State Dependent E↵ects of Monetary Policy:

The Refinancing Channel ∗

Martin Eichenbaum†, Sergio Rebelo‡, and Arlene Wong§
October 2021

Abstract
This paper studies how the impact of monetary policy depends on the dis-
tribution of savings from refinancing mortgages. We show that the efficacy of
monetary policy is state dependent, varying in a systematic way with the pool
of potential savings from refinancing. We construct a quantitative dynamic life-
cycle model that accounts for our findings and use it to study how the response
of consumption to a change in mortgage rates depends on the distribution of
savings from refinancing. These e↵ects are strongly state dependent. We also
use the model to study the impact of a long period of low interest rates on the
potency of monetary policy. We find that this potency is substantially reduced
both during the period and for a substantial amount of time after interest rates
renormalize.

Keywords: monetary policy, state dependency, refinancing.

JEL codes: E52, G21.

∗
We thank the editor, Emi Nakamura, and Adrien Auclert, Martin Beraja, David Berger, Francesco
Bianci, Luigi Bocola, Christoph Boehm, Daniel Greenwald, Erik Hurst, Arvind Krishnamurthy, Moritz
Lenel, Monika Piazzesi, Mikkel Plagborg-Moller, Martin Schneider, Chris Tonetti, Joe Vavra, and Mark
Watson for their comments. We are grateful to Jose Alvarez and Laura Murphy for excellent research
assistance.
†
Department of Economics, Northwestern University and NBER, [email protected]
‡
Finance Department, Kellogg School of Management, Northwestern University, NBER and CEPR,
[email protected]
§
Department of Economics, Princeton University and NBER, [email protected]
1 Introduction
In the U.S., most mortgages have a fixed interest rate and no prepayment penalties.
The decision to refinance depends on the potential savings relative to the refinancing
costs. In this paper, we study how the impact of monetary policy depends on the
distribution of savings from refinancing the existing pool of mortgages. We show that
the efficacy of monetary policy is state dependent, varying in a systematic way with
the pool of savings from refinancing.
We construct a quantitative dynamic life-cycle model that highlights new trade-o↵s
in the design of monetary policy. These results are interesting to the extent that our
model is a credible representation of the data. Our model has a number quantitative
properties that lend support to its credibility. First, it is consistent with the life-
cycle dynamics of home-ownership rates, consumption of non-durable goods, household
debt-to-income ratios, and net worth. Second, it accounts for the probability that a
mortgage is refinanced conditional on the potential savings from doing so. Third, and
most importantly, the model accounts quantitatively for the state-dependent nature
of the e↵ects of monetary policy on refinancing decisions that we document in our
empirical work.
We use our model to study how the impact of a decline in interest rates on con-
sumption depends on the distribution of mortgage rates. One simple measure of the
savings from refinancing is the average gap between outstanding mortgages and current
mortgage rates. When this gap is equal to the average value in the data, a 25 basis
point drop in the mortgage rate leads to a 1 percent rise in consumption. In contrast,
when this gap is one standard deviation above the mean, then consumption rises by 1.4
percent. So, our model implies strong state dependency in the response of consumption
to a fall in mortgage rates.
We also use our model to study how the potency of monetary policy is a↵ected by
the history of interest rates. In response to the financial crisis, the Federal Reserve

1
kept interest rates low for an extended period of time. The potential benefits of this
policy are widely understood (see e.g. Woodford (2012) and McKay, Nakamura and
Steinsson (2016)). Our model points to a potentially important cost: it reduces the
potency of monetary policy during the period of low interest rates as well as during the
renormalization period and its aftermath. The size of these e↵ects is substantial. In
our model-based experiments, when interest rates are below their steady-state values
for six years, monetary policy is less potent for up to two years after renormalization.
Our empirical results are closely related to contemporaneous, independent work by
Berger, Milbradt, Tourre, and Vavra (2018). We view their work as complementary to
ours. In contrast to these authors, we use a quantitative life-cycle model to study the
impact of a lengthy period of low interest rates on the efficacy of monetary policy.
We build on Wong (2020) who studies how the impact of monetary policy shocks on
consumption varies by age and mortgage decisions. She finds that mortgage decisions,
including refinancing, are a key determinant of why consumption responses to monetary
policy shocks vary by age. Our contributions relative to Wong (2020) are two-fold.
First, we document and model the state-contingent nature of refinancing decisions and
its implications for the efficacy of monetary policy. Second, we focus on how the history
of interest rates a↵ects the potency of monetary policy. In pursuing these objectives,
we must overcome the challenge of allowing for state-dependent e↵ects of monetary
policy in a structural model. The basic issue is that to make their decisions people
must form expectations about future income, mortgage rates, house prices, and rental
rates. The stochastic processes for these variables are state dependent, so our model
solution technique must accommodate this dependency.
Our paper is organized as follows. Section 2 discusses the related literature. Sec-
tion 3 describes the data used in our analysis. Section 4 discusses our measures of
potential savings from refinancing. Our basic empirical results are contained in Section
5. We present our quantitative life-cycle model of housing, consumption and mort-
gage decisions in Section 6. In Section 7, we show that our model can account for the

2
state-dependent e↵ects of monetary policy that we document in our empirical work.
In addition, we use the model to study the state-dependent e↵ects of monetary pol-
icy on consumption. Section 8 uses our model to study how the potency of monetary
policy is a↵ected by an extended period of low interest rates. Section 9 provides some
conclusions.

2 Related literature
Our paper relates to five strands of literature. The first strand is a classic literature on
the e↵ect of changes in interest rates on mortgage refinancing. Dunn and McConnell
(1981) develop a theoretical model for pricing mortgage-backed securities that takes
into account the e↵ect of refinancing on the prices, risks and expected returns of such
securities. In an early empirical contribution, Green and Shoven (1986) use a pro-
portional hazard model to estimate the reduction in the probability of prepayment of
fixed-rate mortgages associated with interest rate changes. Schwartz and Torous (1989)
extend the Green and Shoven (1986) analysis to include the e↵ects of seasoning, lagged
refinancing rates, heterogeneity in borrowers, and seasonal e↵ects. They use the re-
sulting model to study the e↵ect of prepayment on the valuation of mortgage-backed
securities. Recent contributions to this literature study the distribution of mortgage
rates across borrowers and emphasize the role of transaction costs and inattention in
explaining refinancing decisions. Examples include Andersen et al. (2015) and Bhutta
and Keys (2016). In this paper, we extend the existing literature by studying how the
distribution of mortgage rates generates state dependency in the e↵ects of monetary
policy.
The second strand of the literature is a large body of empirical work that studies
consumption and refinancing responses to interest rate changes. This literature shows
that households increase their expenditures when they reduce their mortgage payments
and engage in cash-out refinancing (see, Beraja et al. (2018) and the references therein).

3
In this paper, we extend the existing literature by showing that the e↵ects of interest
rate changes on refinancing and real outcomes depend on the distribution of mortgage
rates. This type of state dependency di↵ers from the state dependency based on loan-
to-valuation constraints or home equity emphasized by Beraja et al. (2018).
Because young households tend to be borrowers, while old households tend to invest
in long-term bonds, monetary policy has potentially important distributional e↵ects.
Our paper is related to a third strand of the literature on the distributional e↵ects
of inflation (see Doepke and Schneider (2006) and Doepke, Schneider and Selezneva
(2018) and the references therein). In contrast to these papers, we focus on the state-
dependent e↵ects of monetary policy, and how these e↵ects are shaped by past interest
rate decisions made by the Federal Reserve.
The fourth strand of literature focuses on the role of the mortgage market in the
transmission of monetary policy. Iacovelo (2005), Garriga, Kydland and Sustek (2017)
and Greenwald (2018) model the transmission mechanism using a representative bor-
rower and saver model. In contrast, we use a heterogenous agent, life-cycle model that
features transaction costs and borrowing constraints. Our model is related to work
by Rios-Rull and Sanchez-Marcos (2008), Iacovello and Pavan (2013), Auclert (2017),
Berger, Guerrieri, Lorenzoni, and Vavra (2017), Garriga and Hedlund (2017), Guren,
Krishnamurthy, and McQuade (2017), Guren, McKay, Nakamura, and Steinsson (2018),
Kaplan, Mitman, and Violante (2017), Kaplan, Moll, and Violante (2018), and Wong
(2020)).
Finally, our work is related to a recent literature that stresses the importance of
mortgage refinancing as a key channel through which monetary policy a↵ects the econ-
omy. This literature discusses why the efficacy of monetary policy depends on the
state of the economy because of supply-side considerations. For example, authors like
Greenwald (2018) emphasize the importance of loan-to-value ratios and debt servicing-
to-income ratios. Other authors focus on the e↵ect of changes in house prices on the
ability of households to refinance their mortgages. For example, Beraja, Fuster, Hurst,

4
and Vavra (2018) show that regional variation in house-price declines during the Great
Recession created dispersion in the ability of households to refinance.
In contrast to the previous literatures, we focus on reasons why the efficacy of
monetary policy depends on the state of the economy because of demand-side consid-
erations, i.e. households’ desire to refinance their mortgages. We certainly believe that
supply-side constraints were important in the aftermath of the financial crisis. But we
also think that demand-side considerations were important prior to the crisis and will
become increasingly important as credit markets return to normal.

3 Data
Our empirical work is primarily based on CoreLogic Loan-Level Market Analytics, a
loan-level panel data set with observations beginning in 1995 which includes loan-level
analytics mortgage and origination data. In our benchmark analysis, we end the sample
in 2007. This decision is motivated by the widespread view that credit constraints were
much more prevalent during the financial crisis period than in the preceding period (see
e.g. Mian and Sufi (2014) and Beraja et al. (2018)).
The CoreLogic data includes borrower characteristics (e.g. FICO and ZIP code)
and loan-level information.1 The latter includes the principal of the loan, the mortgage
rate, the loan-to-value ratio (LTV), and the purpose of the loan (whether it refinances
an existing loan or finances the purchase of a new house). There is no information on
the purpose of a small fraction of the loans (5 percent). We compute the fraction of
mortgages refinanced as the ratio of mortgages that are refinanced to the total number
of mortgages outstanding. This measure is conservative in the sense that some of
the loans whose purpose is unknown to us could have been used to refinance existing
mortgages.
For each borrower, we obtain county-level demographic information, including age
1
FICO is the acronym for the credit score computed by the Fair Isaac Corporation.

5
structure, share of employment in manufacturing, lender competitiveness, measures of
home-equity accumulation, educational attainment, unemployment, and per capita in-
come. Appendix A contains a description of these variables. We also obtain county-level
housing permits from the Census Building Permits Survey and county-level unemploy-
ment rates from the Current Population Survey.
We use the Freddie Mac Single Family Loan-Level dataset to study cash-out refi-
nancing, defined as instances in which households increase the loan balance when they
refinance. Cash-out refinances are identified by the Freddie Mac loan-purpose flag.
These data are available since 1999.
Throughout, we confine our analysis to fixed-rate 30-year mortgages. Our results
are robust to considering mortgages of di↵erent maturities. Figure 1 displays the level
and first di↵erences of the fraction of mortgages that are refinanced and the fraction of
refinanced mortgages that are cash-out.
We obtain aggregate time-series variables, including forecasts of unemployment,
inflation and GDP from the Survey of Professional Forecasters. We obtain time-series
of the Federal Funds Rate, and income per capita from the Federal Reserve Bank
of St. Louis. National house prices are obtained from Mack and Martı́nez-Garcı́a
(2011). Rental rates are obtained from the Organization for Economic Cooperation
and Development.
Finally, we obtain measures of expected inflation from the Federal Reserve Bank of
Cleveland. Nominal variables are converted to real variables using the consumer price
index. We obtain county-level house price data from two sources. House price data
is obtained from Hurst et al (2019). These authors obtain the data from the Federal
Housing Finance Authority (FHFA). Our second source of house price data, which is
used to construct home equity, is obtained from the Global Financial Data Real Estate
database. The database contains house price data on monthly loans purchased by
Freddie Mac or Fannie Mae within each region.
Lender competitiveness data was constructed by Scharfstein and Sunderam (2016),

6
using Home Mortgage Disclosure Act (HMDA) data.

