Professional Documents
Culture Documents
Abstract. How important are collateral constraints for reproducing salient features of
the data? To address this question, we estimate two nested versions of a New Keynesian
model: one with collateralized household debt and the frictionless version of the same
model. Both versions of the model are fit to Canadian data using Bayesian methods. We
argue that the presence of collateral constraints improves the performance of the model in
terms of overall goodness of fit. Housing collateral helps to generate a positive correlation
between consumption and house prices. Moreover, housing collateral induced spillovers
boosted consumption growth during the housing market boom-bust cycles of the late
1980s and early 2000s.
We are grateful to Matteo Iacoviello, Fabio Canova and Giorgio Primiceri for useful feedback
on this project. We also wish to thank Allan Crawford, Stephen Murchison, Hajime Tomura
and workshop participants at the Bank of Canada, the 2007 Midwest Macro Meetings, the 2007
Computation in Economics and Finance Meeting and the 2007 North American Summer
Meeting of the Econometric Society for comments and suggestions. David Mandelzys, Linxin
Zhang and Alexander James provided excellent research assistance. Any remaining errors are
ours. This paper was started when Caterina Mendicino and Shin-Ichi Nishiyama were senior
analysts at the Bank of Canada.
Corresponding author: Caterina Mendicino, caterina.mendicino@gmail.com
1. Introduction
Since Kiyotaki and Moore’s (1997) theoretical work, collateralized debt has be-
come a popular feature of macroeconomic models. Recent papers document the
role of collateral constraints for business cycle fluctuations. In estimated mod-
els of the US business cycle, collateral effects help the model to reproduce the
positive response of spending to a house price shock (see Iacoviello 2005) and
generate sizeable housing market spillovers (Iacoviello and Neri 2010). Looser
credit constraints at the household level contributed to the decline in macroe-
conomic volatility experienced in the US in the 1980s and 1990s (Campbell and
Hercowitz 2004). The collateral mechanism has also important implications for
the international transmission of shocks (Iacoviello and Minetti 2007). Collateral
constraints amplify and propagate macroeconomic fluctuations also through the
joint dynamics of land prices and business investment (Liu et al. 2013).
This paper revisits the role of collateral constraints by quantifying their aggre-
gate implications in a model of the Canadian business cycle. We argue that the
collateral mechanism is not only qualitatively but also quantitatively important
for consumption and house price dynamics. We develop this argument by com-
paring the quantitative performance of two nested models: a model with collater-
alized household debt and the frictionless version of the same model. We pursue
a multivariate approach and estimate the entire structure of the model using the
full-information likelihood-based approach. The full-information approach op-
timally adjusts the estimation of the parameters of the models to match the data.
Thus, we give a fair chance to both versions of the model to perform equally well in
matching the data.1 Following the DSGE literature, we estimate the two versions
of the model using Bayesian methods.2 We conduct Bayesian inference and use
posterior probabilities to assess the adequacy of the two alternative modelling
frameworks. We complement previous findings by quantitatively assessing the
role of collateral constraints through model comparison rather than just through
sensitivity analysis as is common in the literature. Further, we also contribute to
the large literature on estimated DSGE models of the Canadian economy.3 In
1 Rather than limiting the analysis to the estimation of single equations derived by the two
alternative specifications of the model, we prefer to compare the models on the bases of a
full-information approach. Model comparisons based on the estimations of single equations,
such as Euler equations or housing demand equations, would test the performance of the model
in only one particular dimension.
2 See, among others, Smets and Wouters (2003, 2007), Iacoviello and Neri (2010), Justiniano et al.
(2010) and Covas and Zhang (2010).
3 Closed-economy DSGE models of the Canadian economy have been estimated by, among
others, Dib (2003, 2006), Dib, Gammoudi et al. (2008), Atta-Mensah and Dib, (2008) and
Covas and Zhang (2010). For estimated small open economy models of the Canadian economy,
see Ambler et al. (2004) and Justiniano and Preston (2010a, 2010b) for standard one-sector
models; see Ortega and Nooman (2006) for a two-sector small open economy model model; see
Dib (2008) for a multi-sector model and Dib, Mendicino et al. (2008) for a multi-sector model
with credit frictions.
