistent with general expectations states in the West and East

coasts tend to be more tenant-friendly whereas Southern states provide more protection

to landlords Rhode Island has the highest index value of 11 followed by Massachusetts

Hawaii and Vermont On the contrary Utah and Colorado are the most landlord-friendly

states with an index lower than 1 followed by Louisiana and Georgia

Figure 3 traces the average Tenant-Right Index over our study period from 2005 to 2016

The small increase observed in this figure is driven by only seven states10 each having a

one-point increase at some time during this period except for Alabama which rose by three

points North Carolina is the only state that registered a lower index value but it is merely

by one point The indexes remain the same over our sample period for the rest of the states

Overall we observe that state laws are consistent over time except for a handful that seems

to move towards slightly more tenant protections

Next we examine the twelve legal provisions used to construct the index Our goal is to

show that every one of these statutes is necessary and reflects a unique aspect of the landlord-

tenant relationship and our index does not contain redundant information We first present

their pairwise correlations in Table 3 The low correlations between them indicate that each

of the twelve legal provisions likely contributes unique information to the final index To

further test whether we reduce the number of contributors to the index we perform Principal

Component Analysis (PCA) and report each componentrsquos eigenvalues in Figure 4

Although the general rule of thumb suggests that we should focus on the first four com-

ponents whose eigenvalues are greater than 1 they explain only 61 of the variation in

10They are Alabama Colorado Georgia Illinois New Hampshire Texas and Utah

19

our Tenant-Right Index This result is not surprising given the low pairwise correlations

observed earlier It implies that we will discard a large amount of information using only

these four leading components in our index Table 4 further reports the loading of the four

components on the original twelve variables For readability we only display absolute values

higher than 03 The first component seems to load heavily on rent-related factors while the

second component correlates strongly with legislation regarding deposits and termination

notice It is less clear for the third and fourth components as they include a wide range of

provisions In conclusion our PCA analysis results validate the importance of every statute

used to construct the Tenant-Right Index as a proxy for tenant protection

4 Data and Methodology

41 Empirical Models

We test the relationship between landlord-tenant laws and various housing outcomes by

estimating the following equation

OutcomeV arct = α + βTenantRightIndexst + λDemographicsControlsct+

+θPropertyControlsct + δYt + εc

(41)

where c and t denote city and year respectively All regressions include a set of year fixed

effects Yt and standard errors are clustered at the city level except in the homeless rate

regression where it is clustered at the state level Given that the Index does not contain a

lot of time series variation for any state we eschew the use of state fixed effects We return

to the question of unobserved state-level variables below

We are interested in five outcomes as the dependent variables in Equation (41) Our first

quantity of interest is the median gross rent in city c in year t defined by the Census Bureau

as the contract rent plus the estimated average monthly cost of utilities We also verify our

results using rent data from Zillow Our independent variable of interest is the state-level

20

TenantRightIndexs that we describe in Section 3 above If more protection for tenant rights

leads to higher rent as predicted by our theoretical model earlier we will observe a positive

β coefficient The regression controls for a set of local demographics variables including

population population growth population density median household income growth of

household income and homeownership rate We also account for property characteristics

at the city level namely the median number of rooms median property age and median

property tax paid

Our second dependent variable is the rental supply For each city in each year we

calculate the number of rental units per 100 residents as the total number of housing units

for rent (both occupied and vacant) divided by population and multiplied by 100 The

control variables include population growth population density median household income

growth of household income homeownership rate and median property tax Our theory

predicts a negative β coefficient on the Tenant-Right Index variable

Next we test for the impact of landlord regulation on the vacancy rate defined as the

number of vacant units divided by the total number of rental units and multiplied by 100

The β coefficient in this case is expected to be negative We control for all the demographics

variables listed earlier in the rent regression The fourth dependent variable is the homeless

rate calculated as the number of persons in homeless shelters divided by total population

and multiplied by 100 We expect β to carry a positive sign as predicted by our theory

Again the model includes a set of control variables population growth population density

median household income growth of household income the poverty rate (the share of the

population whose household income for the past 12 months was below the poverty level)

and average winter temperature We use data at the state-year level in this regression due

to the availability of the homeless population data

Our final regression tests the hypothesis that more substantial tenant protection results

in a lower eviction rate or in other words a negative β coefficient on the Tenant-Right

Index We employ two measures for the dependent variable namely the eviction filing rate

21

and eviction rate The eviction filing rate is the ratio of the number of eviction lawsuits filed

in a city over the number of renter-occupied homes in that city This measure counts all

eviction cases filed in an area including multiple lawsuits filed against the same address in

the same year On the other hand the eviction rate is the subset of those homes that received

an eviction judgment in which renters were ordered to leave which only counts the number

of unique addresses that received eviction judgments in a year The control variables include

population population growth population density median household income growth of

household income poverty rate and the share of the minority (non-white) population There

is the possibility that landlords will more readily evict non-white tenants

42 Data and Summary Statistics

Our sample covers the period from 2005 to 2016 Unless otherwise noted our data come

from the American Housing Survey estimates by the Census Bureau Table 5 presents

their summary statistics We note that the median gross rent for an average city in our

sample is $994 per month while the average Zillow Rent Index (ZRI) obtained from Zillow

is considerably higher at $1386 This is likely due to the fact that the index is computed

from Zillowrsquos estimates of rent for the full universe of houses whether or not they are ever

available for rent On the other hand the median rent estimates by the Census Bureau are

calculated from only rental houses which are often multi-family units Since single-family

homes are on average higher-valued than rental apartments including all of them in the ZRI

likely creates an upward bias in the index For robustness we use both rent measures in our

empirical tests

As a first look at the correlation between our Tenant-Right Index and rent Figure 5

shows the (average) median gross rent for each state and the size of the circles increases with

the average Tenant-Right Index Without controlling for other confounding factors there is

a weak positive correlation coefficient (037) between the two series as our theory suggests

Two other vital measures in our empirical tests are rental supply and vacancy rate

22

As shown in Table 5 the average city in our sample has 1965 units of rental houses for

every 100 persons or about 02 units per person and a vacancy rate of 987 Our fourth

variable of interest is the homeless rate We obtain estimates of the homeless population

from the Annual Homeless Assessment Report provided by the Department of Housing and

Urban Development The data are available at the state level and cover from 2007 to 2016

Notably the District of Columbia registered the highest homeless rate during this period at

123

Turning to our eviction measures we employ a novel database recently released by the

Eviction Lab at Princeton University This is the first comprehensive national eviction

database compiled using more than 80 million formal eviction records including eviction

requests from landlords and eviction orders from judges collected from the courts The

Eviction Lab data contain all the known information on the number of evictions filed in the

United States and made publicly available by municipalities11

The average filing rate and eviction rate across all sample cities from 2005 to 2016 are

772 and 361 respectively The city with the highest eviction filing rate of 6213 is

East Orange (New Jersey) in 2006 which also has the highest supply of rental housing per

capita in our sample and the highest eviction rate of 2098 is observed in Flint (Michigan)

in 2006 Figure 6 shows the average eviction rates over our study period together with the

average Tenant-Right Index for each state Albeit weak we observe a negative correlation

coefficient between them that is consistent with our model prediction

5 Empirical Results

In this section we empirically test the model predictions detailed in Section 2 Recall that

our theoretical model predicts that with an increase in the Tenant-Right Index the median

rent and the homeless population will increase the housing supply the vacancy rate and

the eviction rate will decrease

11For more details see httpsevictionlaborgmethodsmore-questions

23

Repair and Deduct This is similar to the provision above except that instead of with-

holding rent tenants can make the repair themselves and deduct the cost from rent

Nonpayment Termination For nonpayment of rent on average landlords need to give

tenants a 7-day notice to vacate before they can file an eviction lawsuit with the court8

Legally landlords in Alabama can start the eviction proceeding as soon as rent is due while

at the other end of the spectrum those in the District of Columbia must wait 30 days9 A

notice requirement of 7 days or more equals 1 point

Lease Violation Termination Similar to nonpayment landlords must give a proper notice

if they want the tenants to vacate due to a major lease violation which can range from 0 to

30 days We assign one point to states requiring a notice more than the average of 12 days

Self-help Eviction This provision deals with the penalty for landlords engaging in illegal

self-help eviction such as locking out tenants or utility shutoff In most cases tenants can

sue for at least the actual damages they suffer but several states allow more severe penalty

up to 3 times that amount These states are assigned a score of 1 We treat states with no

statute on this issue as leaning towards landlords rather than tenants

Right to Stay We add one point if state law gives tenants the right to remain in the

property after an illegal self-help eviction

Rent Control Preemption This category takes the value of 1 if a state does not have

legislation preventing local governments from passing rent control laws

32 Descriptive and Validation Statistics

Table 2 reports the average score over 2005-2016 for each of the twelve law provisions as

well as the total Tenant-Right Index for each state We regard the 22 states with index

8In addition to the notice requirement some states also have a statutory grace period For example inMaine landlords must wait until the rent is at least 7 days late upon which they can issue a 7-day noticeto the tenants Effectively the total wait period for landlords is 14 days We therefore use the sum of graceperiod and notice requirement in our calculation

9Note that filing a lawsuit with the court is just the beginning of the eviction process Landlords mustthen wait for the court to schedule a hearing (if the tenants do not already leave voluntarily) which maytake anywhere from a few days to several weeks or months in big cities with a huge backlog Only after theyare granted a judgment can they have law enforcement remove the tenants

18

values higher than 6 as tenant-friendly states in the sense that they offer more protection

to tenants in more than half of the aspects that we consider important There are 23 states

with index values lower than 6 which are considered landlord-friendly and the remaining

are neutral states The average index for all states over our sample period is 554 implying

that on average state laws slightly favor landlords to tenants We also map the average index

of each state in Figure 2 Consistent with general expectations states in the West and East

coasts tend to be more tenant-friendly whereas Southern states provide more protection

to landlords Rhode Island has the highest index value of 11 followed by Massachusetts

