112956727 Housing Busts and Household Mobility

7/30/2019 112956727 Housing Busts and Household Mobility

1/15

FRBNY Economic Pol icy Review / For t hcoming 1

Housing Bust s andHousehol d Mobil it y:An Updat e

1. Int r oduct ion

long literature on hou sing econom ics has noted th at a

rise in m ortgage rates could lock-in an owner to his or

her current hou se, thereby slowing or pr eventing a perm anent

mo ve to a new residence if m ortgage interest rates rise

sufficiently to make the n ew debt service payment u naffordab le

(see, for exam ple, Quigley [1987, 2002]). Oth er financial

frictionssuch as the one arising from Californias

Prop osition 13 pr oper ty tax rules, which essent ially imply an

often large increase in property taxes after a movewould

have similar effects on h ou sehold m obility (Ferreira 2010).

Negative equity, by which we m ean th e current value of th e

hou se is less than th e outstan ding m ort gage balance, could also

redu ce mob ility if the owner lacks sufficient liquidity to p ay off

the full loan b alance, which is required for a p erman ent m ove

and sale of the pro perty if the bor rower is to avoid the cost of a

default (Stein 1995; Chan 2001; Engelhard t 2003).

These thr ee poten tial finan cial friction s are all associated

with th e sale of the ho use, so th ere is a tran sfer of econom ic

owner ship, not just a chan ge of residen ce. Thu s, the type of

hou sehold m obility that m ay be impacted by these frictions

involves permanent mo ves in which bot h p hysical location and

econom ic ownership chan ge for th e previous owner. Thehou sing literatu re on finan cial frictions does not ha ve clear

implications for tem porary m oves in which the own er leaves

the hou se for a period of time perhaps to rent it out and

Fernand o Ferreira is an associate professor in th e Depart men ts of Real Estate,

and Business Econom ics and Public Policy, and Joseph Gyour ko is the Mar tin

Bucksbaum Professor of Real Estate at the University of Pennsylvanias

Whar ton Schoo l; Joseph Tracy is an executive vice president and senior

advisor to t he Bank President at th e Federal Reserve Bank of New York.

Corr esponden ce: [email protected]

The authors appreciate the helpful comments of Anthony DeFusco, Andrew

Haughwout, and the referees. Fernando Ferreira and Joseph Gyourko also

than k the Research Sponsor Pr ogram of the Zell-Lurie Real Estate Center at

Wharton for financial support. The views expressed are those of the authors

and do n ot n ecessarily reflect the po sition of the Federal Reserve Bank of

New York o r th e Federal Reserve System.

The relationship between household mobilityand financial frictions, especially thoseassociated with negative home equity, hasattracted greater attention following the recentvolatility in the U.S. housing markets.

The decline in mortgage rates, along with policyinterventions to encourage historically low-raterefinancing, likewise recommend a closer look atmortgage interest rate lock-in effects, which areapt to become important once Federal Reserve

interest rate policy normalizes.

This article updates estimates in a 2010 studyby the authors of the impact of three financialfrictionsnegative equity, mortgage interestrate lock- in, and property tax lock-inonhousehold mobility. The addition of 2009American Housing Survey data to their sampleallows the authors to incorporate the effect ofmore recent house price declines.

The new studys findings corroborate the 2010

results: Negative home equity reduceshousehold mobility by 30 percent, and $1,000of additional mortgage or property tax costslowers it by 10 to 16 percent.

Fernando Ferreira, Joseph Gyourko, and Joseph Tracy

A


2/15

2 Housing Bust s and Househol d Mobil it y: An Updat e

retu rn s at a later date. Overall, m obility reflects perm anen t and

tempo rary moves, but the appro priate mobility measure

depends on the qu estion b eing addressed. Given ou r focus on

the imp act of financial frictions on h omeown ership

transitions, our p referred m easure in the an alytics reported

below reflects only perm anen t m oves as best as possible.

Interest in t he relationship between hom eowner mob ility

and financial frictions, especially friction s associated withnegative home equity, was piqued for researchers and

policymakers by the recent extraord inary boom and b ust in

U.S. hou sing m arkets. With hou se prices falling 30 percent

nat ionally, the prevalence of negative equ ity greatly expand ed

across many mar kets. Mor e recently, the sharp fall in m ortgage

interest rates and t he various policy interventions to encourage

refinancin g at historically low rat es suggest that we also need t o

upd ate our kn owledge of the imp act of mortgage interest rate

lock-in effects, as they seem likely to becom e imp ort ant afterFederal Reserve interest rat e policy nor m alizes.

Because the stud ies cited above were dated or based on

samples from specific geographic regions or population

subgrou ps, our first paper (Ferreira, Gyourko, and T racy 2010)

used th e U.S. Census Bureaus American H ou sing Sur vey

(AHS) panel from 1985-2007 to provide new and m ore general

estimates for th e nation t hat includ e all three form s of financial

frictions in the same econometric specification. Our papers

thr ee prim ary results were: 1) own ers with n egative equity were

on e-th ird less likely to move than oth erwise observation ally

equivalent own ers witho ut n egative equity; 2) for every

add itional $1,000 in m ort gage debt service costs, mob ility wasabout 12 percent lower; and 3) similar increases in p roperty tax

costs from Propo sition 13 in California also r educed m obility

by about 12 percent.

This article upd ates our previous work in two im portan t

ways. It adds data from the m ost recent AHS for 2009,

providing the first evidence from t he beginning of th e bust in

home prices in many markets. It also addresses Schulhofer-

Woh ls (2011) criticism of our samp le selection pr ocedu res

used in Ferreira, Gyourko, an d Tracy (2010). We demo nstrate

that those selection p rocedures are approp riate for studying the

effect of negative equity (an d th e other finan cial frictions

noted) o n perm anent m oves. This upd ate also document s that

our previous findings are robust to the inclusion of new data

and new m easures of perm anent mob ility, which we discuss

mo re fully below.Ou r research is related to an em erging, and potentially very

impo rtant, literature on labor econom ics investigating whether

reduced m obility among hom eowners is imp airing adjustment

in the labor market that m ight prevent the unem ployment rate

from falling as mu ch as it would o ther wise (see, for exam ple,

Aaronson and Davis [2011], Bricker and Bucks [2011],

Don ovan and Schnu re [2011], Modestino and D ennet [2012],

Molloy, Smith, and Woznak [2011], and Valletta [2010]).

Because we focus solely on h ow m obility im pacts hom eowner s,

our results do not directly address potential spillovers into th e

labor m arket. However, our finding o f a large impact o f

negative equity on owner m obility is consistent with t he

preliminary conclusion of the labor literature: There are little

or n o significant imp acts on the u nem ployment rate. As we

discuss, m ost mo ves are within a labor m arket area, so there

can be a significant d ecline in such m oves with n o effect on

access to job op portu nities in t hat area.

Mu ch work is needed to m ore fully understand the linkages

between hou sing an d labor markets on this issue. For examp le,

the likelihood that labor m arkets deteriorate along with

hou sing markets raises the po ssibility that o wners with negative

equity are not m oving in part because good job opportu nities

do n ot exist. Distinguishin g between these two poten tial causesof reduced mobility requires expanding ones theoretical and

empirical horizons to b etter control for labor market

conditions, and th at is the direction in which we urge future

research on th is topic to tur n.

Finally, reduced h om eowner m obility due to financial

frictions has econo m ic and social effects beyond its possible

ram ification s for labor mar kets. For example, locked-in o wners

are more likely to be mismatched relative to their desired

hou sing units and local pub lic service bun dle (such as schoo l

systems and the like). The utility loss just from this m ismatch

could b e significant . Wheth er owner s with n egative equity even

act like tru e owners and pro vide the positive social extern alitiesalleged for homeownership is unknown. Economically, these

owner-occupant s are renters. Mo reover, imm obility

associated with any type of friction could alter the natu re of any

hou sing recovery by shrinking the p otential trade-up market.

All of these issues require fur ther stu dy, because the evidence

suggests that negative equity in p articular is associated with

mu ch lower mo bility, and we suspect that m ortgage interest

Interest in the relationship between

homeowner mobility and financial

frictions, especially frict ions associated

with negative home equity, was piqued

for researchers and policymakers by the

recent extraordinary boom and bust

in U.S. housing markets.