Housing collateral and the business cycle 209
fact, this paper represents the first attempt to quantify the role of credit frictions
at the household level over the business cycle in Canada.4
The analysis proceeds in three steps. First, we introduce the collateral con-
straint mechanism at the household sector into a New Keynesian small open
economy framework. In order to match a broad set of features observed in the
data, the model economy includes a set of nominal and real rigidities that have
been shown useful to match the data (Christiano et al. 2005; Smets and Wouters
2007). At the core of the model is the borrowers–lenders setup developed by
Kiyotaki and Moore (1997). The model features two types of households that
differ in terms of the rate at which they discount the future. In equilibrium, one
type of households borrows, whereas the other lends. Credit constraints arise
because lenders cannot force borrowers to repay. Thus, houses are used as loan
collateral. Houses are elastically supplied. Foreign savers also supply funds to
the domestic economy. Second, to quantify the role of the collateral constraint
mechanism, we fit the model to Canadian data. In particular, we estimate two
versions of the model, one nested in the other. In the benchmark model, housing
collateral value can have an impact on aggregate consumption through collateral
constraints and in the alternative specification this channel is not allowed to op-
erate. Third, we assess the importance of the model’s collateral effect in its ability
to capture key features of consumption and house price data at business cycle
frequencies.
This paper provides several insightful results. We find statistical evidence
suggesting that collateral links between the housing market and the rest of the
economy are quantitatively important. The model with collateral constraints out-
performs the model without them in terms of overall goodness of fit. Indeed,
collateralized household debt helps to generate the positive correlation between
consumption and house prices observed in the data. As the recent experience
in many OECD countries has shown, house prices can fluctuate considerably
over time, making it important to understand how changing house prices influ-
ence consumption behaviour.5 In our general equilibrium framework, there is a
variety of aggregate disturbances, including shocks related to productivity, which
can drive both consumption and house prices.6 Our results highlight that there
are important differences across models in the response of consumption to a par-
ticular shock. In particular, significant differences are found in the transmission
4 For models of the Canadian economy with credit frictions at the firm level, see Dib, Mendicino
et al. (2008), Covas and Zhang (2010) and Zhang (2011).
5 See Hubrich et al. (2013) for evidence on macro-financial linkages in OECD countries.
6 Consumption expenditures and house prices co-move over the business cycle. This positive
correlation can be found in macroeconomic time-series estimates for a variety of countries (Case
et al. 2005) including Canada (Pichette and Tremblay 2003). Micro data studies also show the
important link between consumption and house prices. Campbell and Cocco (2007) find that
consumption expenditures and house prices are more strongly related among household
groupings that are likely to face financial constraints. Attanasio et al. (2005) argue that the
co-movement of consumption and house prices represents their responses to some other
common factor, such as a productivity shock.
210 I. Christensen, C. Mendicino and S.-I. Nishiyama
2. Model
2.1. Households
Households supply labour and derive utility from consumption, ci,t , and housing
services, hi,t :
∞
t L,t ´
maxE0 ¯i b,t ln(ci,t − bCi,t−1 ) + jt lnhi,t − Li,t , (1)
t=0 ´
where i = {1, 2} denotes the two types of households, bCi,t−1 represents external
habits in aggregate consumption, b,t is a shock to the discount rate that affects
the intertemporal substitution of households, ji,t is a shock to the preference for
housing services and L,t is a shock to the disutility of labour. As common in the
literature, housing services are assumed to be proportional to the stock of houses
held by the household.7
Lenders. Patient households accumulate properties for housing purposes, h1,t ,