4 Measuring the potential savings from refinancing

A key variable in our analysis is the potential savings that a household would realize
by refinancing its mortgage at the current mortgage rate. Potential savings depend
on a variety of factors, including old and new mortgage interest rates, outstanding
mortgage balances, and the precise refinancing strategy that a household pursues. In
general, it is impossible to construct a simple, non-parametric summary statistic of
these potential savings. Our benchmark measure is the average interest-rate gap. This
measure is based on the di↵erence between the current and alternative mortgage rate
that household i could refinance at.
We compute the average, across households, of time-t interest-rate gaps between
new and old loan as: nt
1 X
At = rold ritnew . (1)
nt i=1 it
Here, ritnew is the interest rate at time t for a new 30-year conforming mortgage for the
same FICO and region as the original mortgage. The variable nt denotes the number
of mortgages outstanding at time t. At is a real variable, since it is based on the
di↵erence between two nominal interest rates. A virtue of At is that it doesn’t impose
any assumptions about the household’s refinancing decision. The downside is that it
abstracts from relevant information such as outstanding balances or characteristics of
the new mortgage (e.g. duration and fixed versus variable interest rates).
The new mortgage rate depends on borrower characteristics, including FICO and
region. We group FICO scores into the following bins: below 600, 600 620, 620 640,...
,...,760 780, and greater than 780. We condition on the household’s region to take into
account the possibility that mortgage rates vary by region, say because of di↵erences
in income or house price growth.2 We also considered versions of ritnew that condition in
2
As a practical matter, we find that our results are robust to not conditioning on the household’s

7
a non-parametric way on additional variables such as the loan-to-valuation ratio or the
mortgage balance. Adding these measures does not significantly improve the ability of
ritnew to fit the distribution of interest rates across new borrowers.
The annualized unconditional quarterly mean and standard deviation of At is 14
basis points and 70 basis points, respectively. The distribution of the interest rate
gap varies considerably over time. To make this point concrete, Figure 2 displays the
distribution of interest-rate gaps in 1997.Q4 and 2000.Q4. These dates correspond to
local turning points in the average real mortgage rate. The fraction of households with
positive savings and the average savings is much higher in 1997.Q4 than in 2000.Q4.
In Appendix C3, we consider three alternative measures of the savings from refi-
nancing. The first is the fraction of mortgages with a positive interest-rate gap. The
second is based on the average present value of savings from pursuing a simple refinanc-
ing strategy. The third is based on the fraction of loans above a time-varying threshold
for refinancing, defined in Agawal, Driscoll and Laibson (2013). Our results are robust
to using these alternative measures.
The average cross-sectional dispersion in the average rate gap is 1 percent. In
Appendix B, we report the correlation between the average rate gap and observable
characteristics (unemployment rate, per capita income, share of college educated, home
equity accumulation, median age, manufacturing share, and share of males in the pop-
ulation). Other things equal, the average rate gap is higher in areas with a higher
unemployment rate and it is a decreasing function of per capita income, the share of
the population that is college educated, and the amount of home equity accumulation.

5 Empirical results
In this section, we study how the impact on refinancing activity of a change in the
mortgage rate depends on the average savings from refinancing. We report basic cor-
region. This finding is consistent with Hurst, Keys, Seru and Vavra (2016) who find little evidence of
spatial variation in mortgage rates.

8
relations based on ordinary least squares (OLS) in Appendix D. Here, we implement
an instrumental-variable (IV) strategy for measuring the marginal e↵ect of a drop in
mortgage rates on the fraction of loans that are refinanced. Our main results are that
this marginal e↵ect is state dependent as is the impact of a fall in interest rates on
economic activity.

5.1 State dependency and refinancing activity

We begin by considering the regression:

⇢ct+4 = 0X
c
+ 1 RtM + 2 RtM ⇥ Act 1 + c
3 At 1 + 4 Zt 1 + c
5 Zt 1 + ⌘tc . (2)

Here, ⇢ct+4 is the fraction of mortgages refinanced in county c between quarters t and
t + 4, X c is a vector of county fixed e↵ects, and RtM denotes the percentage fall in
our measure of the mortgage rate.3 The variable Act 1 is the average interest-rate gap
for mortgages in county c at time t 1. The vector Zt 1 denotes a set of time-varying
controls. Motivated by results in Nakamura and Steinsson (2018) we include as controls
the average forecast of the Survey of Professional Forecasters (SPF) for the following
variables: real GDP growth (two-year ahead), the civilian unemployment rate (two-
years ahead), and the CPI inflation rate (one and two-years ahead). The variable Ztc 1

includes the following county-level controls: the unemployment rate, average log-change
in real home equity, one-year lag of the refinancing rate, and a Herfindahl index of the
mortgage sector. We include the latter index, developed in Scharfstein and Sunderam
(2013), to capture any variation in pass through by region induced by time variation in
competition across counties. We cluster the standard errors at the county level.
The coefficient 1 measures the e↵ect of a change in mortgage rates on refinancing
rates when Act 1 is zero. The coefficient 2 measures how the e↵ect of an interest rate
change depends on the level of Act 1 . Identification of 1 and 2 comes from both
3
If the mortgage rate falls by 25 basis points, Rt = 0.25. Defining Rt as the fall in the interest
rate, instead of the interest rate change makes the regression coefficients easier to interpret.

9
cross-sectional and time-series variation in the response of refinancing to interest rate
changes.4
We estimate regression (2) using two instruments for RtM that exploit exogenous
changes in monetary policy.5 The instruments are based on high-frequency movements
in the Federal Funds futures rate and the two-year Treasury bond yield in a small
window of time around Federal Open Market Committee (FOMC) announcements.6
In the case of the Federal Funds futures, the monetary policy shock is defined as:
D
"t = (yt+4+ yt 4 ). (3)
D t
Here, t is the time when the FOMC issues an announcement, yt+4+ is the Federal Funds
futures rate shortly after t, yt 4 is the Federal Funds futures rate just before t, and
D is the number of days in the month. The term D/(D t) adjusts for the fact that
Federal Funds futures contracts settle on the average e↵ective overnight Federal Funds
rate. We consider a 60-minute window around the announcement that starts 4 = 15
minutes before the announcement. This narrow window makes it highly likely that
the only relevant shock during that time period (if any) is the monetary policy shock.
Following Cochrane and Piazessi (2002) and others, we aggregate the identified shock
to construct a quarterly measure of the monetary policy shock. This aggregation relies
on the assumption that shocks are orthogonal to economic variables in that quarter.
4
In practice, most of the variation in refinancing rates comes from time-series variation in interest
rates. One way to see this result is to regress the rate of refinancing in county c at time t on time and
county fixed e↵ects. County fixed e↵ects account for less than 20 percent of the variation in refinancing
rates.
5
It is difficult to give a causal interpretation to the OLS-based estimates of 1 and 2 because
of potential endogeneity bias caused by any omitted variable that a↵ects both mortgage rates and
savings from refinancing. For example, suppose that during a recession more people are unemployed
and therefore less willing to incur the fixed costs associated with refinancing. Also, suppose that the
recession occurred because the Fed raised interest rates. Then, RtM and RtM ⇥ Act 1 would be
positively correlated with ⌘tc creating a downward bias in 1 and 2 .
6
This approach has been used by Kuttner (2001), Gürkaynak, Sack and Swanson (2005), Cochrane
and Piazessi (2002), Nakamura and Steinsson (2018), Gorodnichenko and Weber (2015), and Wong
(2020), among others.

10
The standard deviation of the implied monetary policy shock is 12 basis points.
In Appendix C2, we report our empirical analysis when we measure a monetary
policy shock using the 2-year Treasury yield:

"t = yt+4+ yt 4 .

Instrumental variable results. We begin by providing evidence that monetary-

policy shocks are a strong instrument for changes in mortgage rates. First, we show that
monetary policy shocks significantly a↵ect mortgage rates. To this end, we estimate via
OLS the contemporaneous change in the 30-year mortgage rate after a one percentage
point monetary policy shock. Our point estimate is 60 basis points with a standard
error of 28 basis points. Taking sampling uncertainty into account, this estimate is
consistent with those of Gertler and Karadi (2015, Table 1), which range from 17 to 48
basis points, depending on the set of instruments used to identify the monetary shock.
Second, we estimate the following first-stage regressions:

RtM = ↵0 X c + ↵1 "t + ↵2 "t ⇥ Act 1 + ↵ 3 Zt 1 + ↵4 Ztc 1 + ⌘1t ,

RtM Act 1 = 0X
c
+ 1 "t + 2 "t ⇥ Act 1 + 3 Zt 1 + c
4 Zt 1 + ⌘2t .

Table 1 reports our first-stage regression estimates. The F test for the joint significance
of the regression coefficients is greater than ten. This result is consistent with the view
that monetary policy shocks are strong instruments. We also perform the Stock-Yogo
(2005) test for the null hypothesis of weak instruments. The test statistic is 79.5, which
firmly rejects the null of weak instruments.
Panel A of Table 2 reports results based on estimating regression (2) using instru-
mental variables. We cluster the standard errors at the county level. The estimated
values of 1 and 2 are statistically significant at the one percent significance level.
Estimated parameters on controls are reported in Appendix C3. That appendix also
documents the robustness of the estimated value of 1 to including various lagged values

11
of the refinancing rate in our empirical specification. Appendix D contains the analogue
of Panel A of Table 2 generated using OLS.
To interpret the coefficients in Panel A of Table 2, suppose that all independent
variables in regression (2) are initially equal to their time-series averages and that
the average interest-gap is initially equal to its mean value of 14 basis points. The
unconditional average share of loans that is refinanced is 7.9 percent. The estimates in
Panel A of Table 2 imply that a 25 basis point drop in mortgage rates raises the share of
loans that is refinanced to 8.0 percent.7 Now, suppose that the drop in mortgage rates
occurs when the average interest-rate gap is equal to 56 basis points. The latter is the
mean value of 14 basis points plus one standard deviation (70 basis points). Then,
a 25 basis points drop in mortgage rates raises the share of loans that is refinanced
to 12.6 percent.8 So, the marginal impact of a one standard-deviation increase in the
average interest-rate gap is 4.6 percent. This e↵ect is large relative to the average
annual refinancing rate, 7.9 percent.
Figure 3 reports the impulse response function of the fraction of mortgages refi-
nanced to a 50 basis point drop in the mortgage rate. We compute this function for
two cases corresponding to whether the average interest-gap is initially equal to 14
basis points or to 56 basis points. Panel A shows that in the first case there is a weak
response of refinancing to the interest rate cut. Panel B shows that in the second case
there is a persistent rise in refinancing activity over a two-year period.
A natural question is whether there is substantial variation over time in the impact
of mortgage rate reductions on refinancing rates. To answer this question, we use the
estimated version of regression (2) to calculate the e↵ect of a 50 basis points reduction
in the mortgage rate on the refinancing rate at each point in time in our sample. The
solid line in Figure 4 plots this e↵ect and the dashed lines correspond to the 95 percent
confidence interval (computed using the delta method). There is substantial variation
7
This value is given by 7.9% + 0.25 ⇥ ( 1+ 2 ⇥ 0.14) = 8.0%.
8
This value is given by 7.9% + 0.25 ⇥ ( 1+ 2 ⇥ 0.56) = 12.6%.

12
in the potency of monetary policy through the refinancing channel. This e↵ect was most
potent in 1998 and 2003 and least potent in 1995 and 1999. These dates correspond
to periods in which the average rate gap is high and low, respectively. We discuss our
model’s implication for these e↵ects in Section 5.
As a robustness test, we also included in regression (2) interaction terms of the form
Rtm Ztc 1 . These results are reported in Appendix C3. The implied estimates of 2

are statistically indistinguishable from those reported in Panel A of Table 2. The fact
that including the interaction terms does not change the estimated elasticities implies
that the state dependency that we highlight is distinct from other potential mechanisms
explored in the literature. These mechanisms include, for instance, di↵erential responses
in refinancing to a decline in mortgage rates due to di↵erences in competitiveness of the
local lending market. It is also distinct from state dependency related to cross-county
variation in the value of home equity.
We find that our results are robust to including as additional controls the lag of
the refinancing rate, the fraction of mortgages in county c that have adjustable mort-
gage rates, and the interaction of this variable with the monetary policy shock. Our
results are also robust to including interactions between the monetary policy shock and
slow-moving county characteristics such as the median age, share of employment in
manufacturing, and share of college educated workers.