3/15


rate lock-in will become m ore imp ortant in a futu re recovery.

The starting point for that con clusion is a set of robu st

estimates of mo bility effects attrib uta ble to financial frictions.

It is to that analysis that we now tu rn.

2. Financial Fr ict ions andHomeowner Mobil it y:A Br ief Review

High tr ansaction costs of buying and selling a hom e provide an

incentive for peop le to extend their stay in the ho use in ord er

to am ortize these costs over a longer holding period.

Additional financial frictions can arise that exacerbate this

effect. For examp le, Quigley (1987) exam ines the finan cial

friction from fixed-rate mor tgages in an environm ent of rising

mo rtgage rates. Ferreira (2010) an d W asi and W hite (2005)

stud y the im pact of finan cial frictions arising from restr ictionson the rate of pr operty tax increases in C alifornia un der

Proposition 13. A third finan cial friction is created when hou se

prices decline sufficiently to pu sh bor rowers into n egative

equity. Chan (2001) and Engelhardt (2003) study the imp act of

negative equity on hou sehold mob ility.

In Ferreira, Gyourko, and Tracy (2010), we estimat e the

imp act of all thr ee of these financial frictions on hou sehold

m obility using a consistent emp irical meth odo logy and d ata

that span t he 1985-2007 period. It is imp ortan t to keep in

m ind t hat each of th ese frictions applies to th e sale of the

hou se, not just to whether th e owner continu es to live there.Hen ce, we were interested in h ow these financial frictions

impact perm anent m oves that require the house to be sold.

The AHS data are well suited to ad dress this issue. The

data follow a panel of residences thr ough tim e rather than a

pan el of hou seholds. They contain in formation sufficient

for m easu rin g each of the th ree finan cial frictions as well as

other determinan ts of mo bility. A limitation o f these data,

however, is that when an owner sells a hou se and relocates, we

do n ot know where he or she moves to or th e primary mot ive

for the mo ve.

Recall that Ferreira, Gyour ko, and Tr acy (2010) estimate

large impacts of financial frictions on the perm anent mo bility

of homeowners using the AHS panel. Subsequently,

Schulhofer-Wohl (2011) uses our data an d estimation code,

but expands the definition of a m ove beyond clearly perm anentones to include any change in residence between adjacent

American H ou sing Sur veys. Schu lhofer-W ohl is correct in

observing that we und erreported o verall mob ility by censoring

these tran sitions. However, that decision was made by design in

order to d istinguish between perm anent and tempo rary moves,

as th e und erlying theor y from earlier research imp lies that it is

only with respect to perm anent moves that these potential

financial frictions should lead to lower mobility. A temporary

move reflects a situation in which an owner-occupied residence

is reported as vacant or rented for o ne or mo re surveys, with

the or iginal owner subsequently returning t o th e residence.

These moves can occur becau se a hom eowner in fact vacates his

or h er ho me t empo rarily or because vacancy status is

misreported in th e AHS data. Econom ic ownership do es not

chan ge in such cases, so th e costs associated with th e frictions

have not yet been incurred.

Nevertheless, Schulhofer-W ohls (2011) critique led u s to

develop an im proved m easure th at better exploits the pan el

structur e of the AHS to distinguish between th e two types of

mo ves. This raises our repor ted m obility rates substantially,

by more than 25 percent, but it does not m aterially affect

our find ings, as reported in Section 3. We do n ot adop t

Schulho fer-W ohls strategy of countin g all transitions fromownership to ren tal or vacancy status as perm anen t mo ves

because it dr amatically overstates their n um ber. His find ing

of a zero or a slight p ositive correlation between h om eowner

m obility and negative equ ity is likely du e largely to

conflating temporary and permanent moves.1 We show

below that over the 1985-99 period in t he AHS data, more

than 20 percent of Schulhofer-Wohls moves are temporary

in natu re, which m akes his measure prob lematic for use in

research o n lock-in effects. These tem porar y mo ves

correspond to approximately 50 percent of the additional

mo ves that Schulhofer-W ohl tallied in excess of our new,

preferred m obility measure. There is still uncertaint y aboutthe econom ic ownership of the property for the oth er

50 percent o f additional moves.

1 Schulhofer-Wohl used data and codes from our 2010 study to generate his

mobility measure, and he compared his results with our baseline measure of

mobility. He then provided his underlying code, just as we did for him. Our

discussion of his mo bility measure always applies to the first of four such

variables from his 2011 paper.

High transaction costs of buying and sellinga home provide an incentive for people to

extend their stay in the house in order to

amortize these costs over a longer holding

period. Additional financial frictions can

arise that exacerbate this effect.


4/15


Schu lhofer-W oh ls measure of m obility also can be

dynamically inconsistent, with m oves in one period r ecoded at

a later date as non m oves as addition al waves of AHS data are

includ ed in t he estimat ion sample. These issues are especially

worrisome if one is trying to u nderstand the imp act of the

recent hou sing bust on hou sehold m obility, because the errors

from conflating tempor ary and p erman ent m oves are

concentrated n ear the end o f the data, and the AHS does not yethave enou gh po st-crisis surveys to allow researchers to

distinguish between t hese types of mo ves.2

While this update highlights how no isy the data from

American H ou sing Sur veys are, we kno w of no sup erior sour ce

to u se to investigate th is issue. Given that it takes time to resolve

un certainty about whether some transitions are perm anent or

temporary in nature, there is no variable that perfectly reflects

the m obility relevant to an alysis of the im pact of financial

frictions. That includes our impro ved measure reported in th is

article. It still understates true mobility rates to the extent that

any of the moves that we censor du e to un certainty about

whether a change in econom ic ownership of the prop erty

occurr ed actu ally reflects perm anen t m oves. Precisely where to

draw th e line on this measurement issue requires careful

consideration of the costs and benefits of overstating versus

un derstating the num ber of perm anent m oves. We continu e to

advocate for a con servative coding strategy that is dynam ically

consistent o ver time, bu t th is clearly is no t costless. The next

sections detail why we came to th at conclusion .

3. Addit ional Dat a andNew Measur es of Mobil it y

3.1 Changes in the Data and Sum m aryStatistics

There are four changes to th e data used in th is upd ate of

Ferreira, Gyourko, an d Tracy (2010). The first is the addition

of the 2009 AHS samp le, which became available after we had

published our previous study. The 2009 AHS data allow

researchers to begin t o examine th e impact of th e house pricedeclines between 2005 and 2007 on h ousehold m obility from

2007 to 2009. This is straightforward , and we present an d

comp are results with an d without the new data. It does not

result in any m eaningful changes in ou r findings.3

2 The distinction b etween perm anent and temporary m oves will also be a d ata

issue for researchers using househo ld panel dat a sets, such as th e Panel Survey

on Income Dynamics. Exact property address information will be required to

reliably distinguish between t hese two types of moves.

The second chan ge involves the u se of First American-Core

Logic (FACL) r epeat-sales house pr ice indexes in lieu of th e

Federal Housing Finance Admin istration (FHFA) series when we

create instrum ents to address measuremen t error in the creation

of negative equity variables. Unlike the FHFA series, which ar e

based on ly on conformin g loans, the FACL series includ e arm s-

length pu rchases mad e with conforming and no nconform ing

loans, including subprime, Alt-A, and jumbo mortgages. Webelieve this provides a more com plete picture of what was

occurr ing in term s of local house pr ices, especially in recent years,

but t his change also has no m aterial impact on the results.4

The th ird chan ge involves additional cleaning of th e panel

structure of the AHS data. The American Housing Survey was

designed to b e used pr imar ily as a series of cross-sections rath er

than as a panel. For th is reason , a variable that we emp loy to

define th e panel structuret he pu rchase year of the hou se

was not d ependent coded.5 By that, we mean th at the

interviewer does not have access to the respo nses for th is

variable from prio r surveys, so th ere is no way at th e time of the

interview to en sure con sistent codin g across sur veys. As a

result, the pu rchase year can vary in the d ata even for the same

hou sehold. If left uncorr ected, this spurious variation in the

repo rted pu rchase year will ind uce false ho useho ld transition s.