Hawaii and Vermont On the contrary Utah and Colorado are the most landlord-friendly

states with an index lower than 1 followed by Louisiana and Georgia

Figure 3 traces the average Tenant-Right Index over our study period from 2005 to 2016

The small increase observed in this figure is driven by only seven states10 each having a

one-point increase at some time during this period except for Alabama which rose by three

points North Carolina is the only state that registered a lower index value but it is merely

by one point The indexes remain the same over our sample period for the rest of the states

Overall we observe that state laws are consistent over time except for a handful that seems

to move towards slightly more tenant protections

Next we examine the twelve legal provisions used to construct the index Our goal is to

show that every one of these statutes is necessary and reflects a unique aspect of the landlord-

tenant relationship and our index does not contain redundant information We first present

their pairwise correlations in Table 3 The low correlations between them indicate that each

of the twelve legal provisions likely contributes unique information to the final index To

further test whether we reduce the number of contributors to the index we perform Principal

Component Analysis (PCA) and report each componentrsquos eigenvalues in Figure 4

Although the general rule of thumb suggests that we should focus on the first four com-

ponents whose eigenvalues are greater than 1 they explain only 61 of the variation in

10They are Alabama Colorado Georgia Illinois New Hampshire Texas and Utah

19

our Tenant-Right Index This result is not surprising given the low pairwise correlations

observed earlier It implies that we will discard a large amount of information using only

these four leading components in our index Table 4 further reports the loading of the four

components on the original twelve variables For readability we only display absolute values

higher than 03 The first component seems to load heavily on rent-related factors while the

second component correlates strongly with legislation regarding deposits and termination

notice It is less clear for the third and fourth components as they include a wide range of

provisions In conclusion our PCA analysis results validate the importance of every statute

used to construct the Tenant-Right Index as a proxy for tenant protection

4 Data and Methodology

41 Empirical Models

We test the relationship between landlord-tenant laws and various housing outcomes by

estimating the following equation

OutcomeV arct = α + βTenantRightIndexst + λDemographicsControlsct+

+θPropertyControlsct + δYt + εc

(41)

where c and t denote city and year respectively All regressions include a set of year fixed

effects Yt and standard errors are clustered at the city level except in the homeless rate

regression where it is clustered at the state level Given that the Index does not contain a

lot of time series variation for any state we eschew the use of state fixed effects We return

to the question of unobserved state-level variables below

We are interested in five outcomes as the dependent variables in Equation (41) Our first

quantity of interest is the median gross rent in city c in year t defined by the Census Bureau

as the contract rent plus the estimated average monthly cost of utilities We also verify our

results using rent data from Zillow Our independent variable of interest is the state-level

20

TenantRightIndexs that we describe in Section 3 above If more protection for tenant rights

leads to higher rent as predicted by our theoretical model earlier we will observe a positive

β coefficient The regression controls for a set of local demographics variables including

population population growth population density median household income growth of

household income and homeownership rate We also account for property characteristics

at the city level namely the median number of rooms median property age and median

property tax paid

Our second dependent variable is the rental supply For each city in each year we

calculate the number of rental units per 100 residents as the total number of housing units

for rent (both occupied and vacant) divided by population and multiplied by 100 The

control variables include population growth population density median household income

growth of household income homeownership rate and median property tax Our theory

predicts a negative β coefficient on the Tenant-Right Index variable

Next we test for the impact of landlord regulation on the vacancy rate defined as the

number of vacant units divided by the total number of rental units and multiplied by 100

The β coefficient in this case is expected to be negative We control for all the demographics

variables listed earlier in the rent regression The fourth dependent variable is the homeless

rate calculated as the number of persons in homeless shelters divided by total population

and multiplied by 100 We expect β to carry a positive sign as predicted by our theory

Again the model includes a set of control variables population growth population density

median household income growth of household income the poverty rate (the share of the

population whose household income for the past 12 months was below the poverty level)

and average winter temperature We use data at the state-year level in this regression due

to the availability of the homeless population data

Our final regression tests the hypothesis that more substantial tenant protection results

in a lower eviction rate or in other words a negative β coefficient on the Tenant-Right

Index We employ two measures for the dependent variable namely the eviction filing rate

21

and eviction rate The eviction filing rate is the ratio of the number of eviction lawsuits filed

in a city over the number of renter-occupied homes in that city This measure counts all

eviction cases filed in an area including multiple lawsuits filed against the same address in

the same year On the other hand the eviction rate is the subset of those homes that received

an eviction judgment in which renters were ordered to leave which only counts the number

of unique addresses that received eviction judgments in a year The control variables include

population population growth population density median household income growth of

household income poverty rate and the share of the minority (non-white) population There

is the possibility that landlords will more readily evict non-white tenants

42 Data and Summary Statistics

Our sample covers the period from 2005 to 2016 Unless otherwise noted our data come

from the American Housing Survey estimates by the Census Bureau Table 5 presents

their summary statistics We note that the median gross rent for an average city in our

sample is $994 per month while the average Zillow Rent Index (ZRI) obtained from Zillow

is considerably higher at $1386 This is likely due to the fact that the index is computed

from Zillowrsquos estimates of rent for the full universe of houses whether or not they are ever

available for rent On the other hand the median rent estimates by the Census Bureau are

calculated from only rental houses which are often multi-family units Since single-family

homes are on average higher-valued than rental apartments including all of them in the ZRI

likely creates an upward bias in the index For robustness we use both rent measures in our

empirical tests

As a first look at the correlation between our Tenant-Right Index and rent Figure 5

shows the (average) median gross rent for each state and the size of the circles increases with

the average Tenant-Right Index Without controlling for other confounding factors there is

a weak positive correlation coefficient (037) between the two series as our theory suggests

Two other vital measures in our empirical tests are rental supply and vacancy rate

22

As shown in Table 5 the average city in our sample has 1965 units of rental houses for

every 100 persons or about 02 units per person and a vacancy rate of 987 Our fourth

variable of interest is the homeless rate We obtain estimates of the homeless population

from the Annual Homeless Assessment Report provided by the Department of Housing and

Urban Development The data are available at the state level and cover from 2007 to 2016

Notably the District of Columbia registered the highest homeless rate during this period at

123

Turning to our eviction measures we employ a novel database recently released by the

Eviction Lab at Princeton University This is the first comprehensive national eviction

database compiled using more than 80 million formal eviction records including eviction

requests from landlords and eviction orders from judges collected from the courts The

Eviction Lab data contain all the known information on the number of evictions filed in the

United States and made publicly available by municipalities11

The average filing rate and eviction rate across all sample cities from 2005 to 2016 are

772 and 361 respectively The city with the highest eviction filing rate of 6213 is

East Orange (New Jersey) in 2006 which also has the highest supply of rental housing per

capita in our sample and the highest eviction rate of 2098 is observed in Flint (Michigan)

in 2006 Figure 6 shows the average eviction rates over our study period together with the

average Tenant-Right Index for each state Albeit weak we observe a negative correlation

coefficient between them that is consistent with our model prediction

5 Empirical Results

In this section we empirically test the model predictions detailed in Section 2 Recall that

our theoretical model predicts that with an increase in the Tenant-Right Index the median

rent and the homeless population will increase the housing supply the vacancy rate and

the eviction rate will decrease

11For more details see httpsevictionlaborgmethodsmore-questions

23

values higher than 6 as tenant-friendly states in the sense that they offer more protection

to tenants in more than half of the aspects that we consider important There are 23 states

with index values lower than 6 which are considered landlord-friendly and the remaining

are neutral states The average index for all states over our sample period is 554 implying

that on average state laws slightly favor landlords to tenants We also map the average index

of each state in Figure 2 Consistent with general expectations states in the West and East

coasts tend to be more tenant-friendly whereas Southern states provide more protection

to landlords Rhode Island has the highest index value of 11 followed by Massachusetts

Hawaii and Vermont On the contrary Utah and Colorado are the most landlord-friendly

states with an index lower than 1 followed by Louisiana and Georgia

Figure 3 traces the average Tenant-Right Index over our study period from 2005 to 2016

The small increase observed in this figure is driven by only seven states10 each having a

one-point increase at some time during this period except for Alabama which rose by three

points North Carolina is the only state that registered a lower index value but it is merely

by one point The indexes remain the same over our sample period for the rest of the states

Overall we observe that state laws are consistent over time except for a handful that seems

to move towards slightly more tenant protections

Next we examine the twelve legal provisions used to construct the index Our goal is to

show that every one of these statutes is necessary and reflects a unique aspect of the landlord-

tenant relationship and our index does not contain redundant information We first present

their pairwise correlations in Table 3 The low correlations between them indicate that each

of the twelve legal provisions likely contributes unique information to the final index To

further test whether we reduce the number of contributors to the index we perform Principal

Component Analysis (PCA) and report each componentrsquos eigenvalues in Figure 4

Although the general rule of thumb suggests that we should focus on the first four com-

ponents whose eigenvalues are greater than 1 they explain only 61 of the variation in

10They are Alabama Colorado Georgia Illinois New Hampshire Texas and Utah

19

our Tenant-Right Index This result is not surprising given the low pairwise correlations

observed earlier It implies that we will discard a large amount of information using only

these four leading components in our index Table 4 further reports the loading of the four

components on the original twelve variables For readability we only display absolute values

higher than 03 The first component seems to load heavily on rent-related factors while the

second component correlates strongly with legislation regarding deposits and termination

notice It is less clear for the third and fourth components as they include a wide range of

provisions In conclusion our PCA analysis results validate the importance of every statute

used to construct the Tenant-Right Index as a proxy for tenant protection

4 Data and Methodology

41 Empirical Models

We test the relationship between landlord-tenant laws and various housing outcomes by

estimating the following equation

OutcomeV arct = α + βTenantRightIndexst + λDemographicsControlsct+

+θPropertyControlsct + δYt + εc

(41)

where c and t denote city and year respectively All regressions include a set of year fixed

effects Yt and standard errors are clustered at the city level except in the homeless rate

regression where it is clustered at the state level Given that the Index does not contain a

lot of time series variation for any state we eschew the use of state fixed effects We return

to the question of unobserved state-level variables below

We are interested in five outcomes as the dependent variables in Equation (41) Our first

quantity of interest is the median gross rent in city c in year t defined by the Census Bureau

as the contract rent plus the estimated average monthly cost of utilities We also verify our

results using rent data from Zillow Our independent variable of interest is the state-level