5.2 Cash-out refinancing

In this subsection, we use Freddie Mac data on single-family loans to study how cash-out
refinancing responds to changes in mortgage rates. Cash-out refinancing occurs when
the balance of the new mortgage is higher than that of the old mortgage. Mian and
Sufi (2014) show that households predominantly use this type of refinancing to increase
their consumption. So, cash-out refinancing plays an important role in determining the
e↵ects of changes in interest rates on consumption.
We run a version of regression (2) in which the dependent variable is the fraction

13
of total loans with cash-out refinancing in county c between quarters t and t + 4. Panel
B of Table 2 reports our results. Both 1 and 2 are significant at the one-percent
significance level. To interpret these coefficients, suppose that all independent variables
in regression (2) are initially equal to their time-series averages and that the average
interest-gap is initially equal to its mean value of 14 basis points. The estimates
in Panel B of Table 2 imply that a 25 basis point drop in mortgage rates raises the
share of loans with cash-out refinancing by 1.2 percent. Now suppose that the drop in
mortgage rates occurs when the average interest-rate gap is 56 basis points. Then, a 25
basis point drop in mortgage rates raises the share of loans with cash-out refinancing
by 4.3 percent. So, the marginal impact of a one standard deviation increase in the
average interest-rate gap is 3.1 percent. This e↵ect is large relative to the average
annual cash-out refinancing rate, 5.5 percent.
We also estimate a version of regression (2) in which the dependent variable is the
log change in the balance of mortgages with cash-out refinancing. Panel C of Table
2 reports our results. To interpret these coefficients, suppose that all independent
variables in regression (2) are initially equal to their time-series averages and that the
average interest-rate gap is initially equal to its mean value of 14 basis points. The
estimates in panel C of Table 2 imply that a 25 basis points drop in mortgage rates
raises the balance of the mortgages with cash-out refinancing by 5.2 percent. Now
suppose that the drop in mortgage rates occurs when the average interest-rate gap is
56 basis points. Then, a 25 basis point drop in mortgage rates raises the balance of the
mortgages with cash-out refinancing by 8.9 percent. So, the marginal impact of a one
standard deviation increase in the average interest-rate gap is 3.8 percent. The median
mortgage balance in 2007 was roughly $123, 000. It follows that a 3.8 percent increase
in mortgage balance translates into equity extraction of roughly $4, 700, a substantial
amount of cash that becomes available for consumption.

14
5.3 Refinancing and economic activity
We now study how a change in mortgage rates a↵ects economic activity. To be clear,
we are not the first to establish that changes in mortgage rates induced by monetary
policy shocks a↵ect economic activity. We are simply establishing that these e↵ects are
state dependent.
Our first measure of economic activity is the county-level unemployment rate. Our
second measure is the number of county-level permits required for new, privately-owned
residential buildings. This series, produced by the Census Building Permits Survey
since 2000, is of particular interest to us because it is the only component of the Con-
ference Board’s leading indicator index available at the county level. We aggregate
these monthly data to a quarterly frequency.
We begin by considering a regression where the dependent variable is the change in
the unemployment rate between quarter t and t + 4:

Unemploymentct,t+4 = ✓0 X c + ✓1 RtM + ✓2 RtM ⇥ Act c c c

1 + ✓3 At 1 + ✓4 Zt 1 + ✓5 Zt 1 + ⌘t ,

where Zt includes the same controls used in equation (2) except for the unemployment
rate.
Table 3 reports our IV estimates obtained using the instruments discussed in Section
4.1. Standard errors are clustered at the county level. The point estimate of ✓1 is
statistically significant at a 10 percent level while ✓2 is statistically significant at the 1
percent level. To interpret the point estimates, suppose that all independent variables in
regression (4) are initially equal to their time-series averages. Our estimates imply that
a 25 basis point drop in mortgage rates lowers the unemployment rate by 0.6 percent.
Suppose that the drop in mortgage rates occurs when the average interest-rate gap
is equal to 56 basis points. Then a 25 basis point drop in mortgage rates lowers the
unemployment rate by 1.8 percent. So, the marginal impact of a one standard deviation
increase in the average interest-rate gap is 1.2 percent.

15
 We now consider a version of the regression where the dependent variable is the
year-on-year quarterly log-change in new building permits:

log Permitsct,t+4 = ✓0 X c +✓1 RtM +✓2 RtM ⇥Act 1 +✓3 Act 1 +✓4 Zt 1 +✓5 Ztc 1 +⌘tc . (4)

Table 3 reports our IV estimates. Both ✓1 and ✓2 are statistically significant at a 1

percent significance level. To interpret the point estimates suppose that all independent
variables in regression (4) are initially equal to their time-series averages. The estimates
in column 2 imply that a 25 basis point drop in mortgage rates raises the percentage
change in new permits to 17.0. Now suppose that the drop in mortgage rates occurs
when the average interest-rate gap is equal to 56 basis points. Then a 25 basis points
drop in mortgage rates raises the percentage change in new permits to 23.6 percent. So,
the marginal impact of a one standard deviation increase in the average interest-rate
gap is 6.6 percent. This e↵ect is large relative to a one standard deviation change in
housing permits (26 percent).
Overall, we view the results of this section as providing strong support for two
hypotheses. First, the e↵ect of a change in the interest rate on refinancing activity
is state dependent. When measures of the average gains from refinancing are high, a
given fall in the interest rate induces a larger rise in refinancing activity. Second, the
e↵ect of a change in the interest rate on economic activity, as measured by new housing
permits or the rate of unemployment, is state dependent in a similar way. This finding
is consistent with the results in Di Maggio, Kermani, Keys, Piskorski, Ramcharan, Seru,
and Yao (2017). These authors show that households who experience a drop in monthly
mortgage payments increase their car purchases. It is also consistent with results in
Berger, Milibrandt, Tourre and Vavra (2018) who show that there is a state-dependent
rise in auto registrations when interest rates fall.

16
6 A life-cycle model
To analyze the state-dependent e↵ects of monetary policy, we use a life-cycle model
with incomplete markets, short-term borrowing constraints, refinancing costs, and loan-
to-value constraints on mortgages. Our model generalizes the framework in Wong
(2020) to allow for state dependency in the aggregate state process for the interest rate,
income, house prices and rental rates. This generalization allows us to incorporate
state-dependent feedback e↵ects of monetary policy shocks on aggregate variables.
We use the model for three purposes. First, we quantify the structural factors that
drive the state-dependent e↵ects of monetary policy. Second, we estimate the state-
dependent e↵ect of an exogenous change in the interest rate on consumption. Third, we
study how the potency of monetary policy is a↵ected by a long period of low interest
rates.
It is evident that there is a great deal of heterogeneity across households in their
propensity to refinance in response to an interest rate cut. One way to capture that
heterogeneity is to allow for substantial heterogeneity in unobserved fixed costs of refi-
nancing. An alternative is to model heterogeneity in refinancing behavior as reflecting
demographics, initial asset holdings, and idiosyncratic income shocks. We choose the
second strategy to minimize the role of unobservable heterogeneity. An advantage of
this approach is that it is consistent with the positive correlation between consumption
growth and refinancing decisions at the household level. This correlation is important
for generating a response of aggregate consumption to interest rate changes.9

Households. The economy is populated by a continuum of people indexed by j. We

think of the first period of life as corresponding to 25 years of age. Each person can
live up to 60 years. The probability of dying at age a is given by 1 ⇡a . Conditional
9
Recent work that emphasizes the correlation between household consumption and interest rate
changes for understanding the behavior of aggregate consumption includes Auclert (2017), Kaplan,
Violante and Moll (2018), and Wong (2020).

17
on surviving, people work for 40 years and retire for 20 years. The upper bound on a
person’s life is 85 years (T = 85).
The momentary utility of person j who is a periods old at time t is given by:
1
c↵jat h1jat↵ 1
ujat = , > 0.
1
Here, cjat and hjat denote the consumption and housing services of person j who is a
periods old at time t. People derive housing services from either renting or owning a
house. Renters can freely adjust the stock of rental housing in each period. To buy
new
a home of size h, households pay a transaction cost (h). To refinance an existing
refi 10
mortgage, homeowners pay a lump-sum transaction cost . The stock of housing
depreciates at rate . There are no transactions costs from changes in the housing
stock associated with depreciation of an existing home. There are no adjustment costs
associated with selling a house.
Upon death, the wealth of person j who is a periods old at time t, Wjat , is passed
on as a bequest. If a person has an outstanding mortgage upon death, the house is sold
to pay the mortgage and the remainder of the estate is passed on as a bequest. Person
1
j derives utility B Wjat 1 / (1 ) from this bequest. Here B is a positive scalar.
The presence of a bequest motive allows the model to be consistent with the fact that
many people die with large amounts of assets (see e.g. Huggett (1996) and De Nardi
and Yang (2014)). More importantly, this motive helps the model generate the fall in
consumption and house downsizing that we observe in data for older consumers. The
reason is that, as people get older, the bequests receive a higher weight in the utility
function relative to consumption and housing.

Income processes. The time-t labor income of person j who is a periods old at time
t, yjat , is given by:
log(yjat ) = a + ⌘jt + a log(yt ). (5)
10
See DeFusco and Mondragon (2018) for evidence that fixed costs, including closing costs and
refinancing fees, are important determinants of refinancing decisions.

18
Here, a and ⌘jt are a deterministic age-dependent component and a stochastic, id-
iosyncratic component of yjat , respectively. We assume that

⌘jt = ⇢⌘ ⌘jt 1 + "⌘t ,

where |⇢⌘ | < 1 and "⌘t is a white noise process with standard deviation, ⌘. The variable
yt denotes aggregate real income. The term a captures the age-specific sensitivity of
yjat to changes in aggregate real income.
As in Guvenen and Smith (2014), we assume that a person receives retirement
income that consists of a government transfer. The magnitude of this transfer is a
function of the labor income earned in the year before retirement.

Mortgages. Home purchases are financed with fixed-rate mortgages. An individual

j who enters a mortgage loan at age a in date ⌧ , pays a fixed interest rate Rja⌧ and
makes a constant payment Mja⌧ . The mortgage principal evolves according to:

bj,a+1,t+1 = bjat (1 + Rja⌧ ) Mja⌧ .

Mortgages are amortized over the remaining life of the individual. So, the maturity of
a new loan for an a-year old person is m(a) = T a. The fixed interest rate Rja⌧ is
m(a)
equal to r⌧ , which is the time-⌧ market interest rate for a mortgage with maturity
m(a).
The mortgage payment, Mja⌧ , is given by:
bja⌧
Mja⌧ = Pm(a) . (6)
(1 + Rja⌧ ) k
k=1

If a person refinances or buys a new house at time t, the new mortgage rate is given by
the current fixed mortgage rate:11

Rjat = r⌧m(a) .
11
In practice, U.S. mortgage rates can depend on the size of the loan. For example, jumbo mortgages
(loans sizes that exceed the maximum guaranteed by Fannie Mae and Freddie Mac) have higher interest
rates than non-jumbo mortgages. To simplify, we abstract from this feature in our analysis.

19
Bond holdings. A person can invest in a one-year bond that yields an interest rate
rt . The variable sjat denotes the time-t bond holdings of person j who is a years old at
time t. Bond holdings have to be non-negative, sjat 0.

Loan-to-value constraint. At the time of origination, the size of a mortgage loan

must satisfy the constraint:
bjat  (1 )pt hjat .

Here, pt is the time-t price of a unit of housing and pt hjat is the minimum down
payment on a house.

State variables. The state variables in our model are z = {a, ⌘, K, S}. Here, a, ⌘,
and K denote age, idiosyncratic labor income, and asset holdings, respectively. The
vector K includes short-term asset holdings (s), the housing stock (hown for homeown-
ers, zero for renters), the mortgage balance (b for homeowners, zero for renters), and
the interest rate (R) on an existing mortgage. Finally, S denotes the aggregate state
of the economy which consists of the logarithm of real output, y, the logarithm of real
housing prices, p, the real interest rate on short-term assets, r, and the logarithm of
economy-wide average positive savings from refinancing, A. We assume that S is a
stationary stochastic process (see Section 6.2).