Ferreira, Gyou rko, and Tracy (2010) developed several rules

that were used to iden tify and clean these false ho usehold

transitions in the d ata. For this update, we also include h ard-

coded edits to the p urchase year based on an inspection of the

data history for each residence, including information on the

household heads demographic characteristics. This additional

cleaning of the panel structure significantly improves on our

earlier rule-based edits.

6

The fourth and mo st imp ortant change involves the use of

an im proved m easure of m obility, which is the dependen t

variable in our an alysis. This alteration was m otivated by

Schu lhofer-W oh ls (2011) critique of our sam ple selection

pro cedur es. In Ferreira, Gyourko , and Tracy (2010), we

deliberately chose a conservative definition of what constitu ted

a m ove for the reason n oted aboven amely, theory suggests

that financial frictions involving the likes of negative equity or

mo rtgage lock-in should im pact mo bility for perm anent

3 We caution below that th is does not necessarily signal that t he estimated

relationship between mobility and negative equity during this housing market

down tur n will not chan ge as addition al AHS data become available. See thediscussion below for more on this topic.4 The FACL data used here include the impact of distressed transactions. We

have experimented with a series that does not include the data, and it does not

change our results.5 Ideally, for a residence that is own er-occupied, ch anges in the pu rchase year

coincide with changes in ownership o f the residence.6 In the curr ent work, we also follow Schulhofer-W ohl (2011) in setting tenure

to m issing whenever tenu re was impu ted by the AHS. There were 2,183 cases

in which the reported imp uted tenure was reported as owner-occupied and 458

cases repor ted as rental.


5/15


mo ves. To ensure that we did n ot m istakenly include

tempo rary moves (or false transitions attributable to an y

remaining repo rting errors in the survey), we restricted our

samp le to those observation s in which it was im med iately clear

either that th e same ho usehold resided in t he given ho using

un it across consecutive surveys (in which case, there was no

mo ve) or that a different ho usehold lived in an downed the unit

that had been owned by anoth er household in the previous

survey (in which case, there was a perman ent m ove because

both physical location an d econom ic ownership had changed).

Summ ary statistics of ou r original m obility variable, here

called MO VE, are repo rted in the first row of Table 1. This

measure is identical to the on e used in ou r 2010 paper.

Focusing initially on th e top panel, which r eports data for th e

1985-2007 period covered in th at paper, we see that 7.8 percent

of the 61,801 housing transitions used in our regressionanalysis are mo ves accordin g to this definition .7 Tho se 61,801

transitions represent only 82.7 percent of the total nu mber of

observation s poten tially available to u s.8 That is, we treat

17.3 percent of the p otential transitions as censored. In

2.4 percent of th e cases, the m ove is censored because the

7 The reported m obility rate drops from 11.4 percent in ou r previous work to

7.8 percent in this new estimation sample. This decline reflects the removal of

false mo ves as a result of the ad ditional dat a cleaning.

observation is the last in th e panel data for a part icular

residence. The remain ing cases involve tran sitions of the

property from own ership to rental or from ownership to

vacancy where it is possible that th e original owner may still

own the property.

In his first and p referred m obility measure, Schulh ofer-

Woh l (2011) effectively coun ted as a m ove all cases in which a

un it that h ad been owned in a given survey and was now beingrented o r was vacant in the subsequent survey. Using the code

he pr ovided, we created th is variable in ou r dat a. It is labeled

MO VE-ALL in t he second row of Table 1 because it captur es all

transitions, whether perman ent or tran sitory in nature. Note

the m uch higher m obility given th is definition16.4 percent

of transition s are mo ves, versus 7.8 percen t given the definition

in Ferreira, Gyourko, and Tracy (2010).9 A mu ch smaller

fraction of the dat a is censored u sing the MO VE-ALL m easure,

reflecting on ly the 2.4 percen t of cases no ted earlier in which

the observation is the final one in t he data p anel for a p articular

residence.

3.2 Two New Measures of Mobility thatExploit th e AHS Panel Stru cture

Because the conservative coding approach in our 2010 study

could result in dro pping some perm anent m oves in a

non rando m way that might affect our key estimates, we

develop an impro ved m easure of m obility that uses the AHS

panel structure to help m itigate this potential problem. This

new variable is labeled MO VE1 in Table 1. By creating it for all

cases in which th e next survey indicates that th e hou se is vacant

or r ented , we now look forward across all available surveys to

8 There are 74,774 observations on p otent ial tran sitions between 1985 and 2007

for which we have com plete data on all of the contro l variables as well as

instruments used in our regression specification reported below. The

estimation sam ple of 61,801 is nearly identical to ou r earlier estimat ion samp le

of 61,803. This reflects the fact th at the extra o bservations added t o the

estimation sample because of the cleaning of previously uncaught false

transitions in the panel structure nearly balance the number of observations

lost because of the deletion of observations with imputed tenure status.9 As we show, the MOVE-ALL measure would reflect even higher mobility if it

literally did what Schulhofer-Wohl states in his paper (2011, p. 5): As I explain

in the introdu ction, FGT [Ferreira, Gyourko, and Tracy] drop from the sample

all cases where a hou se is own er-occu pied in year tbut is vacant o r rented in yeart+2. I make only one change to FGTs data: I code those cases as moves. Our

study does not actually censor all such cases. For example, if the existing owner

were to temporar ily leave the unit vacant or rent it out an d then com e back to the

unit in a subsequent survey, our data set would n ot censor the initial observation

in that sequence. Our code would recognize that the initial observation in that

sequence was not the last one for t he given h ousehold, and we only allow m oves

for the last observation on the hou sehold. By using the code from ou r 2010 paper,

Schulhofer-Wohl effectively corrects for some temporary moves like this, so that

not every case in which a ho use is owner-occupied in year tbut is vacant or

rented in year t+2 is coun ted as a move in his data.

Tabl e 1

Mobility Measures

1985-2007

Percentage

Moved Noncensored

Percentage

Censored

MOVE 7.8 61,801 17.3

MOVE-ALL 16.4 68,206 8.8

MOVE1 10.0 63,700 14.8

MOVE2 11.0 64,450 13.8

1985-2009

Percentage

Moved Noncensored

Percentage

Censored

MOVE 7.5 66,280 17.7

MOVE-ALL 16.0 73,096 9.2

MOVE1 9.7 68,371 15.1

MOVE2 10.8 69,181 14.1

Source: U.S. Census Bureau , American Ho using Survey.

Notes: Percentage moved is computed conditional on being in our finalregression sam ple, which requires no missing data for all regressors per-

taining to household and housing unit characteristics. It is the ratio of

moves to the sum of moves and nonmoves. Percentage censored is the

ratio of censored moves to the sum of moves, nonmoves, and censored

moves.


6/15


see if the hou se again becom es owner-occupied by anoth er

hou sehold, no t just by th e previous owner. If it d oes, we not e

the year in which the ho use was pu rchased. If the pu rchase year

is between th e curren t survey year and t he next sur vey, we code

this as a perman ent m ove.

In th e example below, the first row report s the Am erican

Housing Survey year, the second indicates tenure status

(owned or r ented), and the third r eports the year the hom e waspur chased by its owner.

In t his case, the ho using unit was owned as of 2003 by som eone

who pu rchased it in 1997. The same housing un it is reported as

rented in th e next two surveys. Then, th e 2009 survey reports

the u nit as again b eing owned, with the owner havingpu rchased th e hom e in 2004. This tells us there was a

perm anent m ove by our p rior owner, with the house being sold

to a n ew owner in 2004 and t hat own er presumably renting it

out for a period of time. In o ur previous coding, situations like

this would have resulted in a censored value for our dependen t

variable in 2003, with t he observation being dropp ed from the

analysis. Our new m obility measure, MO VE1, will code th is as

a mo ve for th e 2003 observation .

We also take advantage of a variable in th e AHS that r ecords

the vacancy status of a unit ( vacancy) to h elp resolve some of

the cases censored un der th e rules creating the M OVE mob ility

indicator. For example, we code MOVE1 as indicating that am ove and sale too k place if the vacancy variable indicates that

the hou se has been sold bu t no t yet occupied (vacancy = 5).

We code MO VE1 to indicate that the or iginal owner has not

m oved if the un it is listed as being held for occasional use,

season al use, or u sual residence elsewhere (vacancy = 6-11).