supply loans to impatient households, b1,t , buy foreign bonds, bÅt , invest in phys-
ical capital, kt , that they rent to the final-good-producing firms at the rate Rtk
and receive dividends, Ft . Thus, they maximize their expected utility subject to
the following budget constraint:
Rt−1 b1,t−1
c1,t + qh,t (h1,t − (1 − ±h )h1,t−1 ) + qk,t (kt − (1 − ±h )kt−1 ) + − b1,t
¼t
Å
Rt−1 &t−1 bÅt−1
+ · · · + st − bÅt = w1,t L1,t + Rtk kt−1 + Ft + Tt , (2)
¼tÅ
where ¼t = Pt =Pt−1 is the gross inflation rate, qh,t is the price of housing, qk,t is
the price of capital, w1,t are real wages and st the real exchange rate. The stock of
housing and capital depreciate at a rates ±h and ±h , respectively. All the variables,
except for the gross nominal interest rates on domestic and foreign bonds, Rt , and
RtÅ , are expressed in real terms. The return on foreign debt depends on a country
specific risk premium & that is required for the model to feature a stationary
distribution.8 Following Adolfson et al. (2007) we assume that the risk premium,
&t , depends on the ratio of net foreign debt to domestic output and the expected
exchange rate:
st bÅt Et st+1 st
&t = exp Á + Ás − 1 + s,t . (3)
Ptd Yt st st−1
The inclusion of the expected exchange rate in the specification of the risk pre-
mium is motivated by empirical findings of a strong negative correlation between
the risk premium and the expected depreciation.9 The demand for foreign funds
combined with the demand function for domestic loanable funds, implies an
uncovered interest parity condition, which in log-linearized form obeys:
r̂t − r̂Åt = (1 + Ás )Et 1st+1 + Ás 1st + Á, (4)
where rt = Rt − Et ¼t+1 .
7 See, for example Iacoviello (2005), Iacoviello and Neri (2010) and Liu et al. (2013).
8 See Schmitt-Grohe and Uribe (2003) for further details.
9 See Duarte and Stockman (2005).
212 I. Christensen, C. Mendicino and S.-I. Nishiyama
10 Consider the
Euler equation of the impatient household evaluated at the deterministic steady
state ¹2 = 1 − ¯¯2¼ Uc2 > 0, where ¹2 is the Lagrange multiplier associated to the borrowing
1
constraint.
11 The primary motivation for this assumption is to obtain a closed-form solution for the steady
state of the model.
Housing collateral and the business cycle 213
2.3.2. Retailers
Finally, retailers combine domestic and imported goods, Y d and Y m, into a
final good Y. Retailers operate in a perfectly competitive market using a CES
production function:
214 I. Christensen, C. Mendicino and S.-I. Nishiyama
Á
1
Á−1
Á 1 Á−1 Á−1
Yt = 1−! Á Ytd +! Á m
Yt Á , (12)
where ! > 0 is the weight on imported goods in the final domestic goods basket and
Á > 0 is the elasticity of substitution between domestic and imported intermediate
goods. Cost minimization entails the following demand curves for Y d and Y m :
−Á m −Á
Ptd P
d
Yt = 1 − ! Yt , Yt = ! t
m
Yt + m,t , (13)
Pt Pt
12 Here we follow, among others, Bernanke et al. (1999), Christiano et al. (2003) and Christensen
and Dib (2008).
Housing collateral and the business cycle 215
2
1 Ãh Ih,t
max qth Ih,t − Ih,t + − ±h ht−1 , (16)
Ih, t Ah,t 2±h ht−1
where Ah,t is housing-specific technology shock, i.e., a shock to the marginal
efficiency of producing housing. New housing capital goods are sold at price qth :
h 1 Ãh Ih,t
qt = 1+ − ±h . (17)
Ah,t 2±h ht−1
This equation is similar to the Tobin’s q relationship for investment, (eq 13),
in which the marginal cost of a unit of housing is related the marginal cost of
adjusting the housing stock. Note that a positive shock to the housing specific
technology will reduce the price of installed housing.
As in Iacoviello and Neri (2010) and Davis and Heathcote (2005), we define GDP
at constant prices and set the relative prices qh, qk, s equal to their steady-state
values.
the posterior distribution can be written as the product of the likelihood function
of the data given the parameters, L(0t |3), and the prior, '(3):
P(3|0t ) L(0t |3)'(3) (27)
We start by estimating the mode of the posterior distribution by maximizing the
log posterior function. Second, we obtain a random draw of size 500,000 from the
posterior distribution using the random-walk Metropolis–Hastings algorithm.
The posterior distribution of the parameters are then used to draw statistical
inference on the parameters themselves or functions of the parameters, such as
second moments.