20

TenantRightIndexs that we describe in Section 3 above If more protection for tenant rights

leads to higher rent as predicted by our theoretical model earlier we will observe a positive

β coefficient The regression controls for a set of local demographics variables including

population population growth population density median household income growth of

household income and homeownership rate We also account for property characteristics

at the city level namely the median number of rooms median property age and median

property tax paid

Our second dependent variable is the rental supply For each city in each year we

calculate the number of rental units per 100 residents as the total number of housing units

for rent (both occupied and vacant) divided by population and multiplied by 100 The

control variables include population growth population density median household income

growth of household income homeownership rate and median property tax Our theory

predicts a negative β coefficient on the Tenant-Right Index variable

Next we test for the impact of landlord regulation on the vacancy rate defined as the

number of vacant units divided by the total number of rental units and multiplied by 100

The β coefficient in this case is expected to be negative We control for all the demographics

variables listed earlier in the rent regression The fourth dependent variable is the homeless

rate calculated as the number of persons in homeless shelters divided by total population

and multiplied by 100 We expect β to carry a positive sign as predicted by our theory

Again the model includes a set of control variables population growth population density

median household income growth of household income the poverty rate (the share of the

population whose household income for the past 12 months was below the poverty level)

and average winter temperature We use data at the state-year level in this regression due

to the availability of the homeless population data

Our final regression tests the hypothesis that more substantial tenant protection results

in a lower eviction rate or in other words a negative β coefficient on the Tenant-Right

Index We employ two measures for the dependent variable namely the eviction filing rate

21

and eviction rate The eviction filing rate is the ratio of the number of eviction lawsuits filed

in a city over the number of renter-occupied homes in that city This measure counts all

eviction cases filed in an area including multiple lawsuits filed against the same address in

the same year On the other hand the eviction rate is the subset of those homes that received

an eviction judgment in which renters were ordered to leave which only counts the number

of unique addresses that received eviction judgments in a year The control variables include

population population growth population density median household income growth of

household income poverty rate and the share of the minority (non-white) population There

is the possibility that landlords will more readily evict non-white tenants

42 Data and Summary Statistics

Our sample covers the period from 2005 to 2016 Unless otherwise noted our data come

from the American Housing Survey estimates by the Census Bureau Table 5 presents

their summary statistics We note that the median gross rent for an average city in our

sample is $994 per month while the average Zillow Rent Index (ZRI) obtained from Zillow

is considerably higher at $1386 This is likely due to the fact that the index is computed

from Zillowrsquos estimates of rent for the full universe of houses whether or not they are ever

available for rent On the other hand the median rent estimates by the Census Bureau are

calculated from only rental houses which are often multi-family units Since single-family

homes are on average higher-valued than rental apartments including all of them in the ZRI

likely creates an upward bias in the index For robustness we use both rent measures in our

empirical tests

As a first look at the correlation between our Tenant-Right Index and rent Figure 5

shows the (average) median gross rent for each state and the size of the circles increases with

the average Tenant-Right Index Without controlling for other confounding factors there is

a weak positive correlation coefficient (037) between the two series as our theory suggests

Two other vital measures in our empirical tests are rental supply and vacancy rate

22

As shown in Table 5 the average city in our sample has 1965 units of rental houses for

every 100 persons or about 02 units per person and a vacancy rate of 987 Our fourth

variable of interest is the homeless rate We obtain estimates of the homeless population

from the Annual Homeless Assessment Report provided by the Department of Housing and

Urban Development The data are available at the state level and cover from 2007 to 2016

Notably the District of Columbia registered the highest homeless rate during this period at

123

Turning to our eviction measures we employ a novel database recently released by the

Eviction Lab at Princeton University This is the first comprehensive national eviction

database compiled using more than 80 million formal eviction records including eviction

requests from landlords and eviction orders from judges collected from the courts The

Eviction Lab data contain all the known information on the number of evictions filed in the

United States and made publicly available by municipalities11

The average filing rate and eviction rate across all sample cities from 2005 to 2016 are

772 and 361 respectively The city with the highest eviction filing rate of 6213 is

East Orange (New Jersey) in 2006 which also has the highest supply of rental housing per

capita in our sample and the highest eviction rate of 2098 is observed in Flint (Michigan)

in 2006 Figure 6 shows the average eviction rates over our study period together with the

average Tenant-Right Index for each state Albeit weak we observe a negative correlation

coefficient between them that is consistent with our model prediction

5 Empirical Results

In this section we empirically test the model predictions detailed in Section 2 Recall that

our theoretical model predicts that with an increase in the Tenant-Right Index the median

rent and the homeless population will increase the housing supply the vacancy rate and

the eviction rate will decrease

11For more details see httpsevictionlaborgmethodsmore-questions

23

our Tenant-Right Index This result is not surprising given the low pairwise correlations

observed earlier It implies that we will discard a large amount of information using only

these four leading components in our index Table 4 further reports the loading of the four

components on the original twelve variables For readability we only display absolute values

higher than 03 The first component seems to load heavily on rent-related factors while the

second component correlates strongly with legislation regarding deposits and termination

notice It is less clear for the third and fourth components as they include a wide range of

provisions In conclusion our PCA analysis results validate the importance of every statute

used to construct the Tenant-Right Index as a proxy for tenant protection

4 Data and Methodology

41 Empirical Models

We test the relationship between landlord-tenant laws and various housing outcomes by

estimating the following equation

OutcomeV arct = α + βTenantRightIndexst + λDemographicsControlsct+

+θPropertyControlsct + δYt + εc

(41)

where c and t denote city and year respectively All regressions include a set of year fixed

effects Yt and standard errors are clustered at the city level except in the homeless rate

regression where it is clustered at the state level Given that the Index does not contain a

lot of time series variation for any state we eschew the use of state fixed effects We return

to the question of unobserved state-level variables below

We are interested in five outcomes as the dependent variables in Equation (41) Our first

quantity of interest is the median gross rent in city c in year t defined by the Census Bureau

as the contract rent plus the estimated average monthly cost of utilities We also verify our

results using rent data from Zillow Our independent variable of interest is the state-level

20

TenantRightIndexs that we describe in Section 3 above If more protection for tenant rights

leads to higher rent as predicted by our theoretical model earlier we will observe a positive

β coefficient The regression controls for a set of local demographics variables including

population population growth population density median household income growth of

household income and homeownership rate We also account for property characteristics

at the city level namely the median number of rooms median property age and median

property tax paid

Our second dependent variable is the rental supply For each city in each year we

calculate the number of rental units per 100 residents as the total number of housing units

for rent (both occupied and vacant) divided by population and multiplied by 100 The

control variables include population growth population density median household income

growth of household income homeownership rate and median property tax Our theory

predicts a negative β coefficient on the Tenant-Right Index variable

Next we test for the impact of landlord regulation on the vacancy rate defined as the

number of vacant units divided by the total number of rental units and multiplied by 100

The β coefficient in this case is expected to be negative We control for all the demographics

variables listed earlier in the rent regression The fourth dependent variable is the homeless

rate calculated as the number of persons in homeless shelters divided by total population

and multiplied by 100 We expect β to carry a positive sign as predicted by our theory

Again the model includes a set of control variables population growth population density

median household income growth of household income the poverty rate (the share of the

population whose household income for the past 12 months was below the poverty level)

and average winter temperature We use data at the state-year level in this regression due

to the availability of the homeless population data

Our final regression tests the hypothesis that more substantial tenant protection results

in a lower eviction rate or in other words a negative β coefficient on the Tenant-Right

Index We employ two measures for the dependent variable namely the eviction filing rate

21

and eviction rate The eviction filing rate is the ratio of the number of eviction lawsuits filed

in a city over the number of renter-occupied homes in that city This measure counts all

eviction cases filed in an area including multiple lawsuits filed against the same address in

the same year On the other hand the eviction rate is the subset of those homes that received

an eviction judgment in which renters were ordered to leave which only counts the number

of unique addresses that received eviction judgments in a year The control variables include

population population growth population density median household income growth of

household income poverty rate and the share of the minority (non-white) population There

is the possibility that landlords will more readily evict non-white tenants

42 Data and Summary Statistics

Our sample covers the period from 2005 to 2016 Unless otherwise noted our data come

from the American Housing Survey estimates by the Census Bureau Table 5 presents

their summary statistics We note that the median gross rent for an average city in our

sample is $994 per month while the average Zillow Rent Index (ZRI) obtained from Zillow

is considerably higher at $1386 This is likely due to the fact that the index is computed

from Zillowrsquos estimates of rent for the full universe of houses whether or not they are ever

available for rent On the other hand the median rent estimates by the Census Bureau are

calculated from only rental houses which are often multi-family units Since single-family

homes are on average higher-valued than rental apartments including all of them in the ZRI

likely creates an upward bias in the index For robustness we use both rent measures in our

empirical tests

As a first look at the correlation between our Tenant-Right Index and rent Figure 5

shows the (average) median gross rent for each state and the size of the circles increases with

the average Tenant-Right Index Without controlling for other confounding factors there is

a weak positive correlation coefficient (037) between the two series as our theory suggests

Two other vital measures in our empirical tests are rental supply and vacancy rate