Each of these instan ces suggests the pr esence of m ultiple hom es

for the household, so that one should n ot interpret a tran sition

as a perman ent m ove and sale of the pro perty. We also code

MO VE1 to ind icate that the un it has not sold if the un it is listed

as noncash rent for one or mo re surveys followed by owner-

occupied status with t he pu rchase year outside of the window

between sur vey years. Finally, we code M OVE1 to in dicate th at

a mo ve and sale have no t taken p lace if the un it is vacant for two

consecut ive surveys and listed as sold bu t no t occupied in t he

second sur vey (vacancy (t+2) = 5).

Table 1 shows that r esolution of previously censored cases in

this mann er results in 10 percent of our regression samp le

transitions now being coded as perm anent mo ves. MOVE1

Example 1

Survey year 2003 2005 2007 2009

Tenure status Own Rent Rent Own

Year purchased 1997 NA NA 2004

mob ility is much h igher than MOVE, by 28 percent, but it

rem ains well below th at for M OVE-ALL. We d iscuss the

differences across measur es mor e fully below; but first, we

introdu ce anoth er m obility variable, MOVE2.

For MOVE2, we maintain th e requirement that we are

certain th at the hou sehold has perm anently moved, but relax

the restriction th at we know that th e house has sold in t he

interval between t he relevant sur veys. Natur ally, this leads to aneven higher p ercentage of tran sitions being classified as

perman ent m oves, as indicated in th is second example.

In this case, we cann ot tell if the owner in 2003 changed residence

and sold the property between 2003 and 2005. It is possible that amo ve and sale did take place and t hat th e new owner decided to

rent ou t the property un til 2008, when the propert y was resold.

That new owner then decides to live in the property and r eports a

purchase year of 2008 in the 2009 AHS. However, it is also possible

that the owner in 2003 decided to mo ve and to rent o ut th e

prop erty, becomin g an absentee land lord. The hou se is then sold

in 2008. Since both situat ions are consistent with the reported

data, this would result in MOVE1 being censored an d recorded as

missing. However, in MOVE2 we classify this as a move in 2003

because we know that the original owner mo ved and did no t

return to t he property. Thus, MOVE2 includes cases in which we

know there was a permanent move, but cannot resolve the timing

of the sale by the original owner. The last row of the top panel of

Table 1 shows that the fraction of MOVE2 transitions is 10 percent

higher than for M OVE1 (11.0 percent versus 10.0 percent). Still,

this more expansive definition do es not generate anything close to

the level of mobility indicated by MOVE-ALL.

The bottom p anel of Table 1 report s the analogous data for

each mobility measure for th e full sample that in cludes the 2009

survey data. Note t hat mo bility is lower for each variable, which

indicates that m easured m obility declined between th e 2007 and

2009 surveys. We exploit this issue in more detail below.

3.3 Trade-Offs across Different Measuresof Mobility

Our concern abo ut Schulhofer-Wohls (2011) em pirical

strategy for t he qu estion we are add ressing is that several of the

Example 2

Survey year 2003 2005 2007 2009

Tenure status Own Rent Rent Own

Year purchased 1997 NA NA 2008

MOVE1 Censored NA NA

MOVE2 Yes NA NA


7/15


hou sing transition s that he con siders moves are false positives

in th e sense that t hey are tempo rary or r eflect coding error s in

the underlying survey. To gauge how serious the potential

pro blem is of conflating these types of mo ves, we evaluat ed th e

likelihoo d of Type I and Type II coding erro rs in his mo bility

measure by coding them in real time in the AHS data. That

is, we begin by read ing in th e cleaned p anel and selecting

observation s for 1985 and 1987. We then code MO VE-ALLbased on his code for 1985 using data from t he 1985 and 1987

surveys. These values for MO VE-ALL are saved an d t he

exercise is repeated u sing the 1987-89 p air of surveys, the 1989-

91 pair, and so o n, un til 1997-99. We end th is exercise in 1999

to en sure that we have enou gh futur e sur veys to assess wheth er

Schulhofer-Wohls moves turn ed out to be perman ent or

temporary. We call this real-time version of the Schulhofer-

Woh l m obility m easure MOVE-ALLR.

It is imp ortant to n ote that th e coding of MOVE-ALLR in

this real-tim e analysis differs from the co ding o f MOVE-ALL in

the estimation sample. Our t hird examp le illustrates why.

When the 2003 AHS data are added to the estimation

sample, MOVE (and o ur t wo other m obility measures),

MO VE-ALL, an d MO VE-ALLR for 2003 will all be censored

because at that tim e this is the last observation in th e panel for

the residence. When the 2005 AHS data are added, MOVE for

2003 will remain censor ed an d M OVE-ALL and MO VE-ALLRfor 2003 will be recoded as a m ove. However, when th e 2007

AHS data are m erged into t he samp le, MO VE for 2003 (as well

as MOVE1 and MO VE2) will be recoded from censored to a

no nm ove, while MO VE-ALL for 2003 will be recoded fro m a

mo ve to a nonm ove and MO VE-ALLR for 2003 will remain

coded as a mo ve (since we do no t allow the real-time m easure

to be r ecoded once it ind icates that a m ove has taken p lace).

The reason for th e recoding of MOVE and M OVE-ALL is that

when con structing these mobility measures, we sort t he data by

residence, household (based on a un ique hou sehold

ident ification n um ber we create), and su rvey year. Based on th e

sorted d ata, a move is only con sidered for the last observation

for that hou sehold. As a result, our coding strategy for MOVE

(as well as for M OVE1 and M OVE2) on ly recodes censored

observations as either n onm oves or m oves and it never recodes

non censored mob ility observations. In contrast, the coding for

MO VE-ALL can be d ynam ically inconsistent o ver time, with

mo ves recoded at a later date as non mo ves. By construction,

Example 3:

Survey year 2003 2005 2007

Tenure status Own Rent Own

Year purchased 1997 NA 1997

MOVE-ALLR maintains dynam ic consistency by not r ecoding

a move as a non mo ve even when inform ation becomes

available indicating that t he original owner has retur ned.

The top panel of Table 2 reports cross-tabulations of our

MO VE2 indicato r, which takes full advantage of the pan el to

differentiate between perm anent and t empor ary transitions,

and M OVE-ALLR.10 We u se MOVE2 for this analysis since our

focus here is whether a m ove is perm anent or n ot, regardless of

when th e property was sold. The first column of the table

docum ents that th ese two mo bility variables confirm th at therewere 70,707 cases in wh ich no m ove occurr ed. There are no

cases in which our MO VE2 m easure con sidered some

transition a move when MOVE-ALLR did n ot (th at is, there is

no eviden ce of Type II erro rs); nor is MOVE2 ever censored o r

missing when M OVE-ALLR indicates that n o m ove took place.

The tables second colu m n is more int eresting because bot h

mo bility measures have 8,550 mo ves, but M OVE-ALLR has an

add itional 8,607 moves. Mo reover, 41.3 percen t (3,557/8,607)

of the addition al moves in MO VE-ALLR turn out to be

tem por ary in natu re because they reflect Type I errors. That is,

using th e full pan el of surveys up to 2009, we observe the owner

return to the u nit at some point in the future, or the surveys

reflect some ot her trait th at leads us to conclude that t here has

not been a permanent m ove.11

10 Here, we use all available transition s from th e AHS for own er-occupied

residents between twen ty-one an d fifty-nine years of age over the 1985-99

period and d o not restrict the observations to tho se with no nm issing values for

all of the regressors that we u se in the final mo bility estimation .

Tabl e 2

Permanent versus Temporary Moves

Cross-Tabulation of MOVE2 with MOVE-ALLR

MOVE-ALLR

0 1

MOVE2 0 70,707 3,5571 0 8,550

. (m issing) 0 5,050

Percentage of False Positives Resolved over Time

Four years or first subsequent survey 66.0

Six years or second subsequent survey 17.4

Eight years or th ird subsequent survey 7.7

Ten years or fourth subsequen t survey 4.7

Twelve years or fifth subsequent survey 1.9

Fourteen years or sixth subsequent survey 1.2

Sixteen-plus years 1.1

Source: U.S. Census Bureau, Am erican H ousing Survey, 1985-2009.