Mortgage interest rate and rental rates. It is difficult for traditional asset pricing
models to account for the empirical properties of mortgage interest rates, rental rates
and housing prices (see Piazzesi and Schneider (2016)). For this reason, we assume
that these variables depend on the aggregate state of the economy via functions that we
estimate. This approach allows the model to be consistent with the empirical properties
of these variables.
The mortgage interest rate, rtm , is given by

rtm = am m m
0 + a1 log(rt ) + a2 log(yt ). (7)

20
This formulation captures, in a reduced-form way, both the term premia and changes
in risk premia that arise from shocks to the aggregate state of the economy.
The real rental rate is given by:

log(prt ) = ↵0 + ↵1 log(rt ) + ↵2 log(yt ) + ↵3 log(pt ). (8)

Value functions. We write people’s maximization problems in recursive form. To

simplify notation, we suppress the dependence of variables on j, a and t. Denote by
V (z)rent , V (z)purchase , V (z)own & refi , and V (z)own & no refi the value functions associated
with renting, purchasing, owning a home and refinancing, and owning a home and not
refinancing, respectively. A person’s overall value function, V (z), is the maximum of
these value functions:

V (z) = max V (z)rent , V (z)purchase , V (z)own & refi , V (z)own & no refi . (9)

A renter maximizes
" 0 1
#
rent rent 0 W 1
V (z) = max u c, h + E ⇡a V (z ) + (1 ⇡a )B , (10)
rent 0
c,h ,s 1

subject to the budget constraint,

c + s0 + pr hrent = y + (1 )phown + (1 + r)s b(1 + R), (11)

and the borrowing constraint on short-term assets,

s0 0.

The discount rate and the probability of survival are denoted by and ⇡a , respectively.
h i
0 1
The term B W 1 / (1 ) represents the utility from bequests. The terms
(1 )phown and b(1 + R) in equation (11) take into account the possibility that the
renter was a home owner at time t 1. The renter’s housing stock and mortgage debt
are both zero:
0
h own = b0 = 0.

21
 A household who decides to purchase a new home maximizes:
" 0 1
#
W 1
V (z)purchase = max u (c, h0own ) + E ⇡a V (z 0 ) + (1 ⇡a )B ,
0 0
c,s ,b 1

subject to the budget constraint

c + s0 + ph0own b0 = y + (1 )phown + (1 + r)s b(1 + R) new

(h0own ),

the borrowing constraint on short term assets,

s0 0,

and the minimal down payment required on the mortgage,

b0  (1 )ph0own .

The new mortgage interest rate is given by:

R0 = r m .

A homeowner who does not refinance his mortgage maximizes:

" 0 1
#
W 1
V (z)own & no refi = max u (c, hown (1 )) + E ⇡a V (z 0 ) + (1 ⇡a )B ,
0
c,s 1
(12)
subject to the budget constraint,

c + s0 = y + (1 + r)s M,

the law of motion for the mortgage principal

b0 = b(1 + R) M,

and the short-term borrowing constraint

s0 0.

22
Since the person doesn’t refinance, the interest rate on his mortgage remains constant

R0 = R.

The mortgage payment is given by equation (6). The law of motion for the housing
stock is
0
h own = (1 )h0 .

A homeowner who refinances, maximizes:

" 0 1
#
own & refi own 0 W 1
V (z) = max u (c, h (1 )) + E ⇡a V (z ) + (1 ⇡a )B ,
c,s0 ,h0 own ,b0 1

subject to the budget constraint

c + s0 b0 + refi
= y + (1 + r)s b(1 + R),

the borrowing constraint on short-term assets,

s0 0,

and the minimal down payment required on the mortgage,

b0  (1 )phown (1 ).

The new mortgage interest rate is given by:

R0 = r m .

The problem for a retired person is identical to that of a non-retired person, except
that social security benefits replace labor earnings.

6.1 Calibration
Our parameter values are summarized in Table 4. We set = 2 and choose B, ,
and ↵ to target key moments of the savings and asset-holding profiles. These moments

23
include the average home ownership rate, the liquid wealth-to-income ratio for working-
age households, and the share of wealth held by older households (aged 65+) according
to the 2007 Survey of Consumer Finances. The idiosyncratic-income parameters ⇢⌘
and ⌘ are chosen to match the annual persistence and standard deviation of residual
earnings in the Panel Study of Income Dynamics. Residual earnings are the error term
in a regression of the logarithm of individual income on age and aggregate income.
The deterministic, age-specific vector a is chosen to match average log earnings by
age estimated by Guvenen et al. (2015). We choose a to match the correlation between
real aggregate income per capita and age-specific earnings in the Current Population
Survey. The house depreciation rate, , is chosen to be consistent with the average ratio
of residential investment to the residential stock from the Bureau of Economic Analysis.
We set so that, in line with Landvoigt, Piazzesi and Schneider (2015), the minimum
mortgage downpayment is 20 percent. We estimate the parameters of the processes for
mortgage rates (am m m
0 ,a1 , and a2 in equation (7)) and and rental rates (↵0 , ↵1 , ↵2 , and

↵3 in equation (8)). We discuss the empirical fit of these processes in Section 6.2.
Recall that we think of the first period of life as 25 years of age. Age-dependent
survival probabilities are given by the U.S. actuarial life-expectancy tables and assume
a maximum age of 85. Assets and income in the first period of a person’s life are
calibrated to match average assets and income for persons of ages 20 to 29 in the 2004
Survey of Consumer Finances. Mortgage rates are initialized according to the initial
empirical distribution of mortgage rates.
refi
We set the fixed cost of refinancing, , equal to $2, 100 (2 percent of median house
price) to match the average quarterly fraction of new loans (4.5 percent). This value is
consistent with the range of costs provided in the Federal Reserve’s “Consumer Guide
to Mortgage Refinancings” and with the evidence in LaCour-Little (2000).12 The fixed
costs of refinancing include the appraisal fee, the inspection fee, and the attorney review
new
fee. The transaction cost function for buying a new home has a fixed component,
12
https://www.federalreserve.gov/pubs/refinancings

24
equal to $1, 100, and a variable component, equal to 4 percent of the house price. These
value is consistent with empirical evidence as well as the values used in the literature.13
Given this calibration, the model is consistent with the average rate of home ownership
in the data.

6.2 The evolution of the aggregate states

To solve their decision problem, people must form expectations about their future
income, mortgage rates, house prices, and rental rates. Because of its partial equilibrium
nature, our model does not imply a reduced-form representation for these variables. It
seems natural to assume that people base their expectations on a time-series model
that has good forecasting properties.
Recall that we model the mortgage rate with maturity m as a function of rt and yt
(see equation (7)). We estimate this function using OLS. Table 5 reports our estimates.
Figure 5 shows that the estimated version of equation (7) does a very good job at
accounting for the time-series behavior of the 30-year mortgage rate over the period
1989-2008.
As discussed above, it is hard for traditional asset-pricing models to account for
the empirical relation between rental rates and housing prices (see, e.g., Piazzesi and
Schneider (2016)). So, we model the real rental rate as a linear function of rt , yt and
pt (see equation (8)). We estimate this function using the national house price and
rent indices obtained from the Federal Reserve Bank Dallas. Figure 6 shows that the
estimated version of equation (8) does a very good job at accounting for the time-series
behavior of the logarithm of the house price-to-rent ratio over the period 1989-2008.
We estimate a suite of quarterly time-series models for the aggregate-state vector
St = {log(rt ), log(yt ), log(pt ), log(At )}. We eliminate from consideration models with
explosive dynamics and judged the remaining models balancing parsimony and the
implied average (over time and across variables) root-mean-square-error (RMSE) of
13
See Diaz and Luengo-Prado (2012) and Berger et al. (2015).

25
one-year-ahead forecasts. Parsimony is important for the computational tractability of
our structural model.
We settled on the following model for quarterly changes in St :

St = B 0 + B 1 S t 1 + B2 log(rt 1 )at 1 + ut . (13)

Here, B1 is a 4 ⇥ 4 matrix, B0 and B2 are 4 ⇥ 1 vectors, and ut is a 4 ⇥ 1 Gaussian dis-

turbance. The variable at 1 is the logarithm of economy-wide average positive savings
from refinancing at time t 1, log(At 1 ).
Appendix E reports the average RMSE for the alternative models that we consid-
ered. These models include specifications with up to two lags of St and log(rt 1 )at 1 .
In addition, we included cross products of all the variables in di↵erent combinations
as well as squares and cubes of the di↵erent variables. We also considered di↵erent
moments of various measures of the gains from refinancing. For example, we replaced
at with the average savings (in levels), the median savings, the average interest-rate
gap, and the logarithm of average positive interest-rate gap, the median of the interest-
rate gap, the fraction of mortgages with positive savings, and the standard deviation
of savings. To conserve on space we do not report these results.
None of the RMSEs associated with the alternative specifications was smaller, taking
sampling uncertainty into account, than the RMSE associated with specification (13).
Specification (13) did have a statistically significant smaller RMSE than many of the
alternatives.
Table 6 reports point estimates and standard errors for B1 and B2 associated with
specification (13) estimated using quarterly data. The coefficients in B2 are statistically
significant at the one percent level for log(rt ) and log(pt ) and at the 5 percent level for
at .
A natural question is whether the inclusion of at and r t 1 at 1 in specification (13)
helps reduce the RMSE for the three aggregate variables, log(rt ), log(yt ), and log(pt ).
Simply adding at to a linear VAR for log(rt ), log(yt ), and log(pt ), reduces the average

26
RMSE for log(rt ), log(yt ), and log(pt ) in a modest but statistically significant way
(from 0.0298 to 0.0258). Adding the interaction term r t 1 at 1 results in an even more
modest, but statistically significant reduction, in the average RMSE for log(rt ), log(yt ),
and log(pt ).
We compute the impulse response of St to a monetary policy shock implied by (13)
as follows. First, the shocks ut are regressed on the monetary-policy shock. Second,
we compute the impact of a monetary-policy shock on log(rt ), log(yt ), log(pt ), and
log(At ). Figure 7 displays the associated impulse response functions for a one-standard-
deviation shock to monetary policy. We see that an expansionary monetary policy shock
is associated with a persistent rise in income and house prices as well as a decrease in
average positive savings from refinancing.
Recall that our model abstracts from growth. To solve the model, we set the con-
stant vector, B0 , in (13) to zero and work with the implied VAR for the level of the
variables. This procedure is equivalent to estimating the VAR using data that have
been demeaned. We approximate this VAR with a Markov chain using the procedure
described in Appendix F. We then convert the quarterly VAR into an annual VAR by
raising the transmission function to the power four. A key property of the Markov
chain is that the implied impulse response functions to a monetary policy shocks are
stationary.

7 Empirical performance of the model

We now compare our model with the data along a variety of dimensions. Model statistics
are computed using simulated data generated as follows. We start the simulation in
1994, assuming that people have the distribution of assets, liabilities and mortgage rates
observed in the data. We feed into the model the realized values of log(rt ), log(yt ), and
log(pt ) for the period 1995-2007. We simulate the idiosyncratic component of income,
yjat , for each household in our model.

27
 Given the estimated VAR, the housing Sharpe ratio implied by the model is 2.46.14
This value is similar to the Sharpe ratio of 2.53 implied by the NIPA Fixed Asset Tables
2.1, line 68 for the period 1994-2007.
One possible concern is that our model abstracts from the idiosyncratic shocks to
home prices emphasized by Landvoigt, Piazzesi, and Schneider (2015). Incorporating
idiosyncratic risk into the model would greatly increase the computational complexity
of our analysis. To assess the robustness of our results to more volatile house prices, we
increase the volatility of house-price shocks by 50 percent. Our main results regarding
the state-dependent nature of monetary policy are robust to this perturbation. More
volatile house prices make housing a riskier asset and deter home ownership. However,
that volatility does not have a first-order e↵ect on refinancing decision.