13 This series is highly correlated with an alternative index of resale housing prices from Royal
Lepage that measures the prices of houses with similar characteristics in different regions across
the country. For this reason we do not think that composition bias is important enough to affect
our results.
14 Unlike the US consumer price index, the Canadian CPI data do not include the cost of imputed
rents. Nonetheless, house prices do affect our measure of inflation through owned
accommodation prices, which include such costs as home insurance, house depreciation and
property taxes.
218 I. Christensen, C. Mendicino and S.-I. Nishiyama
0.05 0 0
0 –0.2 –0.2
Imports
Exports Hours
0.4 0.4 0.1
0.2 0.2 0.05
0 0 0
–0.2 –0.2 –0.05
–0.4 –0.4 –0.1
1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010
15 De Graeve et al. (2007) is another example of a medium-scale DSGE model estimated with yield
curve data (for the United States).
16 A detailed description of the data sources and plots of the detrended data are presented in the
technical appendix, available in the online version of this article.
Housing collateral and the business cycle 219
17 Consistent with the model, to calculate these ratios, we measure gross domestic product as the
sum of the consumption, residential investment, business fixed investment (i.e., excluding
inventory accumulation) and net exports. Since the real national accounts aggregates are
produced on a chain Fisher basis, we calculate these ratios using the nominal series.
18 The discount factor of impatient households is in accordance with estimates of discount factors
for poor or young households (e.g., Lawrance 1991) and falls into the empirical distribution for
discount factors estimated by Carroll and Samwick (1997).
19 The value of the loan-to-value ratio m should reflect the typical loan-to-value ratio for a
constrained household. This household, which we think of as being a first-time home buyer,
borrows the maximum possible against his real estate holdings. Ultimately, we would like to
incorporate information from observed financial variables such as mortgage debt into the
model. However, aggregate measures of mortgage debt in Canada include debt held by
unconstrained households that have had time to accumulate other assets and are better thought
of as patient consumers. In future work, we may be able to exploit information from microdata
to determine a more appropriate target debt-to-asset ratio.
220 I. Christensen, C. Mendicino and S.-I. Nishiyama
share in the production of final goods, ° = 0.23, is lower than is typically used in
models that aggregate all types of capital. The depreciation rate for capital along
with the capital share imply a ratio of business fixed investment to GDP of about
0.165, approximately that seen in the data.
We set the elasticity of demand for individual domestic intermediate goods
so as to give an average markup of 5% in steady state. Individual imported inter-
mediate goods have the same elasticity of demand. Following previous studies,
we set the share of imported goods in the final domestic goods basket, !, to 0.3.
4. Empirical results
In the following, we report the estimates for the two nested models: the model
with collateral constraints and the frictionless version of the same model. We
refer to our benchmark model as the FA model since it has an active financial
accelerator and the model without the borrowing constraints as with the NoFA
model. In this latter framework, the economy is populated only by patient agents,
i.e., ¯2 = ¯1 , that derive utility from housing services, purchase capital and rent it
to the intermediate goods producer. All other features of the model are preserved.
We report the posterior mean and 95% probability interval for the structural
parameters, along with their priors, for the FA model in table 1. Except where
noted, most of the parameters are within the ranges specified in the priors.
The habit-persistence parameter b is estimated to be 0.79, while the long-run
elasticity of labour demand ´ is 0.55; the estimated habits are on the high end of
our prior, and the labour elasticity is on the lower end. The estimate of the wage
share of unconstrained households, ®, is about 0.62. Iacoviello and Neri (2010)
estimates for the US economy report a wage share of unconstrained households
of about 0.68 and 0.81, respectively, over the 1965–1982 and 1982–2006 periods.
The mean estimate of Calvo price stickiness for domestic goods μ d implies
an average duration of price stickiness of about six quarters. Such estimates for
average price stickiness are high relative to findings on price adjustment from
micro studies, but they are in line with previous DSGE models’ estimates–even
with more elaborate systems of nominal stickiness (e.g., Smets and Wouters 2007;
Adolfson et al. 2007). Wage stickiness on the other hand is much lower than in
many other studies, with a typical wage being unchanged for only four months on
average. The price stickiness for imports μ m is about the same as for domestically
produced goods. However, price stickiness for exports μ x is much lower, with the
average export price being sticky for about only 1.25 quarters on average. The
long run elasticity of import demand ¾ is much lower (about 0.3) than that of
export demand ¾ f (about 1.5). Such estimates reflect a much faster response of
exports than imports to a change in the real exchange rate. This may reflect the
composition of exports and imports.