22

As shown in Table 5 the average city in our sample has 1965 units of rental houses for

every 100 persons or about 02 units per person and a vacancy rate of 987 Our fourth

variable of interest is the homeless rate We obtain estimates of the homeless population

from the Annual Homeless Assessment Report provided by the Department of Housing and

Urban Development The data are available at the state level and cover from 2007 to 2016

Notably the District of Columbia registered the highest homeless rate during this period at

123

Turning to our eviction measures we employ a novel database recently released by the

Eviction Lab at Princeton University This is the first comprehensive national eviction

database compiled using more than 80 million formal eviction records including eviction

requests from landlords and eviction orders from judges collected from the courts The

Eviction Lab data contain all the known information on the number of evictions filed in the

United States and made publicly available by municipalities11

The average filing rate and eviction rate across all sample cities from 2005 to 2016 are

772 and 361 respectively The city with the highest eviction filing rate of 6213 is

East Orange (New Jersey) in 2006 which also has the highest supply of rental housing per

capita in our sample and the highest eviction rate of 2098 is observed in Flint (Michigan)

in 2006 Figure 6 shows the average eviction rates over our study period together with the

average Tenant-Right Index for each state Albeit weak we observe a negative correlation

coefficient between them that is consistent with our model prediction

5 Empirical Results

In this section we empirically test the model predictions detailed in Section 2 Recall that

our theoretical model predicts that with an increase in the Tenant-Right Index the median

rent and the homeless population will increase the housing supply the vacancy rate and

the eviction rate will decrease

11For more details see httpsevictionlaborgmethodsmore-questions

23

TenantRightIndexs that we describe in Section 3 above If more protection for tenant rights

leads to higher rent as predicted by our theoretical model earlier we will observe a positive

β coefficient The regression controls for a set of local demographics variables including

population population growth population density median household income growth of

household income and homeownership rate We also account for property characteristics

at the city level namely the median number of rooms median property age and median

property tax paid

Our second dependent variable is the rental supply For each city in each year we

calculate the number of rental units per 100 residents as the total number of housing units

for rent (both occupied and vacant) divided by population and multiplied by 100 The

control variables include population growth population density median household income

growth of household income homeownership rate and median property tax Our theory

predicts a negative β coefficient on the Tenant-Right Index variable

Next we test for the impact of landlord regulation on the vacancy rate defined as the

number of vacant units divided by the total number of rental units and multiplied by 100

The β coefficient in this case is expected to be negative We control for all the demographics

variables listed earlier in the rent regression The fourth dependent variable is the homeless

rate calculated as the number of persons in homeless shelters divided by total population

and multiplied by 100 We expect β to carry a positive sign as predicted by our theory

Again the model includes a set of control variables population growth population density

median household income growth of household income the poverty rate (the share of the

population whose household income for the past 12 months was below the poverty level)

and average winter temperature We use data at the state-year level in this regression due

to the availability of the homeless population data

Our final regression tests the hypothesis that more substantial tenant protection results

in a lower eviction rate or in other words a negative β coefficient on the Tenant-Right

Index We employ two measures for the dependent variable namely the eviction filing rate

21

and eviction rate The eviction filing rate is the ratio of the number of eviction lawsuits filed

in a city over the number of renter-occupied homes in that city This measure counts all

eviction cases filed in an area including multiple lawsuits filed against the same address in

the same year On the other hand the eviction rate is the subset of those homes that received

an eviction judgment in which renters were ordered to leave which only counts the number

of unique addresses that received eviction judgments in a year The control variables include

population population growth population density median household income growth of

household income poverty rate and the share of the minority (non-white) population There

is the possibility that landlords will more readily evict non-white tenants

42 Data and Summary Statistics

Our sample covers the period from 2005 to 2016 Unless otherwise noted our data come

from the American Housing Survey estimates by the Census Bureau Table 5 presents

their summary statistics We note that the median gross rent for an average city in our

sample is $994 per month while the average Zillow Rent Index (ZRI) obtained from Zillow

is considerably higher at $1386 This is likely due to the fact that the index is computed

from Zillowrsquos estimates of rent for the full universe of houses whether or not they are ever

available for rent On the other hand the median rent estimates by the Census Bureau are

calculated from only rental houses which are often multi-family units Since single-family

homes are on average higher-valued than rental apartments including all of them in the ZRI

likely creates an upward bias in the index For robustness we use both rent measures in our

empirical tests

As a first look at the correlation between our Tenant-Right Index and rent Figure 5

shows the (average) median gross rent for each state and the size of the circles increases with

the average Tenant-Right Index Without controlling for other confounding factors there is

a weak positive correlation coefficient (037) between the two series as our theory suggests

Two other vital measures in our empirical tests are rental supply and vacancy rate

22

As shown in Table 5 the average city in our sample has 1965 units of rental houses for

every 100 persons or about 02 units per person and a vacancy rate of 987 Our fourth

variable of interest is the homeless rate We obtain estimates of the homeless population

from the Annual Homeless Assessment Report provided by the Department of Housing and

Urban Development The data are available at the state level and cover from 2007 to 2016

Notably the District of Columbia registered the highest homeless rate during this period at

123

Turning to our eviction measures we employ a novel database recently released by the

Eviction Lab at Princeton University This is the first comprehensive national eviction

database compiled using more than 80 million formal eviction records including eviction

requests from landlords and eviction orders from judges collected from the courts The

Eviction Lab data contain all the known information on the number of evictions filed in the

United States and made publicly available by municipalities11

The average filing rate and eviction rate across all sample cities from 2005 to 2016 are

772 and 361 respectively The city with the highest eviction filing rate of 6213 is

East Orange (New Jersey) in 2006 which also has the highest supply of rental housing per

capita in our sample and the highest eviction rate of 2098 is observed in Flint (Michigan)

in 2006 Figure 6 shows the average eviction rates over our study period together with the

average Tenant-Right Index for each state Albeit weak we observe a negative correlation

coefficient between them that is consistent with our model prediction

5 Empirical Results

In this section we empirically test the model predictions detailed in Section 2 Recall that

our theoretical model predicts that with an increase in the Tenant-Right Index the median

rent and the homeless population will increase the housing supply the vacancy rate and

the eviction rate will decrease

11For more details see httpsevictionlaborgmethodsmore-questions

23

and eviction rate The eviction filing rate is the ratio of the number of eviction lawsuits filed

in a city over the number of renter-occupied homes in that city This measure counts all

eviction cases filed in an area including multiple lawsuits filed against the same address in

the same year On the other hand the eviction rate is the subset of those homes that received

an eviction judgment in which renters were ordered to leave which only counts the number

of unique addresses that received eviction judgments in a year The control variables include

population population growth population density median household income growth of

household income poverty rate and the share of the minority (non-white) population There

is the possibility that landlords will more readily evict non-white tenants

42 Data and Summary Statistics

Our sample covers the period from 2005 to 2016 Unless otherwise noted our data come

from the American Housing Survey estimates by the Census Bureau Table 5 presents

their summary statistics We note that the median gross rent for an average city in our

sample is $994 per month while the average Zillow Rent Index (ZRI) obtained from Zillow

is considerably higher at $1386 This is likely due to the fact that the index is computed

from Zillowrsquos estimates of rent for the full universe of houses whether or not they are ever

available for rent On the other hand the median rent estimates by the Census Bureau are

calculated from only rental houses which are often multi-family units Since single-family

homes are on average higher-valued than rental apartments including all of them in the ZRI

likely creates an upward bias in the index For robustness we use both rent measures in our

empirical tests

As a first look at the correlation between our Tenant-Right Index and rent Figure 5

shows the (average) median gross rent for each state and the size of the circles increases with

the average Tenant-Right Index Without controlling for other confounding factors there is

a weak positive correlation coefficient (037) between the two series as our theory suggests

Two other vital measures in our empirical tests are rental supply and vacancy rate

22

As shown in Table 5 the average city in our sample has 1965 units of rental houses for

every 100 persons or about 02 units per person and a vacancy rate of 987 Our fourth

variable of interest is the homeless rate We obtain estimates of the homeless population

from the Annual Homeless Assessment Report provided by the Department of Housing and

Urban Development The data are available at the state level and cover from 2007 to 2016

Notably the District of Columbia registered the highest homeless rate during this period at

123

Turning to our eviction measures we employ a novel database recently released by the

Eviction Lab at Princeton University This is the first comprehensive national eviction

database compiled using more than 80 million formal eviction records including eviction

requests from landlords and eviction orders from judges collected from the courts The

Eviction Lab data contain all the known information on the number of evictions filed in the

United States and made publicly available by municipalities11

The average filing rate and eviction rate across all sample cities from 2005 to 2016 are

772 and 361 respectively The city with the highest eviction filing rate of 6213 is

East Orange (New Jersey) in 2006 which also has the highest supply of rental housing per

capita in our sample and the highest eviction rate of 2098 is observed in Flint (Michigan)

in 2006 Figure 6 shows the average eviction rates over our study period together with the

average Tenant-Right Index for each state Albeit weak we observe a negative correlation

coefficient between them that is consistent with our model prediction

5 Empirical Results

In this section we empirically test the model predictions detailed in Section 2 Recall that

our theoretical model predicts that with an increase in the Tenant-Right Index the median

rent and the homeless population will increase the housing supply the vacancy rate and

the eviction rate will decrease

11For more details see httpsevictionlaborgmethodsmore-questions

23

As shown in Table 5 the average city in our sample has 1965 units of rental houses for

every 100 persons or about 02 units per person and a vacancy rate of 987 Our fourth

variable of interest is the homeless rate We obtain estimates of the homeless population

from the Annual Homeless Assessment Report provided by the Department of Housing and

Urban Development The data are available at the state level and cover from 2007 to 2016

Notably the District of Columbia registered the highest homeless rate during this period at

123

Turning to our eviction measures we employ a novel database recently released by the

Eviction Lab at Princeton University This is the first comprehensive national eviction

database compiled using more than 80 million formal eviction records including eviction

requests from landlords and eviction orders from judges collected from the courts The