7.1 Life-cycle dynamics.

Consider the model’s ability to account for how the behavior of U.S. households evolves
with age. Figure 8 displays home-ownership rates, as well as the logarithm of non-
durable consumption, the ratio of debt to net wealth, and household net wealth. The
model does a reasonably good job at accounting for the key features of these data.
Notably home ownership rates rise with age and stabilize when people reach their 40s,
both in the model and the data.
To understand the mechanisms that underlie the dynamics of home-ownership, it is
useful to consider a simplified analysis of the cost of owning versus renting.15 The net
benefit of owning a home is given by:
✓ ◆
prt pt+1 pt bt bt m new
(hown )
+ Et rt 1 r rt . (14)
pt pt pt pt t pt

The first term in equation (14) is the savings from not paying rent, which we express
as a fraction of the house price, prt /pt . In our sample, prt /pt is on average 7.7 percent.
14
Returns to housing in the model are computed as log [(prt + pt+1 pt ) /pt ].
15
See Diaz and Luengo-Prado (2012) for a review of the literature on the user cost of owning a home.

28
The second term in this expression is the expected real rate of housing appreciation. In
our calibration, the average value of Et (pt+1 pt ) /pt is one percent per year. The third
term is the opportunity cost of the down payment, 1 bt /pt on a house. The fourth
term is the mortgage payment on the house, where rtm denotes the average mortgage
rate. We estimate that the average value of rt and rtm in our sample is 3.5 percent and
6.5 percent, respectively. The fifth term, , is the rate of depreciation of the housing
stock. We assume that is three percent per annum. The last term in equation (14)
is the fixed cost of buying a house as a percentage of the house price. One di↵erence
between renting and buying not captured by equation (14) is that renters can freely
vary the amount of housing services they purchase while home owners have to pay a
fixed cost to change the size of their house.
A number of observations follow from equation (14). First, other things equal, the
higher is the rental-price ratio and the expected real rate of housing appreciation, the
more attractive it is to own rather than rent a house. Second, other things equal, the
less expensive is the house (i.e. the lower is pt ) the larger is the negative impact of a fixed
new
cost on the desirability of purchasing a home (rt (hown )/pt ). Third, other things
equal, the higher the down payment a household can make, the more attractive it is to
own a home. To see this e↵ect, it is convenient to rewrite the sum of the opportunity
cost of the down payment and the mortgage payment, rt (1 bt /pt ) + (bt /pt ) rtm as:
bt m
rt + (r rt ) . (15)
pt t
The first term (rt ), is the opportunity cost of purchasing a home without a mortgage.
The second term, is the additional interest costs associated with buying a home with a
mortgage of size b, which requires paying the spread (rtm rt ). From the second term,
it is clear that, other things equal, the bigger is the mortgage the less desirable it is to
buy a home.
With these observations as background, consider again Figure 8. The model implies
that home ownership rates rise as people get older. This result follows from the fact

29
that, on average, income rises as a person ages, peaking between 45 and 55 years of
age. As income rises, people want to live in bigger homes, which reduces the impact of
new
fixed costs on the desirability of purchasing a home (rt (hown )/pt ). Also, as income
rises, people can a↵ord bigger down payments on those homes, which reduces the user
cost of owning a home. Taken together, both forces imply that home ownership should
on average rise until people are 55. Thereafter, home ownership rates roughly stabilize.
However, many elderly homeowners downsize. They sell their old homes and use the
proceeds to buy smaller homes which they eventually leave as bequests.
From Figure 8, we also see that household debt declines with age. This fact reflects
two forces. First, people pay down their mortgages over time reducing their debt.
Second, elderly people who are downsizing have small mortgages. Finally, household
net wealth rises on average with age, as people pay o↵ their mortgages and save for
bequests.
Figure 8 also shows that non-durable consumption rises until people reach ages
45 to 55 and then falls. The rise results from two forces. First, people face borrow-
ing constraints which prevent them from borrowing against future earnings. Second,
most households have an incentive to save so they can make a down payment on their
mortgage. The fall in non-durable consumption after age 55 reflects the presence of a
bequest motive. As people age, the weight of expected utility from leaving bequests
rises relative to the weight of utility from current consumption. When we reduce B,
the parameter that controls the strength of the bequest motive, consumption becomes
smoother.

7.2 Refinancing and the interest-rate gap

In the data, the average annual refinancing rate is 7.9 percent with a standard deviation
of 4 percent. In the model, the average annual refinancing rate is 7.9 percent. So,
taking sampling uncertainty into account, the model does a good job of accounting for
the average refinancing rate. The model is also consistent with the fraction of new

30
mortgages issued in each period.16 This fraction is 25 percent both in the model and
the data.
Figure 9 plots the cumulative distribution function of refinanced loans as a function
of the interest-rate gap faced by people in the economy. We display these statistics
both for the data and the model. The data-based statistics are computed as follows.
We bin all the loans according to the interest-rate gap ranges indicated in the figure.
For every bin, we calculate the fraction of loans that were refinanced. Figure 9 displays
these fractions.
The model-based statistics are computed as follows. The initial distribution of age,
assets, mortgage debt and mortgage rates is the same as the actual distribution in
1994. Every period a new cohort of households enters the economy. New households
are randomly assigned to being homeowners or renters in a way that is consistent with
the initial asset distribution. The mortgage rates of home owners in the new cohorts are
drawn from the distribution of new mortgage rates in the data at every point in time.
We assume there are 100, 000 households in the model economy and draw idiosyncratic
shocks for each of these people. At each point in time, we feed in the actual values of the
aggregate state of the economy from 1995 to 2007 for rt , yt , and pt . We use the model to
construct time series for at , the logarithm of economy-wide average positive savings from
refinancing. People use this variable to form expectations for future aggregate states
using the estimated version of equation (13). At every point in time, from 1995 to 2007,
the model generates a distribution of interest-rate gaps and refinancing decisions. So,
we are able to compute the same moments that we estimated from the data. As can
be seen from Figure 9, the model does reasonably well at accounting for the data.
Andersen et al. (2015) and the references therein show that some people do not re-
finance their fixed-rate mortgage when market rates fall below their locked-in mortgage
16
New mortgages include refinancing of existing mortgages and mortgages issued to people who buy
a new home. This pool of people includes people who were renters, new cohorts who enter the housing
market, and people who upgrade or downgrade the size of their house.

31
rate. This phenomenon is more pronounced at the top end of the interest rate gap.
We could account for this “burnout” phenomenon by introducing heterogeneity in re-
financing costs. However, this additional complexity is unlikely to change the model’s
implications for the state-dependent impact of monetary policy. The reason is that
relatively few loans in our sample exhibit the burnout phenomenon. For example, the
fraction of initial mortgages that are never prepaid is 0.3 percent.
We conclude by presenting time-series evidence on the model’s implications for
refinancing activity. The blue line in Figure 10 displays the fraction of loans that were
refinanced in the U.S. between 1998 and 2007. The dashed line displays the analogue
model time series. The model does reasonably well at capturing the broad movements
in the time series, such as the run up in refinancing until 2004 which is associated with
large declines in mortgage rates. The model also captures the subsequent decline in
refinancing activity. The correlation between the data and model-implied paths is 80
percent.

7.3 State dependency of refinancing decisions

We now assess the ability of the model to account for the state-dependent nature of
the e↵ects of monetary policy on refinancing decisions. Using data simulated from
the model, we estimate the following regression using the monetary policy shock as an
instrument:
⇢t+4 = 0 + 1 RtM + 2 RtM ⇥ At 1 + 3 At 1 + ⌘t . (16)

This equation is a version of regression (2) without county fixed e↵ects.17 The variable
⇢t+4 is the fraction of mortgages refinanced in the economy between quarters t and t+4.
Table 7, panel A reports the model-based and data-based estimates of 1 and 2. The
data estimates are reproduced from panel A of Table 2. The model does quite well
at accounting for 2 which governs the state dependent impact of a monetary policy
17
To make the model regressions comparable with the data regressions we include a one-year lag of
the refinancing rate.

32
shock. At the same time, the model somewhat overstates 1, which governs the direct
impact of interest rates on refinancing rates.
Another way to assess the implications of the model is to calculate the e↵ect of a
50 basis points reduction in the mortgage rate on the refinancing rate at each point in
time. Our results are displayed in Figure 4 along with the regression-based estimates
of these e↵ects. Taking sampling uncertainty into account, the model does reasonably
well at tracking the regression-based estimates. Even in periods where the model does
less well, its implications are only a few basis points outside the confidence interval.
We also compute the time series of four cross-sectional moments for the interest rate
gap in the data and the model: the top 25 percentile, the median, the mean, and the
bottom 25 percentile. The time-series correlation between these moments in the model
and the data is 80 percent, 70 percent, 73 percent, and 90 percent, respectively.

7.4 State dependency of mortgages for new home purchases

It is of interest to ask whether the model accounts for the response of purchases of
new homes to an exogenous change in mortgage rates. To this end, we run a version of
regression (2) on data simulated from the model. The dependent variable is the fraction
of new mortgages used to purchase a new home relative to the number of outstand-
ing mortgages. Table 7, panel C reports our results. Both regression coefficients are
statistically significant at the one-percent level. Also, there is strong evidence of state
dependency in the response of new home purchases to changes in mortgage rates.
To interpret these coefficients, suppose that all the independent variables in the
regression are initially equal to their time-series averages and that the average interest-
gap is initially equal to its mean value of 14 basis points. The estimates in column
1 of Panel C of Table 7 imply that a 25 basis points drop in mortgage rates raises
the fraction of loans for new purchases by 3.5 percent. Now suppose that the drop in
mortgage rates happens when the average interest-rate gap is equal to 56 basis points.
Then, a 25 basis points drop in mortgage rates raises the fraction of loans for new

33
purchases to 6.1 percent. So, the marginal impact of a one standard-deviation increase
in the average interest-rate gap is 2.6 percent.
Column 2 of Table 7, panel C shows that the regression coefficients implied by our
model are consistent with the empirical patterns discussed above. However, the model
somewhat understates the direct impact of a change in mortgage rates and the state
dependency of new home purchases.

7.5 Endogenous versus exogenous sources of state dependency

Our model implies a non-linear representation of the data which involves state depen-
dency. So, our theory compels us to estimate a VAR that allows for state dependency.
The VAR which we use to compute agents’ expectations is an exogenous source of state
dependency. But the model embodies other strong, endogenous sources of state depen-
dency. The most important of these sources is the presence of fixed costs of refinancing
or getting a mortgage for a new home.
To substantiate these claims, we shut down the exogenous state-dependence in the
VAR by setting the relevant coefficients to zero. We then re-solve the model using the
resulting linear VAR and re-calculate the statistics reported in Table 7. The results are
reported in column 3 of that Table. Consider the regression results for the fraction of
loans that are refinanced. From panel A, we see that the coefficient on the interaction
term ( Rt ⇥ Average Rate Gap) declines by roughly 33 percent. So, 33 percent of the
state dependence comes from the VAR and two thirds is generated endogenously by the
model. Roughly similar results hold for the fraction of loans that are cash-refinanced
(panel B) and the fraction of loans for home purchases (panel C). Taken together these
results demonstrate that it is quantitatively important to allow for state dependency in
the VAR. At the same time, most of the state dependency is endogenously generated
by the structural model.

34
8 Model implications
In this section, we use our model to study the state-dependent e↵ects of a fall in
interest rates on consumption and how the potency of monetary policy depends on the
past behavior of interest rates.

8.1 State-dependent e↵ects of a fall in interest rates on con-

sumption
We now use model-simulated data to estimate the e↵ect of an exogenous change in the
interest rate on the annual change in the logarithm of consumption for household j
(cjt ):
cjt+1 cjt = j0 + 1 RtM + 2 RtM ⇥ At 1 + 3 At 1 + ⌘tc . (17)

The coefficients in this regression are estimated using the monetary shocks as instru-
ments. Table 8 shows the e↵ect of a 50 basis points fall in interest rates. The total
e↵ect on consumption of an exogenous change in mortgage rates is 1.43 percent. The
direct e↵ect ( 1 RtM ) is 0.97 percent. The state dependent e↵ect ( 2 RtM ⇥ average
interest-rate gap) is 0.46 percent.18
To understand the mechanisms that underlie these e↵ects, we estimate regression
(17) for two separate groups: households that have positive liquid assets (sjt > 0) and
households that do not have positive liquid assets (sjt  0). We call the first group
of households unconstrained and the second group constrained. Forty eight percent
of households are, on average, constrained in our model. This fraction is consistent
with the results in Kaplan, Violante and Weidner (2014). More than 80 percent of the
constrained households are home owners. These households correspond to what Kaplan,
Violante and Weidner (2014) call wealthy hand-to-mouth consumers. The total e↵ect
18
Very little of the consumption response to mortgage rate changes is driven by the associated
changes in house prices. We establish this result by computing the response of consumption to a
change in mortgage rates keeping house prices constant. The implied consumption response is similar
to that obtained when we do allow house prices to change.