Note: 15.1 p ercent of false po sitives are resolved using vacancy statu s.


8/15


Ou t of all the false positives from MO VE-ALLR, in two-th irds

of the cases the Type I erro r could be elimin ated by looking at

only one subsequent American H ousing Survey, as shown in the

botto m p anel of Table 2. To better un derstand th is, presum e that

we are un certain about wheth er a transition in the 1985 data is

perm anent or tem porar y. That is, the data clearly show a given

owner-occup ant in 1985, but a different occupan t or a reported

vacancy in 1987. In 66 percent of cases, the 1989 survey fullyresolves the u ncert ainty. In these false positive cases, we see the

same hou sehold living and owning the same un it in 1985 and

1989. Another 17.4 percent o f the false positives are resolved by

the next available survey (that is, after six years have passed), so

that mo re than 83 percent of cases are clarified by 1991 in th is

examp le. The rem ainin g cases are clarified by futur e surveys,

with some own ers being absent for lon g periods of time.

Ho wever, the n um ber of th ose cases is quite small.12

It is also import ant to note t hat for 5,050 transitions,

MO VE2 is assigned a censored value while MO VE-ALLR

considers th em m oves. Wh ile no ne of these cases can be

definitively identified as perman ent m oves with th e curren tly

available data, some of th em u ndo ubtedly are and will be

revealed and coded as such over tim e as add ition al sur vey data

becom e available. In practice, this means that MO VE2 still does

not include all true perm anent moves. This highlights the fact

that there is no perfect m easure o f such mo bility as long as the

data do not allow for the imm ediate recognition of whether an

econom ic change in ownership has occurred.

4. Result s

4.1 Estimation Method ology

In Ferreira, Gyourko , and Tracy (2010), we sho wed that each of

ou r finan cial friction variables, which are ba sed on self-

reported values, is subject to substantial measurement error

that causes severe attenuation bias in estimated m obility

11 As noted above, the lack of dependent coding for this variable means th at some

of these cases could be attributable to coding error by the AHS survey taker in the

sense that he or she does see or int erview the or iginal owner an d m istakenlyconcludes that the un it is not occupied by the same person. T he best examp le of

this involves units described as being vacant and held for occasional or seasonal

use. This group represents 14 percent of the 3,557 cases. There is a mu ch smaller

fraction of units (1.2 percent) for which there is noncash rent and a subsequent

sale outside th e relevant sam ple interval. There is an even smaller share of units

(0.3 percent) that are vacant across two consecutive surveys, with the second

survey listing the h ousing unit as sold but not yet occupied.12 Subsequent to a tem porary move, the mean (m edian) duration of the owner

in the residence is 6.1 (5.0) years. In 38 percent of cases, the post-tem porar y

move dur ation is censored by the end of the data in 2009.

effects.13Such m easurem ent error can be m itigated by using aninstrum ental-variable appro ach.14 In th e case of hou se equity

variables, we use the pu rchase price of the hou se and an y house

price appreciation implied by the First American-Core Logic

repeat-sales house price index for the r elevant metrop olitan

area in order to calculate our instrum ent for the self-reported

measur e of negative equity. The instru m ental variable for

mo rtgage lock-in is based on th e average rate on th irty-yearfixed-rate mo rtgages durin g the year in which th e hou se was

purchased for th e self-reported interest rate. The real annu al

differen ce in m ortgage paymen ts is calculated using the

difference between this rate and the prevailing mortgage rate

variable. In bo th cases, our instrum ent relies on the intu ition

that aggregate information averages out individual-level

measurement error.

The Proposition 13 property tax subsidy variable is

constructed from two self-reported variables. To ad dress the

likely measurement error, we create an in strum ent defined as

the difference between th e growth in th e metro politan area

repeat-sales house price index and the m aximu m allowable

growth in th e prop erty tax over the same period, all multiplied

by the fully assessed pr oper ty tax on th e pur chase value of the

hou se. Needless to say, the value of the im plied subsidy still is

zero for non -California h ouseholds.

To accomm odate our data structure, we use a recursive

mixed-process model that expands upo n the classic mo bility

specifications intro duced by Han ushek and Quigley (1979) an d

Venti and Wise (1984), which also served as the found ation for

ou r earlier empirical work. The following fou r-equ ation system

describes our m obility outcome and ou r three instrumen tal

variables:

if

0 otherwise

if

0 otherwise

13 Kain and Q uigley (1972) is the seminal work on this issue. Mo re recently,

Bayer, Ferreira, and McM illan ( 2007) observe that self-reported values are less

accurate the longer ago the occupant moved in. Hence, wide swings in prices

like those seen over ou r samp le period increase the dispersion of self-repor ted

home values. Schwartz (2006) also reports measurement error in interest rates.14 See Ashenfelter and Krueger (1994) for a classic reference on ho w to create

an alternative measure of the treatment variable of interest, and then to use

that measure as the instrumental variable.

Imi Xi P13 XP13i FR M XF R M i NINi1

1i+ + + +=*

XP13i Xi P13ZP13i FR MZF R M i NINi2

2i+ + + +=

XF R M i Xi P13ZP13i FR MZF R M i NINi2 3i+ + + +=

INi1 Xi P13ZP13i FR MZF R M i NINI

2 4i

+ + + +=*

Imi 1= Imi 0*

INi1 1= INi

1 0*


9/15


, where ,

where is our observed mo bility indicator, a continu ous

latent index for the prop ensity to move, our negative equityindicator based on the self-reported hou se value, our

alternative negative equity indicator based on the m etro area

hou se price index, a contin uou s latent index for whether the

borrower is in negative equity, our instrument for the annu al

property tax cost of moving attributable to Proposition 13 for

California residents, and our instrum ent for the ann ual

interest rate cost associated with refinancing for households

with a fixed-rate mortgage.

We estimate this system using Roodman s Cmp program in

STATA. A description o f the pr ogram , its implem entat ion, and

applications is given in Rood man (2009). For com parison with

ou r earlier findin gs, we also p resent results for a single-

equation Probit (u sed in Ferreira, Gyourko, and Tracy [2010])

and a standard linear-probability model.15

4.2 Negative Equity

In th is section, we first present updat ed results on the relationship

between mob ility and n egative equity using n ew data from the

2009 AHS and for th e five different m obility variables described

above. For the rest of ou r discussion, we code M OVE-ALLR for

the full sample period from 1985 to 2007 or to 2009. Table 3

begins by providing summ ary statistics on the distribu tion o f self-

reported negative equity according to whether th ere was a mo ve.

Table 4 then r eport s the results of re-estimatin g the core m obility

specification from Ferreira, Gyourko, and T racy (2010) using the

five mob ility measures described above as the depen dent variable.

The top pan el of Table 4 reports mar ginal effects from that

specification estimated with the cleaned and edited AHS data

from 1985 to 2007. Results for the expanded 1985-2009 AHS data

are reported in the bottom pan el.

15 Schulhofer-Woh l (2011) correctly notes that ou r n egative equity indicator was

a dichotomou s dumm y and thus did not have the requisite properties for theIV Probit estimation procedur e as carried out in our 2010 study. Consequently,

our main results of this update are based on th e IV Probit m arginal effects from

the joint estimation of the four-equ ation system o utlined above. For comparison,

we also report estimates from a single-equation IV Probit (u sed in our p revious

paper) as well as an IV linear-probability version of the mod el, with those results

reported in th e second and third colum ns of Table 4. Schulhofer-Wohl does not

instrum ent for the measurem ent error. As our paper showed, there is never any

significant corr elation between a financial friction and perman ent m oves unless

attenuation bias is dealt with in som e fashion.