Eviction Lab data contain all the known information on the number of evictions filed in the

United States and made publicly available by municipalities11

The average filing rate and eviction rate across all sample cities from 2005 to 2016 are

772 and 361 respectively The city with the highest eviction filing rate of 6213 is

East Orange (New Jersey) in 2006 which also has the highest supply of rental housing per

capita in our sample and the highest eviction rate of 2098 is observed in Flint (Michigan)

in 2006 Figure 6 shows the average eviction rates over our study period together with the

average Tenant-Right Index for each state Albeit weak we observe a negative correlation

coefficient between them that is consistent with our model prediction

5 Empirical Results

In this section we empirically test the model predictions detailed in Section 2 Recall that

our theoretical model predicts that with an increase in the Tenant-Right Index the median

rent and the homeless population will increase the housing supply the vacancy rate and

the eviction rate will decrease

11For more details see httpsevictionlaborgmethodsmore-questions

23

51 Rent Affordability

511 Baseline Results

We begin with relating the Tenant-Right Index to rent affordability by estimating Equation

(41) We hypothesize that landlords may perceive higher costs associated with rental activi-

ties in areas where the landlord regulation is strict implying a positive relationship between

the index and rent levels The results presented in Table 6 strongly support this hypothesis

Using the median gross rent from the Census Bureau we estimate that a one-unit increase in

our Tenant-Right Index is associated with a $17 rent increase in column (1) Given that the

average median rent in our sample is $994 per month this is equivalent to a 17 increase

in rents More notably it amounts to an 18 difference in rent costs when we compare

the most tenant-friendly state (index value of 11) to the most landlord-friendly state (index

value of 042) holding all other factors constant

In the second column of Table 6 we exclude 38 cities with active rent control policies

Although such laws may keep rents below market levels for tenants in controlled units the

uncontrolled sector may see increased rents as a result of constrained supply (Early 2000

Diamond et al 2019) and therefore distort the median rent observed in these cities It is

reassuring to find that eliminating rent control cities has little effect on our indexrsquos coefficient

except for a slight reduction in its magnitude

We repeat the model specifications of column (1)-(2) in column (3)-(4) in which we replace

the Census measure of rent with ZRI The coefficient magnitudes are consistently larger in

column (3)-(4) since ZRI has higher values on average than the median rent estimates used

in the first two columns As explained earlier ZRI likely overestimates rent levels because its

sample includes all houses instead of only rental units Nonetheless the coefficients suggest

that the average monthly rent ($1386) increases by 25-3 for every one-unit increase in

our Tenant-Right Index

Turning to the coefficient estimates of other control variables we find that they are

24

largely consistent with prior expectations Cities with denser population higher income

and higher property tax tend to have higher rent Rent levels are also higher where rental

units have more rooms and lower where properties are older In addition places with high

home-ownership rates are less costly to rent Finally we find that rent levels are negatively

related to population and income growth but this is possibly caused by affordable cities

experiencing higher growth from migration from more expensive areas

512 Instrumental Variable Estimation

Despite the many control variables included in the tests our empirical findings from the

OLS models could be spurious For example a city with overwhelmingly high rental costs

or scarce rental units may be more likely to increase the level of tenant protection against

landlords whereas a city with affordable and abundant rental housing may not find the need

to pass such laws As a result looking across communities we could observe that cities with

expensive rental houses also have a higher Tenant-Right Index without causality running

from the index to higher rents

We employ an instrumental variable strategy utilizing voting outcomes in presidential

elections to address the endogeneity issue discussed above In particular we make use of

the opposite standpoints of the two major political parties in the US The Democratic

Party is commonly known for being proactive in increasing social welfare benefits including

improving tenant protection such as passing rent control and just-cause eviction bills12 On

the other hand the real estate industry has long relied on Republican politicians to block

pro-tenant legislation and promote business-friendly policies13 Thus we argue that the

more support a state has for the Democratic Party (the so-called ldquobluerdquo states) the more

likely their laws lean in the direction of protecting tenants more than landlords For this

12Andrew Jeff 2020 ldquoWhere the Democratic primary candidates standon housingrdquo Curbed February 25 httpswwwcurbedcom201961918644798

democratic-primary-housing-policy-2020-president13Campanile Carl 2018 ldquoLandlords are terrified Democrats could take con-

trol over New Yorkrdquo New York Post September 17 httpsnypostcom20180917

landlords-are-terrified-democrats-could-take-control-over-new-york

25

largely consistent with prior expectations Cities with denser population higher income

and higher property tax tend to have higher rent Rent levels are also higher where rental

units have more rooms and lower where properties are older In addition places with high

home-ownership rates are less costly to rent Finally we find that rent levels are negatively

related to population and income growth but this is possibly caused by affordable cities

experiencing higher growth from migration from more expensive areas

512 Instrumental Variable Estimation

Despite the many control variables included in the tests our empirical findings from the

OLS models could be spurious For example a city with overwhelmingly high rental costs

or scarce rental units may be more likely to increase the level of tenant protection against

landlords whereas a city with affordable and abundant rental housing may not find the need

to pass such laws As a result looking across communities we could observe that cities with

expensive rental houses also have a higher Tenant-Right Index without causality running

from the index to higher rents

We employ an instrumental variable strategy utilizing voting outcomes in presidential

elections to address the endogeneity issue discussed above In particular we make use of

the opposite standpoints of the two major political parties in the US The Democratic

Party is commonly known for being proactive in increasing social welfare benefits including

improving tenant protection such as passing rent control and just-cause eviction bills12 On

the other hand the real estate industry has long relied on Republican politicians to block

pro-tenant legislation and promote business-friendly policies13 Thus we argue that the

more support a state has for the Democratic Party (the so-called ldquobluerdquo states) the more

likely their laws lean in the direction of protecting tenants more than landlords For this

12Andrew Jeff 2020 ldquoWhere the Democratic primary candidates standon housingrdquo Curbed February 25 httpswwwcurbedcom201961918644798

democratic-primary-housing-policy-2020-president13Campanile Carl 2018 ldquoLandlords are terrified Democrats could take con-

trol over New Yorkrdquo New York Post September 17 httpsnypostcom20180917

landlords-are-terrified-democrats-could-take-control-over-new-york

25

reason we use the percentage of votes for the Democratic presidential candidates at the state

level as the instrument for our Tenant-Right Index and re-estimate our base models using

2SLS The identifying assumption is that conditional on the numerous control variables in

our specification the impact that party affiliation has on rents works only through policy

innovations of the type summarized in our Index The first stage results are shown in Table 7

confirm that a higher voting percentage for the Democratic Party is a strong predictor of our

Tenant-Right Index suggesting that our instrument passes the relevance test Additionally

the Kleibergen-Paap Wald F statistics for the voting instrument are well above the commonly

accepted threshold of 10 indicating that the instrument is not weak

We obtain the vote percentage from the MIT Election Data and Science Lab which

provides vote data at the state level for every presidential election since 1976 For non-

election years we use the values from the nearest previous election The summary statistics

of this variable is reported in Table 5 On average 4875 of votes support candidates from

the Democratic Party but this share can range from 2163 in ldquoredrdquo states to 9246 in

ldquobluerdquo states

Column (5)-(8) in Table 6 display the second-stage coefficients from our instrumental

variable analysis Again we find that the instrumented Tenant-Right Index is positively

and significantly correlated with rents and the magnitude of the coefficient estimates are

approximately three times that of the corresponding OLS specifications Our preferred

specification in column (5) suggests that median rent increases by 61 for every one-unit

increase in our index To summarize consistent with the theory model prediction our

empirical results in this section indicate that tighter landlord regulation results in higher

equilibrium rents since landlords need to be compensated for the increased costs to separate

from bad tenants Our finding is robust to the inclusion of a large set of control variables

as well as the use of an instrumental variable to address the potential endogeneity issue

26

52 Rental Housing Supply Vacancy and Homeless Rates

Next we discuss our empirical estimates for the relationship between landlord-tenant laws

and several other housing outcomes addressed in our theoretical model Note that for

brevity we only report selected control variables whose coefficients are statistically and

economically significant in all the following tables As expected column (1) of Table 8

shows that tenant rights protection is negatively correlated with the stock of rental houses

A one-unit increase in our index leads to 00028 fewer rental units available for every resident

or 14 less supply on average all else constant14 Prior research on rent control policies has

also found similar evidence of landlords exiting the market in response to such ordinances

(Autor et al (2014) Diamond et al 2019 Sims 2007) For example Diamond et al (2019)

documents that rental supply in San Francisco decreases by 15 due to landlords selling to

owner-occupants or redeveloping their properties

In the second column of Table 8 the coefficient on our Tenant-Right Index suggests that

vacancy rates tend to be lower in areas with higher tenant protection This effect follows

from the lower supply result found above We turn to examine homelessness in the third

regression Although the estimate is only statistically significant at the 10 level the sign of

the coefficient is consistent with our theoretical prediction of a positive relationship between

tenant protections and the homeless rate In other words we document that tenant-favorable

laws inadvertently lead to more homelessness which is again a result of reduced rental supply

To address the concern that our OLS estimates may suffer from the endogeneity issue

described in the previous subsection we present the results from the corresponding IV spec-

ifications in the last three columns All the coefficients remain statistically significant and

consistent with our model predictions but the point estimates are notably bigger in mag-

nitude Increasing the Tenant-Right Index by one unit reduces the rental supply by 21

decreases the local rental vacancy by 89 and increases the local homeless rate by as much

14This is calculated as (0002801965)100 where the numerator is the avegage rental supply per personas reported in Table 5