35
on consumption of an exogenous change in mortgage rates is 4.6 and 0.78 percent for
constrained and unconstrained households, respectively. So, the consumption response
is predominantly driven by the constrained households.
Roughly 84 percent of the households who refinance engage in cash-out refinancing.
This value is in line with the evidence presented by Chen, Michaux, and Roussanov
(2020). Using a conservative estimate based on conforming mortgages, these authors
argue that, over the period 1993-2010, on average about 70 percent of refinanced loans
involve cash-out.
In response to a one-percent decline in mortgage rates, households who engage in
cash-out refinancing in our model increase their loan balances by 19.2 percent. This
e↵ect is broadly consistent with the empirical estimates of Bhutta and Keys (2016).19
To assess the model’s implications for the state-dependent nature of cash-out re-
financing, we use model-simulated data to estimate a version of regression (16) using
monetary policy shocks as instruments. Here, the dependent variable is the fraction
of total loans with cash-out refinancing. Our results are reported in Table 7, panel B.
Comparing columns II in panels A and B, we see that in the data the state-dependent
e↵ect of an interest rate cut on refinancing and cash-out refinancing is about the same.
The model captures, qualitatively, the state-dependent nature of cash-out refinancing,
i.e. the larger are potential savings, the larger is the response of cash-out refinancing
to an interest rate cut.
Our model abstracts from the e↵ects of refinancing decisions on bank owners. If
those owners are constrained and the profits of the bank rise or fall one to one by the
amount that consumers save by refinancing, the refinancing channel has no aggregate
e↵ect on consumption. However, it is natural to assume that bank owners behave like
unconstrained households. Under this assumption, the negative e↵ect of refinancing on
the consumption of bank owners is much smaller in absolute value than the positive
19
Using Equifax data, these authors estimate that, in response to a one-percent decline in mortgage
rates, households who engage in cash-out refinancing increase their loan balances by 23 percent.

36
e↵ect on the consumption of constrained households.20 As a result, the overall e↵ect
of refinancing on aggregate consumption is positive. To explore the potential size of
this e↵ect we do a simple back-of-the-envelope calculation. We compute the change
in consumption implied by a permanent-income-style calculation by multiplying the
present value of total savings from refinancing by the steady-state risk-free rate (3.5
percent). We subtract this value from total consumption. This adjustment has a
negligible impact with the growth rate of consumption falling by less than 0.01 percent.
It is possible that lenders respond to a fall in their income by reducing the flow
of credit to borrowers. The precise way in which interest rates and the quantity of
credit change adjust depends on monetary policy, the nature of financial frictions and
whether the economy is open or closed. For example, in an open economy like the U.S.,
international capital flows could adjust to compensate for any decline in the flow of
credit from domestic lenders to domestic borrowers.
Finally, we do not explicitly model the wealth e↵ects of duration risk on mortgage
investors. This risk is embedded in the mortgage rates that we use to estimate equation
(7). In solving the model, the mortgage rate that people face and their expectations
of future mortgage rates reflect changes in the duration-risk premia arising from the
state of the economy. So, even though we don’t explain those premia, we do account
for them when solving the model.

8.1.1 The importance of life-cycle dynamics

Life-cycle dynamics play an important role in our quantitative results. To illustrate this
role, we consider an alternative version of the model with no income life-cycle dynamics.
Unlike in our benchmark model, young people have no systematic incentive to borrow
against future income. As a result, mortgage balances are smaller and fewer people are
financially constrained. So, the gains from refinancing are smaller.
20
The negative e↵ect on U.S. consumption of the decline in profits due to refinancing is mitigated
by the fact that some of stock shares of U.S. banks are owned by foreigners.

37
 In this alternative economy, both the direct and state-dependent e↵ect of monetary
policy on refinancing are smaller. The direct e↵ect of a change in mortgage rates
as measured by 1 (the coefficient on RtM in equation (2)) drops from 0.06 in the
benchmark model to roughly zero. The state dependent e↵ect as measured by 2 (the
coefficient on RtM ⇥ average interest-rate gap in equation (2)) drops from 0.232 to
0.081. We obtain similar results for the fraction of loans that are cash-out refinanced.

8.2 State dependency and the potency of monetary policy

In this subsection, we provide intuition for the state-dependent e↵ects of monetary
policy in our model by comparing the impact of a given interest rate cut in di↵erent
scenarios. We then use our model to quantify an important cost of prolonged low-
interest rate periods.

8.2.1 Model experiments

In all of the experiments considered in the subsection the model economy starts in
steady state, i.e. the aggregate state variables have been constant and equal to their
unconditional means. However, people have been experiencing ongoing idiosyncratic
shocks to their income.
The three paths that we consider are displayed in Figure 11. At each point in
time, people form expectations about aggregate states according to the Markov-chain
approximation to the demeaned level representation of the aggregate states associated
with (13). Our results are summarized in Panel A of Table 9.
In the first scenario, we consider the e↵ect of an interest rate cut when the economy
starts in steady state and remains there until period four. In period five, we feed an
interest-rate shock into the model that generates a 50-basis point fall in the interest
rate. We refer to this scenario as the benchmark scenario. From row (i) of Table 9,
Panel A we see that 18 percent of people refinance in the impact period of the shock and
aggregate consumption increases 1.7 percent. There are two reasons why these e↵ects

38
are so large. First, all existing homeowners with a mortgage have a positive rate gap
after the interest rate cut because they obtained their mortgages at the steady-state
mortgage interest rate. Second, people expect the interest rate to revert to the mean,
so period seven is a good time to refinance.
In the second scenario, the central bank steadily raises interest rates starting in
period one until they peak in period four. The central bank then cuts the interest rate
by 50 basis points in period five. From row (ii) of Table 9, Panel A we see that only
7 percent of households refinance in the impact period of the shock and there is only
a 0.2 percent rise in consumption. The reason for these small e↵ects is that only 23
percent of people face a positive interest-rate gap in period five. These are the people
who entered new mortgages despite rising interest rates due to life-cycle considerations
or idiosyncratic income shocks.
In the third scenario, the central bank steadily lowers interest rates starting in
period one until they trough in period four. The central bank then cuts interest rates
by 50 basis points in period five. From row (ii) of Table 9, Panel B we see that in this
scenario, 12 percent of people refinance in the impact period of the shock and there is
a 0.6 percent rise in consumption. The consumption e↵ect is smaller than in the first
scenario because a subset of people refinanced as interest rates declined and engaged
in cash-out refinancing. Those people are generally not liquidity constrained in period
seven.21
These results show that, in our model, the current impact of monetary policy
through the refinancing channel depends on the past actions of the Fed. The fun-
damental reason is that those actions a↵ect the distribution of potential savings from
refinancing.
21
The average interest rate is a key determinant of when households refinance. When they do so,
it is often optimal to take advantage of the new loan and cash out. Because of the fixed costs of
refinancing, it is not in general optimal to refinance just to cash out.

39
8.2.2 A downside of long periods of low interest rates

Here we use our model to quantify an important cost of keeping interest rates low for
a long period of time: it makes monetary policy less powerful for an extended period
thereafter.
We begin by addressing the question: after rates have been normalized, when does
monetary policy regain its initial potency? To address this question we consider the
paths displayed in Figure 12. In all of these cases, the economy starts from steady state
and the interest rate falls from 3.5 percent to 1 percent for four periods. The interest
rate then normalizes back to 3.5 percent. The di↵erence between the cases is that t
periods after the normalization, t 2 {1, 2, 3}, the interest rate falls by 50 basis points.
Results are reported in Table 10. From this table we see that aggregate consumption
rises by 0.8, 1.1 and 1.9 percent for t = 1, 2, and 3, respectively. So the sooner the
interest rate cut after normalization occurs, the smaller is its impact. For reference,
recall that aggregate consumption rises by 1.7 percent if the interest rate falls by 50
basis points in the benchmark scenario where the economy is in steady state. So, the
potency of an interest rate cut is substantially reduced for the first two years after the
interest rate is normalized.22
The key factor driving this result is that fewer people face a positive interest-rate
gap when the interest rate is cut relative to the benchmark scenario. Many people
face a negative-rate gap because they entered mortgages at rates that were lower than
the steady-state mortgage rate. So fewer people have an incentive to refinance their
mortgages in period seven than in the benchmark scenario.
Over time, people enter new loans in response to life-cycle-related income changes
and idiosyncratic-income shocks. So, the share of people with a mortgage rate equal to
22
The results of this experiment are not driven by the model’s low burnout rate. We could match
the empirical burnout rate by adding large refinancing costs for a subset of the population. But if
people with high refinancing costs did not refinance between periods 2 and 10, they would be unlikely
to refinance in period 11 through 14.

40
the steady-state rate increases over time. That increase in turn implies that a larger
fraction of people have a positive-rate gap after a 50 basis points rate cut. The potency
of an interest rate cut rises over time. According to Table 10, it takes roughly three
years for monetary policy to have the same e↵ect on consumption as in the benchmark
case.
Consistent with the post-normalization results, our model also implies that mon-
etary policy is less potent after a series of interest declines. To illustrate this point,
suppose that starting from steady state, the interest rate falls from 3.5 percent to 1 per-
cent and stays at 1 percent between periods two and six.23 Then, in period seven, the
interest rate falls by an additional 50 basis points (not shown on the Figure). According
to our model, only 5 percent of people refinance in the impact period of the shock and
there is only a 0.5 percent rise in consumption. These modest e↵ects contrast sharply
with the e↵ect of an interest rate fall in the benchmark scenario where 16 percent of
people refinance in the impact period of the shock and there is a 1.7 percent rise in
consumption. The intuition for this result is the same as the one underlying Panel B of
Table 9. In this exercise, most people already have low mortgage rates either because
they refinanced an existing mortgage or purchased a new home. Either way, the stock
of people with a large, positive interest rate gap is small. For this reason, a further
interest rate cut has a small impact on refinancing and consumption.
What can policy makers do to deal with the potency problem? One possibility is
to take advantage of the nonlinear response of consumption to a fall in the interest
rate. Recall that a 50 basis point interest rate cut in the first period after interest rates
normalize leads to a 1.7 percent rise in consumption. Suppose instead that the interest
rate fell by 100 basis points. Then consumption would rise by 2.9 percent. This increase
is roughly the same if, starting from steady state, the interest rate falls by 100 basis.
23
As above, we model changes in interest rates via sequences of interest rate shocks. In all cases,
agents form expectations about aggregate states according to the Markov-chain approximation to the
demeaned level representation of the aggregate states associated with equation (13).

41
 These results suggest that policy makers can deal with the potency problem in one
of two ways. If they cut interest rates by relatively small amounts, e.g. 50 basis points,
then they wait until policy regains the same impact as in the steady state. However,
if they are prepared to cut interest rates by large amounts, e.g. 100 basis points, the
potency problem is not an issue as long as monetary policy is not constrained by the
e↵ective lower bound (ELB).
There is considerable debate about the importance of the ELB. For example, De-
bortoli, Gali, and Gambetti (2020) argue that the ELB was not a binding constraint
in the aftermath of the financial crises. The conventional view (e.g. Eggertsson (2003),
Eggertsson and Woodford (2004), and Christiano, Eichenbaum and Rebelo (2011)) is
that the ELB does constrain monetary policy. Swanson and Williams (2014) argue that
the ELB was not binding before 2011 but did constrain monetary policy thereafter. Ki-
ley and Roberts (2017) argue that the ELB will be binding 30 to 40 percent of the time
in the future. To the extent that the ELB is a constraint on monetary policy, it would
be difficult for the Fed to lower rates by large amounts after a prolonged period of low
interest rates.