1i

2i

3i

4i

N 0

1 12

13

14

22

2324

32

34

l

=

Imi

IMi*

INi1

INi2

INi1*

ZP13i

ZF R M i

Focusing first on the m ulti-equation Pr obit m arginal effects

in co lum n 1, we observe a statistically significant n egative

relationship between the presence of negative equity and

mo bility for o ur o riginal MOVE indicato r as well as for our

impr oved MOVE1 indicator. For ou r earlier sample period

from 1985 to 2007, our preferred M OVE1 indicator implies

that negative equity is associated with a two -year mo bility rate

that is 3 percentage points lower, ceteris paribu s. This is

30 percent of the baseline mobility rate of 10 percent, which is

similar to the relative impact reported in Ferreira, Gyourko,and Tr acy (2010). The MO VE variable used in ou r earlier paper

generates a slightly larger imp act, but it is not statistically or

econom ically different from that for MO VE1. The m ore

expansive definition of permanent mobility reflected in

MO VE2 yields a slightly lower mar ginal effect of 2.8 percent age

points, or abou t on e-fourth o f the baseline m obility rate. It is

different from zero at a 10 percent confidence level for th e

Tabl e 3

Cross-Tabulations of Negative Equityand Mobility Indicators

Negative Equity

Mobility Indicator No Yes

MOVE

No 74.02 2.11

Yes 6.05 0.15

Censored 16.22 1.46

MOVE1

No 74.51 2.14

Yes 8.04 0.23

Censored 13.74 1.34

MOVE2

No 74.51 2.14

Yes 8.99 0.28

Censored 12.79 1.29

MOVE-ALL

No 74.02 2.11Yes 14.15 0.51

Censored 8.12 1.09

MOVE-ALLR

No 68.60 1.91

Yes 14.71 0.57

Censored 12.98 1.23

Source: U.S. Census Bureau, Am erican Ho using Survey (1985-2009).

Notes: Negative equity is based on self-reported hou se values.

MOVE-ALLRis the real-time calculation o f MOVE-ALL over the full

sample period in wh ich we do no t allow moves to be subsequently

recoded as nonmoves. Cell percentages are shown.


10/15


1985-2007 sam ple, and we cann ot reject the nu ll hypot hesis

that the effects are the sam e across all three m easures.

A comparison of results across column s in the to p pan el of

Table 4 indicates that im plied marginal effects from the m ulti-

equat ion Pr obit specification ar e consistently lower than effects

from th e single-equation Pr obit and the linear-probability

specifications, although the pattern of findings is quite

consistent. In add ition, the standard error s are such th at we

cann ot con clude th at th e levels of the im plied effects differ by

estimation strategy.

The first column of Table 4s second pan el adds in the dat a

from the 2009 survey. We find m odestly lower m arginal effects

here com pared with th e 1985-2007 results, and negative equity

is no longer associated with statistically significant lower

mobility for the MOVE2 variable. However, these marginal

effects are no t significantly different from tho se of the earlier

sample period, so there is no evidence yet that the m ost recent

housing bust has materially changed the relationship between

negative equity and o wner m obility. That said, on e cannot and

should no t conclude th at the r elationship will not chan ge overthis cycle as mo re data b ecom e available, as cautioned in ou r

original paper. The pr evious section im plies that it takes four to

six years for the vast major ity of the censor ed ho using

transitions to be resolved. Hence, it will be much later in this

decade before we can m ore confidently know how n egative

equity affected perm anent mo bility in this latest down turn .

No te that th e coefficient on the MO VE-ALL ind icator as

constructed by Schulhofer-Wohl (2011) suggests a positive

correlation between negative equity and mob ility. In n either

sample period is this statistically different fro m zero , but th e

poin t estimates are po sitive, not n egative. The m isclassification

of so many tem porary m oves as permanen t on es is likely to be

critical here. Recall that theo ry does no t suggest a n egative

correlation between tempo rary mo ves and negative equity.

Hence, it should not be surprising to find a weak and im precise

correlation when m ore than o ne-fifth of the coded m oves may

not involve a perman ent m ove and sale of the home.16

This intu ition that th e conflation of tempor ary and

perman ent moves is the driving factor behind the d ifference

between o ur negative equity results and th ose reported by

Schulhofer-Wohl is corroborated by comparin g the different

estimates associated with MO VE-ALL and MO VE-ALLR.

Recall that th e distinction between t hese two measures is thatMOVE-ALLR retains moves identified by Schulhofer-Wohl

that are kn own ex post to be temp orary, whereas MOVE-ALL

allows these temporary m oves to be recoded as no nm oves.

Retaining these tempo rary m oves increases the measured

mob ility rate from 16.1 percent for MO VE-ALL to 17.8 percent

for M OVE-ALLR. The estimates in Table 4 indicate that th e

inclusion of these additional temporary moves raises in each

case the estimated p ositive effect of negative equity on m obility.

Of course, the un derlying sample used in generating these

estimates is the result of censorin g all cases in wh ich we cann ot

tell whether physical location an d economic own ership

16 We also estimated all mo dels with the original FHFA price series used to help

determ ine negative equity. Focusing on th e system IV Prob it results, we not e

that MO VE-ALL rem ains positive but is still statistically insignificant. MO VE

contin ues to b e positive and statistically significant. Th e m arginal effects for

MOVE1 and MOVE2 decline by aroun d 25 percent for the 1985-2007 samp le

and aroun d 40 percent for the 1985-2009 sample, and th ey are no longer

statistically significant. This dro p in the m agnitude of m arginal effects likely

reflects the inability of the FHFA ho use price indexes to accurately track the

declining prices due to the ind exes narr ow focus on ho uses finan ced with

conforming mor tgages.

Tabl e 4

Empirical Estimates

1985-2007

IV Probit

(Multi-Equation)

IV Probit

(Single-Equation)

IV Linear

Probability

MOVE -0.043** -0.050** -0.062**

N=61,801 (0.012) (0.014) (0.017)

MOVE1 -0.030** -0.047** -0.056**

N=63,700 (0.014) (0.016) (0.019)

MOVE2 -0.028* -0.047** -0.043**

N=64,450 (0.015) (0.020) (0.020)

MOVE-ALL 0.019 0.029 0.029

N=68,206 (0.021) (0.024) (0.024)

MOVE-ALLR 0.029 0.063** 0.061**

N=64,181 (0.021) (0.029) (0.029)

1985-2009

MOVE -0.037** -0.046** -0.054**

N=66,280 (0.011) (0.017) (0.016)

MOVE1 -0.024* -0.044** -0.048**

N=68,371 (0.014) (0.016) (0.018)

MOVE2 -0.022 -0.037** -0.036*

N=69,181 (0.014) (0.017) (0.019)

MOVE-ALL 0.027 0.032 0.035

N=73,096 (0.018) (0.023) (0.023)

MOVE-ALLR 0.037* 0.066** 0.066**

N=69,079 (0.020) (0.027) (0.027)

Source: U.S. Census Bureau , American Hou sing Sur vey.

Notes: Probit m arginal effects are average differences. Stand ard erro rs are

in parentheses. MOVE-ALLR is the real-tim e version of M OVE-ALL over

the full sample period in wh ich we do not allow m oves to be subse-

quently recoded as nonmoves.

** Statistically significant at the 95 percent confidence level.

* Statistically significant at the 90 percent confidence level.


11/15


chan ged. That is rou ghly half of the excess moves in MO VE-

ALLR relative to M OVE2 based on ou r real-tim e analysis of the

1985-99 period. Pr actically speaking, m ost of the censor ed

cases in ou r full data set are from recent waves of the AHS, and

Table 2s results suggest t hat if past pat tern s persist, the vast

m ajority will be resolved within four t o six years. Ho wever, it

seems likely that at least some of th e cases in which t he pr evious

owner is coded as no longer living in the u nit over m ultiple

surveys, but for which t here is still no clear evidence of a sale,

actually are perman ent m oves.17

17 This raises the question of whether we could im prove the m easure M OVE2

by count ing as moves situation s in which it seems likely (bu t not certain) that

a perman ent m ove has taken place. Intuition m ight suggest that t he longer the

ownership gap observed in which the residence is reported as rental or vacant,

the m ore likely that the previous owner will not retur n. To check on this

possibility, we looked at own ership gaps of different lengths and com put ed the

fraction of cases in which the m ove turned ou t to be temporary, conditional on

having the inform ation to m ake this determination. For situations in which the

residence was rented or vacant for at least three surveys, the tran sition tur ned

out to be temporary in 59 percent of the cases in which we could determine the

final outcome. If we lengthen the ownership gap to four or more surveys, the

percentage of tempor ary moves actually increases to 62 percent. Th is patterncontin ues for ownership gaps of five or m ore and six or move surveys. Thu s,

the simple intuition that the longer the current o wnership gap, the more likely

the m ove will turn out to be perman ent, is not supported in th e data. For this

reason, we do not think on e can improve on MOVE2 by recoding censored

tran sitions as moves given an own ership gap of some specified length.