27

as 26

53 Evictions

Thus far our analysis points to the conclusion that more stringent landlord regulation is

paradoxically but unsurprisingly damaging to tenants in terms of higher rent lower supply

lower vacancy and more homelessness Advocates of tenant rights however have argued

for its benefits in the fight against the eviction crisis in the US which we now turn to in

this subsection Table 9 presents our empirical results on the relationship between evictions

and the Tenant-Right Index The dependent variables are eviction filing rate and eviction

rate in column (1) and (2) respectively As described in the data section the filing rate

includes multiple filings against the same address while the eviction rate variable only counts

unique eviction judgments and is a more accurate measure of evictions Both coefficients are

statistically significant at the 1 level but the magnitude in the case of the eviction rate is

58 smaller

We use the IV approach discussed in the previous section in the last two columns and

obtain larger point estimates for the coefficients The magnitude suggests that every one-

unit increase in the Tenant-Right Index is associated with a 032 percentage point decrease

in eviction rate Given that the average eviction rate across all cities in our sample is

361 (Table 5) this effect is undoubtedly economically significant equivalent to an 89

decrease Overall these results imply that pro-tenant laws are effective in reducing the

eviction problem landlords are less likely to obtain eviction judgments where there is strong

protection for tenant rights

In summary our empirical results in this section suggest that while tenant-friendly cities

on average have lower eviction rates than landlord-friendly areas the former has worse

rent affordability than the latter Thus it is important for policy makers to recognize the

delicate trade-off between tenant protections and rent affordability imposing strict landlord

regulations may protect tenants from potential hardships associated with eviction but at

28

the cost of higher overall rent levels lower supply and higher homeless rates in the long run

The welfare effects of tenant rights depend on the presumably large benefits for those who

avoid eviction versus a loss of consumer surplus for other housing consumers

6 Conclusion

Every minute four renters are evicted from their rental homes Policymakers are seeking

solutions to reduce the level of eviction-triggered residential mobility This study provides

economic analyses that shed light on the net impact of landlord-tenant laws on eviction and

rent affordability Our paper offers three major contributions to the literature on affordable

rent First we are the first to study the net impact of landlord-tenant laws on rent afford-

ability using both ex ante regulations and ex post eviction judgments Second we are the

first to construct a novel state-level index to proxy for the level of tenant protections Third

we provide both theoretical suggestions and empirical evidence showcasing the relationship

between tenant rights eviction rates and rent prices

On the one hand we find that rent levels supply and vacancy rates are strongly corre-

lated with the extent of tenant protection A one-unit increase in the Tenant-Right Index

is responsible for a 61 percent increase in rent prices a 21 percent decrease in the supply

of rental units and an 89 percent decrease in vacancy rate in a city On the other hand

our analysis of tenant rights and eviction outcomes reveals that a one-unit increase in the

Tenant-Rights Index is associated with an 89 percent decrease in eviction rates Hence our

findings highlight an important trade-off between tenant protections and rent affordability

imposing strict landlord regulations may protect tenants from potential hardships associated

with eviction in the short-run but at the cost of higher overall rent levels for everyone in

the long-run This has important implications for landlord-tenants regulations that should

be of great interest to policymakers

29

References

Arnott R and E Shevyakhova (2014) Tenancy rent control and credible commitment inmaintenance Regional Science and Urban Economics 47 72ndash85

Autor D H C J Palmer and P A Pathak (2014) Housing market spillovers Ev-idence from the end of rent control in cambridge massachusetts Journal of PoliticalEconomy 122 (3) 661ndash717

Been V I Gould Ellen and S House (2019) Laboratories of regulation Understandingthe diversity of rent regulation laws Fordham Urb LJ 46 1041

Bennett M (2016) Security of tenure for generation rent Irish and Scottish approachesVictoria U Wellington L Rev 47 363

Borsch-Supan A (1986) Household formation housing prices and public policy impactsJournal of Public Economics 30 (2) 145ndash164

Coulson N E D Laing and P Wang (2001) Spatial mismatch in search equilibriumJournal of Labor Economics 19 (4) 949ndash972

Cunningham C K Gerardi and L Shen (2020) Double trigger for mortgage default Thefracking boom Management Science

Desmond M (2012) Eviction and the reproduction of urban poverty American Journal ofSociology 118 (1) 88ndash133

Desmond M (2016) Evicted Poverty and profit in the American city Broadway Books

Diamond R T McQuade and F Qian (2019) The effects of rent control expansion ontenants landlords and inequality Evidence from San Francisco American EconomicReview 109 (9) 3365ndash94

Dong M R F Anda V J Felitti D F Williamson S R Dube D W Brown and W HGiles (2005) Childhood residential mobility and multiple health risks during adolescenceand adulthood the hidden role of adverse childhood experiences Archives of pediatricsamp adolescent medicine 159 (12) 1104ndash1110

Early D W (2000) Rent control rental housing supply and the distribution of tenantbenefits Journal of Urban Economics 48 (2) 185ndash204

Gerardi K K F Herkenhoff L E Ohanian and P S Willen (2018) Canrsquot pay or wonrsquotpay Unemployment negative equity and strategic default The Review of FinancialStudies 31 (3) 1098ndash1131

Glaeser E L and E F Luttmer (2003) The misallocation of housing under rent controlAmerican Economic Review 93 (4) 1027ndash1046

Gyourko J and P Linneman (1989) Equity and efficiency aspects of rent control Anempirical study of new york city Journal of urban Economics 26 (1) 54ndash74

30

Kim K (2018) Payment performance and residency discounts in the rental housing marketReal Estate Economics

Manheim K (1989) Tenant eviction protection and the takings clause Wis L Rev 925

Molloy R et al (2020) The effect of housing supply regulation on housing affordability Areview Regional Science and Urban Economics 80 (C)

Oishi S (2010) The psychology of residential mobility Implications for the self socialrelationships and well-being Perspectives on Psychological Science 5 (1) 5ndash21

Pissarides C A (2000) Equilibrium unemployment theory MIT press

Pribesh S and D B Downey (1999) Why are residential and school moves associated withpoor school performance Demography 36 (4) 521ndash534

Sampson R J and P Sharkey (2008) Neighborhood selection and the social reproductionof concentrated racial inequality Demography 45 (1) 1ndash29

Sharkey P and R J Sampson (2010) Destination effects Residential mobility and trajec-tories of adolescent violence in a stratified metropolis Criminology 48 (3) 639ndash681

Sims D P (2007) Out of control What can we learn from the end of massachusetts rentcontrol Journal of Urban Economics 61 (1) 129ndash151

South S J and K D Crowder (1998) Housing discrimination and residential mobilityImpacts for blacks and whites Population Research and Policy Review 17 (4) 369ndash387

31

Figure 1 Effect of a Rise in Eviction Cost on Contact Rates

Notes This figure demonstrates the effect of an increase in d (eviction cost) on the contact rates for tenants

(micro) and landlords (λ)

32

Fig

ure

2T

enan

t-R

ights

Index

by

Sta

te

Not

es

Th

isfi

gure

feat

ure

sa

map

ofth

eU

nit

edS

tate

sw

ith

colo

r-co

din

gto

hig

hli

ght

the

Ten

ant-

Rig

ht

Ind

exva

lue

inea

chst

ate

T

he

data

use

din

this

map

are

show

nin

the

firs

tco

lum

nof

Tab

le2

33

Figure 3 Tenant-Right Index by Year

Notes This figure shows the trend of average Tenant-Right Index by year during our sample period The

average Tenant-Right Index are steadily increasing from 2005 to 2016

Figure 4 Tenant-Right Index Principal Component Analysis Eigenvalues

Notes This figure features the scree plot of eigenvalues from the Principal Component Analysis The first

four principal components have eigenvalues greater than 1 These top four eigenvalues account for 61 of

the variance in the Tenant-Right Index

34

Fig

ure

5T

enan

t-R

ight

Index

and

Med

ian

Gro

ssR

ent

Not

es

Th

isfi

gure

feat

ure

sa

map

ofth

eU

nit

edS

tate

sw

ith

colo

r-co

din

gto

hig

hli

ght

the

leve

lsof

Med

ian

Ren

tin

each

state

T

he

size

of

the

circ

les

corr

esp

ond

sto

the

valu

eof

Ten

ant-

Rig

ht

Ind

exin

that

state

35

Fig

ure

6T

enan

t-R

ight

Index

and

Evic

tion

Rat

e

Not

es

Th

isfi

gure

feat

ure

sa

map

ofth

eU

nit

edSta

tes

wit

hco

lor-

cod

ing

toh

igh

light

the

evic

tion

rate

inea

chst

ate

T

he

size

of

the

circ

les

incr

ease

s

wit

hth

eva

lue

ofth

eT

enan

t-R

ight

Ind

exin

that

state

36

Tab

le1

Defi

nit

ion

ofL

andlo

rd-T

enan

tL

awP

rovis

ions

Abbre

via

tion

Defi

nit

ion

Exam

ple

Max

Dep

osit

Max

imum

secu

rity

dep

osit

for

anunfu

rnis

hed

apar

tmen

ton

aon

eye

arle

ase

One

mon

thrsquos

rent

Dep

osit

Inte

rest

Whet

her

landlo

rdm

ust

keep

secu

rity

dep

osit

sin

anin

tere

st-b

eari

ng

acco

unt

Yes

No

Dep

osit

Ret

urn

Dea

dline

for

retu

rnin

gse

curi

tydep

osit

when

no

ded

uct

ions

are

imp

osed

by

landlo

rd45

day

sR

egula

rT

erm

inat

ion

Min

imum

not

ice

toch

ange

term

sor

term

inat

ea

mon

th-t

o-m

onth

tenan

cy30

day

sR

ent

Incr

ease

Not

ice

Min

imum

not

ice

toin

crea

sere

nt

for

am

onth

-to-

mon

thte

nan

cy30

day

sR

ent

Wit

hhol

din

gT

enan

thas

the

opti

onto

wit

hhol

dre

nt

for

failure

topro

vid

ees

senti

alse

rvic

esY

esN

oR

epai

ran

dD

educt

Ten

ant

isal

low

edto

repai

ran

dded

uct

cost

sfr

omre

nt

Yes

No

Non

pay

men

tT

erm

inat

ion

Ter

min

atio

nnot

ice

requir

edfo

rnon

pay

men

tof

rent

5d

ays

Lea

seV

iola

tion

Ter

min

atio

nT

erm

inat

ion

not

ice

requir

edfo

rle

ase

vio

lati

on10

day

sSel

f-hel

pE

vic

tion

Am

ount

tenan

tca

nsu

ela

ndlo

rdfo

rse

lf-h

elp

evic

tion

Act

ual

dam

age

Rig

ht

toSta

yW

het

her

tenan

thas

the

righ

tto

stay

afte

rille

gal

evic

tion

Yes

No

Ren

tC

ontr

olP

reem

pti

onW

het

her

stat

ela

ws

explici

tly

pre

-em

pt

loca

lgo

vern

men

tsto

contr

olre

nt

Yes

No

Not

es

Th

ista

ble

pro

vid

esth

ed

efin

itio

nof

the

twel

ve

law

pro

vis

ion

sre

gard

ing

lan

dlo

rd-t

enant

rela

tion

ship

su

sed

toco

nst

ruct

ou

rT

enant-

Rig

ht

Ind

ex

37

Tab

le2

Ten

ant-

Rig

ht

Index

by

Sta

tes

(200

5-20

16)