9 Conclusion
This paper provides evidence that the efficacy of monetary policy is state dependent,
varying in a systematic way with the pool of savings from refinancing. We construct
a quantitative life-cycle model of refinancing decisions that is consistent with the facts
that we document.
Our model points to an important cost of fighting recessions with a prolonged period
of low interest rates. Such a policy reduces the potency of monetary policy in the
period after interest rates are normalized. So, if the economy is a↵ected by a negative
shock during that period, policy makers will have less ammunition at their disposal to
counteract the e↵ects of that shock. This observation raises the conundrum: should

42
monetary policy makers use their ammunition to fight an ongoing recession or the next
one?

10 References
Agarwal, Sumit, John Driscoll, and David Laibson, 2013, “Optimal Mortgage Refinanc-
ing: A Closed Form Solution,” Journal of Money, Credit, and Banking 45, 591—622.
Andersen, S., Campbell J., Nielsen K., and Ramadorai T. “Inattention and Inertia
in Household Finance: Evidence from the Danish Mortgage Market,” National Bureau
of Economic Research working paper No. 21386, 2015.
Auclert, Adrien “Monetary Policy and the Redistribution Channel,” manuscript,
Stanford University, 2017.
Bailey, Michael, Cao, Rachel, Kuchler, Theresa, Stroebel, Johannes, and Wong, Ar-
lene. Replication data for: “Social Connectedness: Measurement, Determinants, and
E↵ects.” American Economic Association, 2018. Ann Arbor, MI: Inter-university Con-
sortium for Political and Social Research, 2019-10-12. https://doi.org/10.3886/E114016V1
Beraja, Martin, Andreas Fuster, Erik Hurst, and Joseph Vavra “Regional Hetero-
geneity and the Refinancing Channel of Monetary Policy,” Quarterly Journal of Eco-
nomics, 2018.
Berger, David and Joseph Vavra, “Consumption Dynamics During Recessions,”
Econometrica, 2015, 83 (1), 101—154.
Berger, David, Guerrieri, Veronica, Lorenzoni, Guido and Vavra, Joseph, 2017.
“House Prices and Consumer Spending.” The Review of Economic Studies, 85(3),
pp.1502-1542.
Berger, David, Konstantin Milbradt, Fabrice Tourre, and Joseph Vavra (2018),
“Mortgage Prepayment and Path-Dependent E↵ects of Monetary Policy,” manuscript,
Northwestern University.
Bhutta, Neil and Benjamin J. Keys. 2016. “Interest Rates and Equity Extraction

43
during the Housing Boom.” American Economic Review 106 (7):1742–1774.
Board of Governors of the Federal Reserve System (US), Federal Funds E↵ective
Rate [DFF], retrieved from FRED, Federal Reserve Bank of St. Louis; https://fred.stlouisfed.org/series/D
2021.
Board of Governors of the Federal Reserve System (US), 3-Month Treasury Bill
Secondary Market Rate [TB3MS], retrieved from FRED, Federal Reserve Bank of St.
Louis; https://fred.stlouisfed.org/series/TB3MS, 2021.
Chen, Hui, Michael Michaux, and Nikolai Roussanov. “Houses as ATMs: Mortgage
Refinancing and Macroeconomic Uncertainty,” The Journal of Finance 75, no. 1 (2020):
323-375.
Christiano, Lawrence, Martin Eichenbaum, and Sergio Rebelo. “When is the Gov-
ernment Spending Multiplier Large?” Journal of Political Economy 119, no. 1 (2011):
78-121.
Cochrane, John H., and Monika Piazzesi. ”The Fed and Interest rates-a High-
frequency Identification.” American Economic Review 92, no. 2 (2002): 90-95.
Corelogic Loan-Level Analytics Mortgage Data and Corelogic Origination data, ac-
cessed 2018/01/22.
Debortoli, Davide, Jordi Gali, and Luca Gambetti. “On the Empirical (Ir) relevance
of the Zero Lower Bound Constraint,” NBER Macroeconomics Annual 34, no. 1 (2020):
141-170.
DeFusco, Anthony A. and John Mondragon “No Job, No Money, No Refi: Frictions
to Refinancing in a Recession,” manuscript, Northwestern University, 2018.
De Nardi, Mariacristina and Fang Yang “Bequests and Heterogeneity in Retirement
Wealth,” European Economic Review, Elsevier, vol. 72(C), pages 182-196, 2014.
Di Maggio, Marco, Amir Kermani, Benjamin J. Keys, Tomasz Piskorski, Rodney
Ramcharan, Amit Seru, and Vincent Yao “Interest Rate Pass-through: Mortgage rates,
Household Consumption, and Voluntary Deleveraging,” American Economic Review
107, no. 11 (2017): 3550-88.

44
 Diaz, Antonia, and J. Luengo-Prado, Maria. 2012. “The User Cost, Home Owner-
ship and Housing Prices: Theory and Evidence from the US,” International Encyclo-
pedia of Housing and Home, 228-234.
Diaz, Antonia and Maria J. Luengo-Prado “The User Cost, Home Ownership and
Housing Prices: Theory and Evidence from the U.S.” in Smith, S.J.; Elsinga, M.;
Fox-O’Mahony, L.; Ong, S.E.; Wachter, S. (editors) The International Encyclopedia of
Housing and Home, Elsevier (2012).
Diep, Peter, Andrea L. Eisfeldt, Scott Richardson, 2017, “Prepayment Risk and
Expected MBS Returns,” NBER Working Paper No. 22851
Doepke, Matthias, and Martin Schneider. “Inflation and the Redistribution of Nom-
inal Wealth.” Journal of Political Economy 114, no. 6 (2006): 1069-1097.
Doepke, Matthias, Martin Schneider, and Veronika Selezneva. ”Distributional Ef-
fects of Monetary Policy,” manuscript, Northwestern University, 2018.
Dunn, Kenneth B., and John J. McConnell. “Valuation of GNMA Mortgage-backed
Securities,” The Journal of Finance 36, no. 3 (1981): 599-616.
Eggertsson, Gauti B. “Zero Bound on Interest Rates and Optimal Monetary Policy,”
Brookings papers on economic activity 2003, no. 1 (2003): 139-233.
Eggertsson, Gauti B., and Michael Woodford. “Policy Options in a Liquidity Trap,”
American Economic Review 94, no. 2 (2004): 76-79.
Federal Reserve Bank of Cleveland, 30-year expected inflation. https://www.clevelandfed.org/our-
research/indicators-and-data/inflation-expectations.aspx., 2021.
Federal Reserve Bank of Philadelphia. 2021. “Survey of Professional Forecasters.”
Freddie Mac Single Family Loan-Level Dataset and Quarterly Refinance Report.
Garriga, C. and Hedlund, A., 2017. Mortgage Debt, Consumption, and Illiquid
Housing Markets in the Great Recession. FRB St. Louis Working Paper, (2017-30).
Garriga, Carlos, Finn E. Kydland, and Roman Sustek. “Mortgages and monetary
policy.”The Review of Financial Studies 30, no. 10 (2017): 3337-3375.
Gertler, Mark, and Peter Karadi. “Monetary Policy Surprises, Credit Costs, and

45
Economic Activity.” American Economic Journal: Macroeconomics 7.1 (2015): 44-76.
Gilchrist, Simon, David López-Salido, and Egon Zakrajek. “Monetary Policy and
Real Borrowing Costs at the Zero Lower Bound,” American Economic Journal: Macroe-
conomics 7, no. 1 (2015): 77-109.
Global Financial Data Real Estate database.
Gorodnichenko, Yuriy, and Michael Weber “Are Sticky Prices Costly? Evidence
from the Stock Market.,” American Economic Review 106, no. 1 (2016): 165-99.
Green, Jerry, and John B. Shoven. 1986. “The E↵ects of Interest Rates on Mortgage
Prepayments,” Journal of Money, Credit and Banking 18, no. 1: 41-59.
Greenwald, Daniel “The Mortgage Credit Channel of Macroeconomic Transmis-
sion,” manuscript, Sloan School of Management, MIT, 2016.
Guren, A.M., Krishnamurthy, A. and McQuade, T.J., 2018. “Mortgage Design in
an Equilibrium Model of the Housing Market ” (No. w24446). National Bureau of
Economic Research.
Guren, A.M., McKay, A., Nakamura, E. and Steinsson, J., 2018. “Housing Wealth
E↵ects: The Long View ” (No. w24729). National Bureau of Economic Research.
Gürkaynak, Refet S., Brian Sack, and Eric Swanson. “The Sensitivity of Long-
term Interest Rates to Economic News: Evidence and Implications for Macroeconomic
Models.” American Economic Review 95, no. 1 (2005): 425-436.
Guvenen, Fatih, and Anthony A. Smith. 2014. “Inferring Labor Income Risk and
Partial Insurance from Economic Choices.” Econometrica, 82(6): 2085–2129.
Guvenen, Fatih, Fatih Karahan, Serdar Ozkan, and Jae Song. 2015. “What Do
Data on Millions of U.S. Workers Reveal about Life-Cycle Earnings Risk?” National
Bureau of Economic Research Working Paper 20913.
Huggett, Mark, ”Wealth Distribution in Life-cycle Economies”, Journal of Monetary
Economics, Volume 38, Issue 3, December 1996, Pages 469-494.
Hurst, Erik, Keys, Benjamin, Seru, Amit and Vavra, Joseph, “Regional Redistribu-
tion Through the U.S. Mortgage Market,” American Economic Review, 2016, 106(10),

46
2982-3028.
Iacoviello, M., 2005. House Prices, Borrowing Constraints, and Monetary Policy in
the Business Cycle. American Economic Review, 95(3), pp.739-764.
Iacoviello, M. and Pavan, M., 2013. Housing and Debt over the Life Cycle and Over
the Business Cycle. Journal of Monetary Economics, 60(2), pp.221-238.
Khandani, Amir E., Andrew W. Lo, and Robert C. Merton, 2013, “Systemic Risk
and the Refinancing Ratchet E↵ect”, Journal of Financial Economics 108, 29-45.
Kaplan, Greg, Kurt Mitman and Giovanni L. Violante “The Housing Boom and
Bust: Model Meets Evidence,” manuscript, Princeton University, 2017.
Kaplan, Greg, Benjamin Moll and Gianluca Violante “Monetary Policy According
to HANK,” American Economic Review, 108(3), 697-743, 2018.
Kaplan, Greg and Giovanni L. Violante & Justin Weidner, 2014. “The Wealthy
Hand-to-Mouth,” Brookings Papers on Economic Activity, vol 2014(1), pages 77-138.
Kiley, Michael T., and John M. Roberts. ” Monetary Policy in a Low Interest Rate
World.“Brookings Papers on Economic Activity 2017, no. 1 (2017): 317-396.
Kuttner, Kenneth N. ”Monetary Policy Surprises and Interest Rates: Evidence from
the Fed Funds Futures Market.” Journal of Monetary Economics 47, no. 3 (2001): 523-
544.
LaCour-Little, Michael. “The Evolving Role of Technology in Mortgage Finance.”
Journal of Housing Research (2000): 173-205.
Landvoigt, Tim, Monika Piazzesi, and Martin Schneider. 2015. “The Housing
Market(s) of San Diego.” American Economic Review, 105(4): 1371-1407.
Mack, A., and E. Martı́nez-Garcı́a. 2011. ”A Cross-Country Quarterly Database
of Real House Prices: A Methodological Note.” Globalization and Monetary Policy
Institute Working Paper No. 99, Federal Reserve Bank of Dallas.
McKay, A., Nakamura, E. and Steinsson, J., 2016. “The Power of Forward Guidance
Revisited,” American Economic Review, 106(10), pp.3133-58.
Mian Atif, Kamelesh Rao and Amir Sufi “Household Balance Sheets, Consumption,