Ho wever, it is still useful to un derstand that th e potent ial fragility of our results

(and possibly of previous researchers) arises from the fact that it is difficult to

properly measure mobility in a number of cases.

4.3 Fixed-Rate Mortgages and Property TaxLock-Ins

Upd ated results on th e impact of two additional finan cial

frictions on h ousehold m obility are presented in Table 5.

The first friction pertains to hom eowners with a fixed-rate

mo rtgage. In a rising interest rate environ men t, if a

hom eowner with this type of mor tgage moves, the m onth ly

cost of an ident ically sized mor tgage can b e higher. The second

friction pertains to hom eowners in Californ ia whose propertytax increases have been limited over time d ue to Proposition

13. If the hom eowner m oves to a similarly valued pro perty,

taxes would be set t o th e fully assessed value of th e ho use. In

bot h cases, we examin e the m arginal effect of an addit ional

$1,000 annual cost on th e likelihood that th e household m oves.

We provide estimates for specifications containing our two

impro ved mobility indicators for the expanded sample period,

in which we use the FACL overall house prices to up date ho m e

values. The data confirm ou r earlier finding that both frictions

give rise to reduced hou sehold m obility10 percent to

16 percent less per $1,000 using our preferred mobility

measure MO VE1. In non e of the specifications do th e datareject the n otion that th e mo bility friction is the same whether

it is generated b y rising rates for fixed-rat e borr owers or higher

prop erty taxes for California hom eowners.

We suspect th at th is interest-raterelated lock-in effect

will become increasingly imp ortant as mon etary policy is

norm alized in th e future. To illustrate, we consider th e

Tabl e 5

Impact of Other Financial Frictions on Household Mobility

IV Probit (Mu lti-Equ at io n) IV Probit (Sin gle-Equ at io n) IV Lin ear Pro bability

Mobility indicator: MOVE1

Fixed-rate m ortgage lock-in ($1,000) -0.016** -0.018* -0.013

(0.009) (0.009) (0.009)

Proposition 13 property tax lock-in ($1,000) -0.010** -0.010** -0.008**(0.005) (0.004) (0.004)

Mobility indicator: MOVE2

Fixed-rate m ortgage lock-in ($1,000) -0.023** -0.024** -0.019**

(0.009) (0.009) (0.009)

Proposition 13 property tax lock-in ($1,000) -0.009* -0.009* -0.008**

(0.005) (0.005) (0.004)

Source: U.S. Census Bureau , American Ho using Survey, 1985-2009.

Note: Probit m arginal effects are average derivatives, with standard errors in p arenth eses.

** Statistically significant at the 95 percent confidence level.

* Statistically significant at the 90 percent confidence level.


12/15


hypot hetical case of a 250 basis poin t increase in th e average

thir ty-year fixed-rate m ort gage interest rate as a result of the

nor malization o f mon etary policy. For hom eowners in 2009

with a fixed-rate m ortgage, this results in a m ean (m edian)ann ual paym ent d ifference of $2,300 ($1,710). Accord ing to

the Pro bit mar ginal effects for MO VE1, this imp lies a mean

(m edian) reduction in the two-year m obility rate of

3.7 (2.7) percentage points. If we calculate using th e estimates

for MOVE2,we obtain a reduction in the two-year mo bility

rate of 5.3 (3.9) percentage points. This suggests that as

negative equity (hop efully) d iminishes in impor tance over the

com ing years, it well may be offset by an increasing fixed-rate-

mo rtgage friction.18

5. Spil l over s int o t he Labor Mar ketand Ot h er Impl icat ions

Policymakers naturally have been interested in whether

reduced m obility amon g homeowners (from negative equity

especially) might be p laying a role in what has hereto fore been

a very sluggish em ploymen t recovery. Perhaps being stuck in

on es hom e because of th e high costs of curin g negative equity

prevent s a sufficiently large nu m ber of people from m oving to

accept jobs, which affects the measured u nem ployment r ate.

Our analysis is restricted to the h ousing m arket because the

AHS follows residences rather th an ho usehold s and t herefore it

is no t suited to add ressing job mo bility. Ho wever, the

18 This is particularly true for borro wers who received a below-m arket

mor tgage rate throu gh a private modification or a Hom e Affordable

Modification Program m odification (cond itional on the borrower not

redefaulting on the modified mortgage). If these low-rate mortgages were

either assumable or portable, there would be no associated mobility friction.

preliminary answer on this question from th e initial set of

research in labor econ om ics is no. Since long-distan ce moves

are mo re likely to be job related, these stud ies tend to focus on

mo ves across states or cou nt ies.19 The AHS files are also usefulfor examin ing the types of moves likely to be im pacted b y

hou sing market friction s. For examp le, the AHS asks recent

movers (that is, those who m oved within th e last two years)

about t he prim ary reason for their move and, u ntil 1995, the

distance of the m ove. A high per centage of moves

73 percentare local, while only 13 percent cross a state

border. Table 6 provides more detail on t he prim ary reason for

mo ves, both overall and broken d own by distance. Most moves

are for qu ality-of-life, person al/family, and financial reasons,

and d o no t app ear to be prim arily job related. This is especially

true for local moves. In contrast, longer-distance moves,

part icularly across states, tend to b e job related. One

imp lication of these data th at is consistent with th e initial labor

ma rket an alysis results is that financial frictions affecting

hou sehold m obility may well be mor e likely to red uce local

mo ves that need n ot h ave significant spillover effects into th e

labor m arket. Nevertheless, it is too early to conclu de that t his

is the final word on poten tial spillovers into the labor m arket.

That co nclusion shou ld await a fuller recovery as well as

confirming evidence from studies using micro d ata and

mo deling individual household behavior.

We em phasize that even if reduced mob ility attributable to

financial frictions has no spillovers into the labor market, thatdoes not m ake them econom ically unim portan t. The fewer

19 Several of these papers (for instan ce, Aaronson and D avis [2011], Mod estino

and Dennett [ 2012], and Molloy, Smith, and Woznak [2011]) also estimate

aggregate models of migration rates rather than micro models of whether a

hou sehold moves. Dono van and Schnur e (2011) also pursue an aggregate-level

analysis, but theirs is mor e comp rehensive in the sense that it investigates the

impact of negative equity within and across counties (and also within and

across states).

Tabl e 6

Main Reason for Move: Overall and by Distance of Move

1985-95

Reason 1985-2009 All Sam e Metropolitan Statistical Area Sam e State Different State Out of Country

Job-related 12.58 13.23 3.85 21.20 60.53 66.10

Quality-of-life 26.70 23.94 26.67 24.97 8.18 3.39

Personal/fam ily 23.88 20.44 19.73 16.64 10.22 6.78

Financial 21.83 25.55 33.00 20.55 4.25 6.78

Other 11.84 13.18 12.90 13.13 14.94 15.25

All equal 3.17 3.67 3.85 3.51 1.89 1.69

Sources: U.S. Census Bureau, Am erican H ousing Survey; autho rs calculations.

Note: The sample is restricted to owner-occupied respondents between the ages of twenty-one and fifty-nine.


13/15


within-metro politan-area mo ves that we see due to n egative

equity have direct effects on own er econo mic welfare and

potent ially impo rtant im plications for th e natur e of the

hou sing sector recovery. Being locked into on es curren t

residence because of the high costs of curin g negative equity

means th e hou sehold is imp erfectly matched in its residence.

The welfare losses from being mism atched ar e not just from

having the wron g-sized hou se (such as not enough bedr oom snow th at there is an add itional child), but also from being in

the wron g location . Many families, for instan ce, m ay not be

able to m ove to their p referred school district, even if ther e is

no desire to change jobs.