Sta

teT

enan

tM

axD

epos

itD

epos

itR

egula

rR

ent

Wit

hhol

dR

epai

rN

onpay

men

tL

ease

Sel

f-hel

pR

ight

Ren

tR

ights

Dep

osit

Inte

rest

Ret

urn

Ter

min

atio

nIn

crea

seR

ent

Ded

uct

Ter

min

atio

nV

iola

tion

Evic

tion

toC

ontr

olIn

dex

Not

ice

Ter

min

atio

nSta

yP

reem

pti

on

Ala

bam

a3

420

830

00

830

075

00

170

830

00

Ala

ska

800

00

11

11

10

01

11

Ari

zona

800

10

11

11

10

01

10

Ark

ansa

s2

000

00

10

00

01

00

0C

alif

ornia

800

00

11

11

10

01

11

Col

orad

o0

580

00

00

058

00

00

00

Con

nec

ticu

t7

000

10

00

11

11

11

0D

elaw

are

885

10

11

11

10

00

851

1D

istr

ict

ofC

olum

bia

800

11

01

01

01

11

01

Flo

rida

265

00

10

01

00

00

650

0G

eorg

ia1

250

00

10

00

250

00

00

Haw

aii

900

10

11

11

10

01

11

Idah

o2

000

01

10

00

00

00

0Il

linoi

s4

920

10

11

10

920

00

00

India

na

300

00

01

10

01

00

00

Iow

a5

000

00

11

11

00

01

0K

ansa

s6

001

00

10

10

11

01

0K

entu

cky

700

00

01

11

10

11

10

Lou

isia

na

100

00

00

00

10

00

00

Mai

ne

600

00

01

11

11

00

01

Mar

yla

nd

600

01

01

11

00

10

01

Mas

sach

use

tts

100

01

10

11

11

11

11

0M

ichig

an6

851

00

10

11

01

085

10

Min

nes

ota

615

01

11

01

10

00

151

0M

issi

ssip

pi

300

00

01

00

10

10

00

Mis

souri

300

00

01

01

10

00

00

Mon

tana

800

00

11

01

10

11

11

Neb

rask

a9

001

01

10

11

01

11

1N

evad

a7

000

00

11

11

00

11

1

38

Tab

le1

Ten

ant-

Rig

ht

Index

by

Sta

tes

(200

5-20

16)

(Con

tinue)

Sta

teT

enan

tM

axD

epos

itD

epos

itR

egula

rR

ent

Wit

hhol

dR

epai

rN

onpay

men

tL

ease

Sel

f-hel

pR

ight

Ren

tR

ights

Dep

osit

Inte

rest

Ret

urn

Ter

min

atio

nIn

crea

seR

ent

Ded

uct

Ter

min

atio

nV

iola

tion

Evic

tion

toC

ontr

olIn

dex

Not

ice

Ter

min

atio

nSta

yP

reem

pti

on

New

Ham

psh

ire

708

11

01

11

00

008

11

0N

ewJer

sey

800

11

01

11

11

00

01

New

Mex

ico

700

11

01

11

00

01

10

New

Yor

k5

900

10

10

11

00

09

01

Nor

thC

arol

ina

325

10

00

00

250

10

01

0N

orth

Dak

ota

585

11

01

10

10

00

850

0O

hio

400

01

01

01

00

00

01

Okla

hom

a6

000

00

10

11

01

11

0O

rego

n7

000

00

10

11

11

11

0P

ennsy

lvan

ia6

000

10

00

11

11

00

1R

hode

Isla

nd

110

01

01

11

11

11

11

1Sou

thC

arol

ina

600

00

01

01

10

11

10

Sou

thD

akot

a8

001

01

11

11

00

11

0T

ennes

see

700

00

01

01

11

11

10

Tex

as3

450

00

10

01

00

045

10

Uta

h0

420

00

00

00

420

00

00

Ver

mon

t9

000

01

11

11

11

01

1V

irgi

nia

500

01

01

01

00

10

10

Was

hin

gton

500

00

10

11

10

00

10

Wes

tV

irgi

nia

200

00

01

00

00

00

01

Wis

consi

n2

000

01

00

10

00

00

0W

yom

ing

200

00

00

01

00

00

01

Avera

ge

554

031

027

029

080

041

078

064

026

039

046

055

035

Not

es

Th

ista

ble

rep

orts

the

ind

ivid

ual

scor

esfo

rth

etw

elve

legal

pro

vis

ion

sas

wel

las

the

fin

al

Ten

ant-

Rig

ht

Ind

exin

each

state

A

llare

aver

age

valu

esov

er20

05-2

016

For

the

twel

vele

gal

pro

vis

ion

sa

score

of

1m

ean

sa

state

favors

ten

ants

tola

nd

lord

sT

he

Ten

ant-

Rig

ht

Ind

exis

the

sum

of

all

twel

vep

rovis

ions

39

Tab

le3

Pri

nci

pal

Com

pon

ent

Anal

ysi

sC

orre

lati

onM

atri

x

Max

Dep

osit

Dep

osit

Reg

ula

rR

ent

Wit

hhol

dR

epai

rN

onpay

men

tL

ease

Sel

f-hel

pR

ight

Ren

tD

epos

itIn

tere

stR

eturn

Ter

min

atio

nIn

crea

seR

ent

Ded

uct

Ter

min

atio

nV

iola

tion

Evic

tion

toC

ontr

olN

otic

eT

erm

inat

ion

Sta

yP

reem

pti

on

Max

imum

Dep

osit

100

0D

epos

itIn

tere

st0

157

100

0D

epos

itR

eturn

012

5-0

301

100

0R

egula

rT

erm

inat

ion

022

90

087

-00

011

000

Ren

tIn

crea

seN

otic

e0

300

011

00

247

031

81

000

Ren

tW

ithhol

din

g0

174

022

50

243

012

50

258

100

0R

epai

ran

dD

educt

-00

10-0

094

021

70

212

029

00

257

100

0N

onpay

men

tT

erm

inat

ion

018

50

139

-01

84-0

061

005

30

111

005

51

000

Lea

seV

iola

tion

Ter

min

atio

n0

064

005

6-0

164

019

8-0

255

022

90

107

035

41

000

Sel

f-hel

pE

vic

tion

036

60

0181

023

40

288

022

40

357

040

3-0

035

019

41

000

Rig

ht

toSta

y0

282

-01

490

326

025

50

278

042

70

419

006

90

254

056

51

000

Ren

tC

ontr

olP

reem

pti

on0

037

009

70

244

016

30

216

029

80

128

012

7-0

003

010

4-0

073

100

0

Not

es

Th

ista

ble

rep

orts

the

corr

elat

ion

mat

rix

of

the

twel

veva

riab

les

incl

ud

edin

the

con

stru

ctio

nou

rT

enant-

Rig

ht

Ind

ex

Th

eco

rrel

ati

on

bet

wee

n

diff

eren

tva

riab

les

are

low

bec

ause

each

ofth

emre

pre

sents

au

niq

ue

legal

asp

ect

of

ten

ant

rights

40

Table 4 Principal Component Analysis Eigenvectors

(1) (2) (3) (4) (5)Variable Comp1 Comp2 Comp3 Comp4 Unexplained

Maximum Deposit -0428 0466Deposit Interest 0433 0457 0369Deposit Return -0479 0339Regular Termination -0333 0603Rent Increase Notice 0320 0438 0345Rent Withholding 0369 0325 0425Repair and Deduct 0339 0547Nonpayment Termination 0492 0483Lease Violation Termination 0497 -0430 0224Rent Control Preemption 0426 0373Right to Stay 0434 0245Rent Control Preemption 0428 0566 0273

Notes This table reports eigenvectors of the four leading components In these results the first principal

component primarily loads on rent related factors The second component primarily loads on deposits and

termination related factors The third component and the fourth component include a wide range of tenant

right proxies Blanks are abs(loading)lt 03

41

Table 5 Descriptive Statistics

(1) (2) (3) (4) (5)Variables N Mean Std Dev Min Max

City Level DataPopulation (rsquo000) 5928 19987 45025 5667 855041Population Growth () 5928 137 454 -4890 3995Median Household Income (rsquo000) 5928 $5424 $1916 $2119 $15137Income Growth () 5928 224 835 -3274 4417Poverty Rate () 3784 1521 627 296 4063Share of Minority Population () 5785 3394 1768 302 9686Median Number of Rooms 5928 421 043 130 820Median Property Age (years) 5928 3658 1418 400 7700Median Property Tax 5928 $2673 $1526 $185 $10000Median Gross Rent 5928 $994 $298 $466 $3042Density (personssquare miles) 5928 4229 3936 163 54027Home-Ownership Rate () 5928 5651 1220 1655 9671Zillow Rent Index 3761 $1386 $570 $486 $4750Rental Supply per 100 Person (units) 2602 1965 506 527 3473Vacancy Rate () 2602 987 416 110 3314Eviction Filing Rate () 3198 772 776 000 6213Eviction Rate () 3143 361 268 000 2098