47
and the Economic Slump” Quarterly Journal of Economics, 2013, 1687-1726
Mian, Atif, and Amir Sufi. 2014. “House Price Gains and U.S. Household Spending
from 2002 to 2006.” National Bureau of Economic Research, Inc NBER Working Papers
20152.
Nakamura, Emi and Jón Steinsson “High Frequency Identification of Monetary Non-
Neutrality: The Information E↵ect. ” Quarterly Journal of Economics, 133(3), 1283-
1330, August 2018.
OECD. 2021, Housing prices (indicator). doi: 10.1787/63008438-en
Piazzesi, M. and M. Schneider (2016): “Housing and Macroeconomics,” Handbook
of Macroeconomics, 2, 1547–1640.
Rios-Rull, J.V. and Sanchez-Marcos, V., 2008 “An Aggregate Economy with Di↵er-
ent Size Houses,” Journal of the European Economic Association, 6(2-3), pp.705-714.
Rouwenhorst, Geert K., “Asset Pricing Implications of Equilibrium Business Cycle
Models,” in Thomas Cooley, ed., Structural Models of Wage and Employment Dynam-
ics, Princeton: Princeton University Press, 1995.
Scharfstein, David and Adi Sunderam, “Market Power in Mortgage Lending and
the Transmission of Monetary Policy,” 2013 NBER working paper 19156.
Schwartz, Eduardo S. and Walter N. Torous 1989. “Prepayment and the Valuation
of Mortgage-Backed Securities.” Journal of Finance 44(2), pp. 375-392.
Stock, James H., and Motohiro Yogo. 2005. Testing for Weak Instruments in Linear
IV regression. In Identification and Inference for Econometric Models: Essays in Honor
of Thomas Rothenberg, ed. D. W. K. Andrews and J. H. Stock, 80–108. New York:
Cambridge University Press.
Swanson, Eric T., and John C. Williams. “Measuring the E↵ect of the Zero Lower
Bound on Medium-and Longer-Term Interest Rates.” American Economic Review 104,
no. 10 (2014): 3154-85.
U.S. Bureau of Labor Statistics, Local Area Unemployment Statistics, Unemploy-
ment rate, not seasonally adjusted.

48
 U.S. Bureau of Economic Analysis, Personal Income, Population, Per Capita Per-
sonal Income, and Per capita personal income.
U.S. Bureau of Economic Analysis, Real Disposable Personal Income: Per Capita
[A229RX0Q048SBEA], retrieved from FRED, Federal Reserve Bank of St. Louis;
https://fred.stlouisfed.org/series/A229RX0Q048SBEA, 2021.
U.S. Board of Governors of the Federal Reserve System, “Survey of Consumer Fi-
nances”, https://www.federalreserve.gov/econres/scfindex.htm., 2021.
U.S. Census Bureau. 2016. “Building Permits Survey.”
Woodford, M., 2012 “Methods of Policy Accommodation at the Interest-rate Lower
Bound.” In Proceedings-Economic Policy Symposium-Jackson Hole (pp. 185-288). Fed-
eral Reserve Bank of Kansas City.
Wong, Arlene. 2020 “Population Aging and the Transmission of Monetary Policy
to Consumption,” manuscript, Princeton University.

49
Tables and Figures

Table 1: First-stage estimates

Notes: Regression equation (2), first-stage estimates based on futures shock. Standard errors are in
parentheses. ⇤ , ⇤⇤ , and ⇤⇤⇤ give the significance at the 10, 5, and 1 percent levels.

1
 Table 2: State dependency of monetary policy and refinancing

Notes: The table reports the response to a decline in interest rates. It therefore reports the estimates
from regression equation (2), multiplied by -1. The IV is based on futures. Standard errors are in
parentheses. ⇤ , ⇤⇤ , and ⇤⇤⇤ give the significance at the 10, 5, and 1 percent levels.

2
 Table 3: State dependency of monetary policy, unemployment and housing permits

Notes: The table reports the response to a decline in interest rates. It therefore reports the estimates
from regression equations (4) and (5), multiplied by -1. IV is based on futures. Standard errors are in
parentheses. ⇤ , ⇤⇤ , and ⇤⇤⇤ give the significance at the 10, 5, and 1 percent levels.

Table 4: Model parameter values

Notes: Table depicts parameter values. See text for more detail.

3
 Table 5: Estimated Aggregate Process for Mortgage and Rental Rates

Regression with dependent variable

Variables 30-year ratet log rental ratet
log yt -3.475 0.843
(14.090) (0.119)
log rt 0.334 -0.002
(0.069) (0.001)
log pt -0.022
(0.014)
constant -0.030 3.187
(0.020) (0.488)

Notes: Standard errors are in parentheses. ⇤ , ⇤⇤ , and ⇤⇤⇤ give the significance at the 10, 5, and 1
percent levels. These are the estimated coefficients for equations (8) and (9). See text for more detail.

Table 6: Estimated Aggregate Process

Regression with dependent variable

Variables log yt log rt log pt log savingst
log yt-1 -0.297 13.520 -0.019 -3.486
(0.108) (2.845) (0.047) (2.365)
log rt-1 0.000 0.612 0.010 0.223
(0.002) (0.051) (0.002) (0.147)
log pt-1 0.131 2.461 0.810 0.928
(0.075) (2.114) (0.070) (1.710)
log savingst-1 -0.009 0.157 -0.030 -0.529
(0.006) (0.142) (0.008) (0.296)
log savingst-1 x log rt-1 -0.002 0.133 -0.009 -0.189
(0.002) (0.035) (0.002) (0.086)

Notes: Regression equation (15). Standard errors are in parentheses. ⇤ , ⇤⇤ , and ⇤⇤⇤ give the significance
at the 10, 5, and 1 percent levels. See text for more detail.

4
 Table 7: State dependency of monetary policy and refinancing: Model vs Data

Data Model
No state-
Benchmark dependency
in VAR
Panel A: Fraction of loans that refinanced
∆R(t) 0.040 0.097 0.102
(0.023)
∆R(t) x Average rate gap 0.266*** 0.209 0.136
(0.076)
Panel B: Fraction of loans that are cash-out refi
∆R(t) 0.074*** 0.098 0.104
(0.007)
∆R(t) x Average rate gap 0.176*** 0.211 0.137
(0.027)
Panel C: Fraction of loans for home purchases
∆R(t) 0.095*** 0.114 0.066
(0.006)
∆R(t) x Average rate gap 0.115*** 0.134 0.078
***

Notes: The table reports the response to a decline in interest rates. It therefore reports the estimates
from regression equation (2), multiplied by -1. Standard errors are in parentheses. ⇤ , ⇤⇤ , and ⇤⇤⇤ give
the significance at the 10, 5, and 1 percent levels.

5
 Table 8: State dependency of monetary policy

Effect on refinancing:
Overall effect of a 50 bp expansionary shock 5.23%
β1ΔRt 4.83%
β2ΔRt times mean(ϕt) 0.40%
Effect on consumption:
Overall effect of a 50 bp expansionary shock 1.43%
β1ΔRt 0.97%
β2ΔRt times mean(ϕt) 0.46%

Notes: The table reports the response to a decline in interest rates. It therefore reports the estimates
from regression equation (2), multiplied by -1. See text for more detail.

Table 9: Alternative paths of monetary policy

Average
Fraction with Effect on Change in Fraction ST
Rate path prior to a 50bp cut rate gap
positive rate gap refinancing consumption constrained
before cut

Panel A: Effects of Flat vs Rising History

(i) Flat at about 3.5% 0.00% 100% 18% 1.6% 0.68
(ii) Rising from 3.5% to 6.5% over 4 pds -0.62% 23% 7% 0.2% 0.83
Difference (i)-(ii) 0.62% 77% 11% 1.4% -0.15

Panel B: Effects of Flat vs Falling History

(i) Flat at about 3.5% 0.00% 100% 18% 1.6% 0.68
(ii) Falling from 3.5% to 1% over 4 pds 0.45% 100% 12% 0.6% 0.66
Difference (i)-(ii) -0.45% 0% 6% 1.0% 0.02

Notes: Alternative paths of monetary policy. See text for more detail.

6
 Table 10: Alternative paths of monetary policy

Average Fraction with

Effect on Change in Fraction ST
Rate path prior to a rate cut rate gap positive rate
refinancing consumption constrained
before cut gap
Reloading Effect with 50bp cut
(a) Benchmark case: continuously flat at 0.00% 100% 18% 1.6% 68%
3.5% prior to a 50bp rate cut
(b) 3.5% cut to 1% for 4 pds, rise for 3 pds -0.48% 67% 10% 0.8% 56%
to 3.5%, flat at 3.5% for 1 pd
(c) 3.5% cut to 1% for 4 pds, rise for 3 pds -0.27% 86% 14% 1.1% 66%
to 3.5%, flat at 3.5% for 2 pds
(d) 3.5% cut to 1% for 4 pds, rise for 3 pds -0.11% 88% 16% 1.9% 70%
to 3.5%, flat at 3.5% for 3 pds

Reloading Effect with 100bp cut

(e) Benchmark case: continuously flat at 0.00% 100% 19% 2.7% 72%
3.5% prior to a 100bp rate cut
(f) 3.5% cut to 1% for 4 pds, rise for 3 pds -0.48% 68% 18% 3.1% 56%
to 3.5%, flat at 3.5% for 1 pd

Notes: Alternative paths of monetary policy. See text for more detail.

7
 Figure 1: Time Series of the Refinancing Rate
(a) Refinancing Rate (b) Change in Refinancing Rate

(c) Share of Refinancing With Cash-Out (d) Change in Share of Refi With Cash-Out

Notes: Panels (a) and (b) depicts the time series of the refinancing rate (fraction of loans that were
refinanced in that quarter) and the quarterly change in the refinancing rate, respectively. Panels (c)
and (d) depicts the time series of the share of refinanced loans that involved a balance increase (i.e.
cash-out) and the quarterly change in this share, respectively.

8
 Figure 2: Distribution of interest rate gaps in 1997q4 and 2000q4

Notes: The figure depicts the distribution of interest-rate gaps across borrowers. The interest-rate gap
is defined as the di↵erence between the existing mortgage rate and the current market rate. See text
for more details.

Figure 3: Cumulative Refinancing Rate in Response to a 50bp Mortgage Rate Cut

(a) Rate gap = 0 (b) Rate gap = 50bps

Notes: The figure depicts the cumulative response function of refinancing to a 50 bp decline in mortgage
rates. Panel A shows the response if the initial average rate gap is zero. Panel B shows the response
if the initial average rate gap is instead 50 basis points.

9
 Figure 4: E↵ect on the Refinancing Rate of a 50bp cut, at each point in time

Figure 5: Time series of fitted and actual mortgage rates

30-year Mortgage Rate

8
Actual Fitted
7
6
5
4
3
2
1
0
1989 1993 1997 2001 2005

Notes: The figure depicts the fitted and actual mortgage rate data. See text for more details.

10
 Figure 6: Time series of fitted and actual house price to rent ratios

Log House Price to Rent Ratio

4.7
Actual

4.6 Fitted

4.5

4.4

4.3

4.2

4.1
1989 1993 1997 2001 2005

Notes: The figure depicts the fitted and actual house price to rental ratios. See text for more details.

Figure 7: Impulse response function of aggregate variables

Notes: The figure depicts the impulse response function to a 1 sd interest rate shock. See text for
details.

11
 Figure 8: Life-cycle moments

Notes: The figure depicts the fitted and actual life-cycle moments. See text for more details.

Figure 9: Refinancing, given the interest-rate gap

0.6

Model Data
0.5

0.4

0.3

0.2

0.1

0
0-0.5% 0.5-1% 1-2% >2%
Rate gap

Notes: The figure depicts propensity to refinance for each given interest-rate gap in the data and the
model. See text for more details.

12
 Figure 10: Time Series of Refinancing

Fraction of loans that refinanced

0.35

0.30 Data

Model
0.25

0.20

0.15

0.10

0.05

0.00
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Notes: The figure depicts propensity to refinance over time in the data and the model. See text for
more details.

Figure 11: Alternative interest rate paths

Notes: The figure depicts three alternative interest rate paths, starting at steady state. See text for
more details.

13
 Figure 12: Alternative interest rate paths
4.0%

3.5%

3.0%

2.5%

2.0%

1.5%

1.0%
Periods after
0.5%
normalization
0.0%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Notes: The figure depicts three alternative interest rate paths, starting at steady state. See text for
more details.

14