In add ition to these welfare consequences are the poten tial

impacts on th e scale and intensity of trade-up (and trade-

down ) pu rchases. There are vastly m ore sales of existing hom es

than new ho mes in a t ypical year, so lower tran saction levels in

the existing stock mat erially affect the state of th e hou sing

market, including the incomes of realtors and others who work

in th e hou sing sector an d in dur able goods sales that coincide

with tu rn over of owned h ou sing, as well as the finan ces of

man y state and local governmen ts that rely on transfer taxes.20

Finally, it is nat ur al to focus on t he pot ential ram ifications

of lower m obility due t o n egative equity, but we shou ld no t

forget that the mortgage interest rate lock-in effect could

become mu ch more impo rtant in the future. We find

econo m ically meanin gful int erest rate lock-in effects in p ast

cycles, and the stage is set for th em t o becom e emp irically

relevant. Federal Reserve inter est rate policy and oth er pu blic

policies have been successful at en cour aging refinancing at

historically low rates. When rate p olicy no rm alizes, we may

find m any owners constrained from moving because of themu ch higher debt service payments they would incur from

buying a different ho me.

6. Summar y an d Impl icat ionsfor Fut ur e Resear ch

Our inclusion of th e most recent American H ousing Survey for

2009, which reflects initial data from the recent h ou sing bust,

does not materially change previously reported estimates of

how n egative equity and o ther finan cial frictions are correlated

with hom eowner m obility. Hom eowners with n egative equityremain about one-th ird less likely to m ove than o therwise

observationally equivalent owners. However, the uncertainty

surroun ding changes in economic ownership involving various

tran sitions concen trat ed in t he last few surveys suggests that we

cannot really know for sure h ow the recent h ousing bust

impacted perm anent m obility until a few years into the futur e.

Then, the additional survey data will reveal the tru e natu re of

man y of those transitions.

A critique of th e sample selection pr ocedures used in o ur

earlier work (Ferreira, Gyourko , and Tr acy 2010), which claims

to reverse this result, appears largely du e to th e incorrect

classification o f man y transition s as m oves that are likely to b e

tempo rary and n ot perm anent, or simply reflect coding error in

the individual surveys. Whether negative equity can be

positively associated with tem por ary mo ves is a question that

we did not attemp t to answer then. That said, our im proved

measur e still does not reflect m obility perfectly because of ou r

conservative policy of censoring transitions that cannot be

definitively defined as permanent in nature. Hopefully,

researchers will develop o ther d ata sour ces or ways to redu ce

this noise in th e AH S pan els.

Going forward, it is more im portan t for scholars to tackle

the qu estion of whether this correlation is causal in n ature.That will requ ire new theoret ical and em pirical strat egies to

better contro l for labor m arket conditions. As long as labor an d

hou sing markets mo ve together (and there is soun d reason

concep tually and emp irically to believe they do) , the

correlation docum ented here could be driven predo minan tly

by the lack of good job op portu nities to attract po tential

mo vers. Un til we addr ess this issue, we will no t kno w the tru e

social cost of highly leveraged ho m e pur chases that are m ore

likely to lead to n egative equity situation s.

20Low transaction volumes in housing markets also complicate the appraisal

pro cess because of a lack of com parables. This likely leads to conservative

appraisals and therefore the need for households to make larger down-

payments in order to purchase a home. This creates the possibility of an

adverse-feedback effect th at can fur ther r educe hom e sales.


14/15

Refer ences


Aaronson, D., and J. Davis. 2011. How Mu ch Has H ouse Lock

Affected Labor Mob ility and the U nemp loymen t Rate? Federal

Reserve Bank o f Chicago, Chicago Fed Let t er no. 290,

September.

Ashenfelter, O., and A. B. Krueger. 1994. Estimates of the Economic

Returns to Schooling from a New Sample of Twins. Amer ican

Economic Review 84, no. 5 (December): 1157-73.

Bayer, P., F. Ferreira, and R. McMillan. 2007. A Unified Framework

for Measuring Preferences for Schools and Neighborhoo ds.

Jour nal of Polit ical Economy 115, no. 4 (August): 588-638.

Bricker, J., and B. Bucks. 2011. Negative Hom e Equity, Econ om ic

Insecurity, and H ousehold M obility: Evidence from th e 2007-

2009 SCF Panel. Paper presented at the Federal Reserve Bank

of St. Louis Applied M icro Econom ics Conference, St. Louis,

Missouri, May 5-6.

Chan, S. 2001. Spatial Lock-In: Do Falling House Prices Constrain

Residential M obility? Jour nal of Urban Economics 49, no. 3

(May): 567-86.

Donovan, C., and C. Schnure. 2011. Locked in the H ouse:

Do Un derwater Mor tgages Reduce Labor M arket Mobility?

Unp ublished paper, May 31.

Engelhardt, G. V. 2003. Nomin al Loss Aversion, Ho using Equity

Constraints, and Household Mobility: Evidence from the United

States. Jour nal of Ur ban Economics 53, no. 1 ( January): 171-95.

Ferreira, F. 2010. You Can T ake It with You: Proposition 13 Tax

Benefits, Resident ial Mo bility, and W illingn ess to Pay for Ho usin g

Amenities. Jour nal of Publ ic Economics 94, nos. 9-10

(October) : 661-73.

Ferreira, F., J. Gyourko, and J. Tracy. 2010. Housing Busts and

Ho usehold Mo bility. Jour nal of Ur ban Economics 68, no. 1

(July): 34-45.

Hanushek, E. A., and J. M. Quigley. 1979. The Dynamics of the

Ho using Market: A Stock Adjustment M odel of Housing

Consumption. Jour nal of Urban Economics 6, no. 1

(January): 90-111.

Kain, J. F., and J. M. Quigley. 1972. Note o n Owners Estimate of

Ho using Value. Journa l of t he American St at ist ical

Associat ion 67, no. 340 (D ecember) : 803-6.

Modestino, A. S., and J. Dennett. 2012. Are American H om eowners

Locked into Th eir Hom es? The Im pact of Hou sing Market

Cond itions on State-to-State Migration. Federal Reserve Bank of

Boston W orking Paper no. 12-1, February. Available at ht tp://

www.bostonfed.org/economic/wp/wp2012/wp1201.pdf.

Molloy, R., C. L. Smith, and A. Woznak. 2011. Internal Migration in

the U nited States. Jour nal of Economic Per spect ives 25, no. 3

(summer): 173-96.

Quigley, J. M. 1987. Interest Rate Variation s, Mortgage Prepaym ents,

and H ousehold Mobility. Review of Economics an d

St at ist ics 69, no. 4 (No vember): 636-43.

. 2002. Hom eowner Mo bility and M ortgage Interest Rates:

New Evidence from t he 1990s. Real Est at e Economics 30,

no. 3 (fall): 345-64.

Roodman, D. 2009. Estimat ing Fully Observed Recursive Mixed-

Process Models with Cm p. Center for Global Development

Wor king Paper no . 168, April.

Schulhofer-Wohl, S. 2011. Negative Equity Does Not ReduceHo meown ers Mob ility. NBER Working Paper no. 16701,

January. Available at http://www.nber.org/papers/w16701.

Schwartz, A. 2006. Household Refinancing Behavior in Fixed-Rate

Mor tgages. Unp ublished paper, H arvard Un iversity, November/

December.

Stein, J. C. 1995. Prices and Trading Volum e in the H ousing Market:

A Mod el with D own-Paym ent Effects. Quar t erl y Journal of

Economics 110, no. 2 (May): 379-406.

Valletta, R. G. 2010. House Lock and Structu ral Unemp loyment.Paper p resented at th e Federal Reserve System Co mm ittee on

Regional Analysis, New Orleans, Louisiana, November 14-16.


15/15

Refer ences (Cont inued)

FRBNYEconomic Pol icy Review / For t hcoming 15

Venti, S. F., and D. A. Wise. 1984. Moving and H ousing Expenditure:

Transaction Costs and Disequilibrium. Jour nal of Public

Economics 23, nos. 1-2 (Febru ary-March): 207-43.

Wasi, N., and M. J. White. 2005. Property Tax Limitations and

Mobility: Lock-In Effect of Californias Proposition 13.

Br ookings-Wha r t on Paper s on Urban Aff air s2005, 59-97.

The views expressed are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York

or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to theaccuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained indocuments produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.

112956727 Housing Busts and Household Mobility

Documents

Transcript of 112956727 Housing Busts and Household Mobility