State Level DataTenant Rights Index 612 554 265 000 1100Share of Homeless Population () 500 019 016 005 123Share of Democratic Party Votes () 612 4875 1130 2163 9246Average Winter Temperature (F) 500 5294 914 2430 7870

Notes This table reports the summary statistics of variables used in the empirical tests

42

Tab

le6

Ten

ant-

Rig

ht

Index

and

Ren

tA

ffor

dab

ilit

y

OL

SIV

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dep

en

dent

Vari

ab

le

Med

ian

Med

ian

Zil

low

Zil

low

Med

ian

Med

ian

Zil

low

Zil

low

Ren

tR

ent

Ind

exIn

dex

Ren

tR

ent

Ind

exIn

dex

Ten

ant

Rig

hts

Ind

ex16

933

140

24

42

732

353

15

60

935

534

28

13

839

4

114

156

(2

443

)(2

371

)(5

834

)(5

448

)(6

757

)(6

074

)(1

696

9)(1

469

6)P

opu

lati

on-0

012

-00

43

-0

037

-01

45

-0

001

-00

25-0

016

-01

11

(00

12)

(00

12)

(00

38)

(00

33)

(00

10)

(00

19)

(00

30)

(00

49)

Pop

ula

tion

Gro

wth

-09

21

-10

24

-05

13-0

754

-03

71-0

559

-00

84-0

685

(04

07)

(04

18)

(20

46)

(19

52)

(04

12)

(04

15)

(22

65)

(20

62)

Med

ian

Hou

seh

old

Inco

me

121

91

10

953

210

60

16

212

100

88

9

517

17

125

139

49

(0

612

)(0

670

)(1

544

)(1

506

)(0

704

)(0

754

)(1

684

)(1

614

)In

com

eG

row

th-2

403

-21

52

-3

706

-28

23

-1

814

-17

13

-2

513

-20

78

(0

188

)(0

207

)(0

521

)(0

529

)(0

209

)(0

226

)(0

562

)(0

569

)M

edia

nN

um

ber

ofR

oom

s61

936

811

29

-1

897

79

-1

304

15

35

204

50

433

-263

042

-205

150

(13

750)

(13

701)

(37

592)

(34

921)

(17

943)

(17

695)

(44

990)

(42

789)

Med

ian

Pro

per

tyA

ge-2

207

-34

63

2

328

-11

02-4

040

-48

47

-1

831

-40

37

(0

637

)(0

514

)(1

547

)(1

164

)(0

708

)(0

695

)(1

707

)(1

549

)M

edia

nP

rop

erty

Tax

000

90

018

-0

015

002

40

003

000

8-0

031

-0

001

(00

05)

(00

07)

(00

13)

(00

17)

(00

06)

(00

08)

(00

15)

(00

18)

Hom

e-O

wn

ersh

ipR

ate

-41

76

-3

593

-10

692

-7

583

-19

87

-17

97

-64

69

-4

521

(0

839

)(0

828

)(2

116

)(1

825

)(1

025

)(1

031

)(2

438

)(2

171

)D

ensi

ty0

014

0

028

0

030

0

064

0

011

0

022

0

024

0

053

(0

006

)(0

003

)(0

013

)(0

007

)(0

005

)(0

003

)(0

011

)(0

008

)

K-P

F-s

tat

147

7217

321

146

2815

680

Ob

serv

atio

ns

592

85

508

376

13

487

591

25

492

376

13

487

Yea

rF

EY

esY

esY

esY

esY

esY

esY

esY

esIn

clu

de

Ren

tC

ontr

olC

itie

sY

esN

oY

esN

oY

esN

oY

esN

oA

dju

sted

-R0

765

077

70

678

070

80

660

069

10

550

060

8

Not

es

Th

ista

ble

rep

orts

our

esti

mat

ion

resu

lts

of

Equ

ati

on

41

inS

ecti

on

4

Colu

mn

(1)ndash

(4)

dis

pla

yth

eO

LS

resu

lts

an

dC

olu

mn

(5)-

(8)

dis

pla

y

the

seco

nd

stag

ees

tim

atio

nre

sult

sfr

omth

eIV

spec

ifica

tion

sT

he

Kle

iber

gen

-Paap

Wald

ran

kF

stati

stic

sfo

rea

chIV

regre

ssio

nare

rep

ort

edin

the

tab

le

We

contr

olfo

ra

set

oflo

cal

dem

ogra

ph

ics

vari

ab

les

incl

ud

ing

pop

ula

tion

p

op

ula

tion

gro

wth

p

op

ula

tion

den

sity

m

edia

nh

ou

seh

old

inco

me

grow

thof

hou

seh

old

inco

me

and

hom

eow

ner

ship

rate

W

eals

oco

ntr

ol

for

the

med

ian

nu

mb

erof

room

sm

edia

np

rop

erty

age

an

dm

edia

np

rop

erty

tax

pai

dat

the

city

leve

lSta

nd

ard

erro

rsar

eca

lcu

late

dat

the

city

leve

lC

lust

ered

stan

dard

erro

rsare

show

nin

pare

nth

eses

(

plt

00

1

plt

00

5

plt

01)

43

Table 7 First Stage Results for 2SLS IV Specifications

Dependent Variable Tenant-Right Index(1) (2) (3) (4)

Democratic() 0150 0158 0139 0145(0012) (0012) (0011) (0011)

Population -0000 -0000 -0000 0000(0000) (0000) (0000) (0000)

Population Growth -0004 -0002 -0003 0003(0004) (0004) (0011) (0011)

Median Household Income 0028 0031 0024 0026(0006) (0008) (0006) (0008)

Income Growth -0006 -0007 -0007 -0007(0002) (0002) (0002) (0003)

Median Number of Rooms 0386 0462 0519 0600(0186) (0203) (0198) (0212)

Median Property Age 0015 0020 0018 0022(007) (0007) (0007) (0007)

Median Property Tax -0000 -0000 0000 0000(0000) (0000) (0000) (0000)

Density 0000 0000 0000 0000(0000) (0000) (0000) (0000)

Home-ownership Rate -0036 -0041 -0032 -0034(0009) (0009) (0009) (0010)

K-P F-stat 14772 17321 14628 15680Observations 5912 5492 3761 3487Year FE Yes Yes Yes Yes

Notes This table displays the first stage estimation results for column (5)-(8) in Table 6 The mean

value of the instrument variableldquoDemocratic ()rdquo is 4819 percent Clustered standard errors are shown in

parentheses ( plt001 plt005 plt01)

44

Table 8 Tenant-Right Index Housing Supply Vacancy and Homeless Rate

OLS IV

(1) (2) (3) (4) (5) (6)Dependent Variable Rental Supply Vacancy Homeless Rental Supply Vacancy Homeless

(Per 100 ) Rate () Rate () (Per 100 ) Rate () Rate ()

Tenant Rights Index -0277 -0258 0008 -0411 -0870 0041(0053) (0067) (0004) (0165) (0224) (0021)

Median Household Income 0011 -0065 0004 0018 -0033 1496(0013) (0011) (0002) (0014) (0016) (2892)

Income Growth -0001 0023 -0003 -0004 0012 -0003(0006) (0011) (0003) (0006) (0011) (0002)

Density -0000 -0000 0000 -0000 0000 0000(0000) (0000) (0000) (0000) (0000) (0000)

Home-ownership -0474 0036 -0478 0016(0014) (0017) (0014) (0021)

K-P F-stat 3446 3630 5522Observations 2602 2602 500 2597 2597 500Data level City City State City City StateYear FE Yes Yes Yes Yes Yes YesAdjusted-R 0814 0200 0689 0811 00945 0451

Notes This table reports our estimation results of Equation 41 in Section 4 We include a full set of con-

trol variables population population growth population density home-ownership rate median household

income growth of household income median property tax poverty rate and average winter temperature

The instrument variable for the IV specifications is Democratic () Clustered standard errors are shown

in parentheses ( plt001 plt005 plt01)

45

Table 9 Tenant-Right Index and Eviction

OLS IV

(1) (2) (3) (4)Dependent Variable Eviction Eviction Eviction Eviction

Filing Rate Rate Filing Rate Rate

Tenant Rights Index -0383 -0163 -0769 -0320(0126) (0044) (0263) (0088)

Median Household Income -0189 -0067 -0154 -0052(0039) (0013) (0040) (0014)

Income Growth -0016 -0008 -0020 -0010(0013) (0004) (0013) (0004)

Share of Minority Population 0156 0029 0162 0032(0029) (0008) (0030) (0008)

K-P F-stat 12369 12502Observations 3198 3143 3190 3135Year FE Yes Yes Yes YesAdj-R 0179 0154 0165 0136

Notes This table reports our estimation results of Equation 41 in Section 4 The control variables include

population population growth population density median household income growth of household income

poverty rate and the share of the minority (non-white) population The instrument variable for the IV

specifications is Democratic () Clustered standard errors are shown in parentheses ( plt001

plt005 plt01)

46

• Introduction
• Theoretical Motivation
• • Landlords
  • Tenants
  • Rents
  • Quantities in the Rental Market
  • Nonpayment
  • Model Predictions
  • • Tenant-Right Index
    • • Index Construction
      • Descriptive and Validation Statistics
      • • Data and Methodology
        • • Empirical Models
          • Data and Summary Statistics
          • • Empirical Results
            • • Rent Affordability
              • • Baseline Results
                • Instrumental Variable Estimation
                • • Rental Housing Supply Vacancy and Homeless Rates
                  • Evictions
                  • • Conclusion