The estimated range of ½¼−1 implies a range for the inflation weight in the
monetary policy rule, ½¼ , from 1.4 to 2.1, while the weight on output, ½y , is fairly
small (between 0.01 and 0.06). The interest rate smoothing term, ½R , is between
0.41 and 0.68. The estimated policy rule coefficients are similar to those reported
in other studies (Ortega and Rebei 2006, Dib 2008, Dib, Gammoudi et al. 2008).
Table 1 also reports the results for the NoFA model. The estimates of the two
versions of the model do not highlight significant differences across parameters’
222 I. Christensen, C. Mendicino and S.-I. Nishiyama
TABLE 1
Estimation results
Our objective is to document the role that collateral constraints play to enable
the model to fit the data over the business cycle. Initially, we compare the two
estimated versions of the model in terms of the overall goodness of fit over the
entire sample. Then, we investigate how collateral constraints affect the trans-
mission mechanism of a variety of shocks. Last, we illustrate the implications
for a key set of moments of interest in the data. To this purpose we compare the
two models with the performance of a Bayesian Vector Autoregression (BVAR)
estimated using the same data set.
Results presented below show that the data strongly favour the model with col-
lateral constraints. Collateral constraints mainly affect the transmission of shocks
that are directly linked to house prices. In particular, they enable the model to gen-
erate a positive response of aggregate consumption to housing demand shocks.
As a result, the model with collateral constraints performs better in matching
some key facts about consumption, particularly its correlation with GDP and
house prices.
for the number of estimated parameters, which should reduce the advantage of
the FA model. We also calculate the log ratio of the Laplace approximations of
the marginal likelihood of both models, assuming flat priors on all parameters,
including standard deviation terms.
The log Bayes factor as measured by the Schwarz criterion is 22.4; the implied
posterior odds ratio of FA and NoFA is approximately 5.5 × 109 to 1 in favour of
FA. The log of the Laplace odds ratio is about 20.9, implying a posterior odds ra-
tio of 1.2 × 109 to 1 in favour of FA. The Laplace approximation and the Schwarz
approximation of the Bayes factor are within the same order of magnitude, and
both suggest that the data strongly favour the FA model over the NoFA model.21
Interestingly, the maximum likelihood estimate of alpha is about 0.63, similar to
the posterior mean estimated with the use an informative prior.
relaxes the borrowing constraint and improves the debt capacity of borrowers.
Thus, expansion of households’ credit allows them to also increase consumption
expenditures.
To illustrate the importance of different features of the model for the trans-
mission of housing preference shocks, figure 3 plots impulse responses from the
estimated FA model along with the responses obtained when: (i) the borrow-
ers make up a negligible fraction of the wage bill, i.e., ® = 1 (dashed line), (ii)
wages are flexible, i.e., μw = 0 (dotted line) and (iii) access to the international
bond market is more costly and Á rises from 0.0006 to 0.1 (dash-dotted line).
Regarding the presence of borrowing constrained households, we find that this
is a crucial assumption to obtain a positive response of consumption to housing
preference shocks (top left panel). Further, it also implies a more pronounced
rise in residential investment (top right panel).
The degree of access to the foreign bond market is another important factor
for the response of aggregate consumption to the shock. In fact, reduced access
to the foreign bond market (higher Á) increases the response of the the yield
spread and, thus, the real interest rate to this shock (not reported in the figure)
such that in equilibrium it is possible to satisfy the demand for loans through
a larger amount of domestic savings.23 Moreover, the reduced amount of loans
also dampens the increase in consumption expenditures and generates a larger
response in hours worked.
Finally, under flexible wages the increase in hours worked is reduced while the
interest rate rises by more. This occurs since, in the absence of wage stickiness,
the impact of the housing demand shock on the marginal cost and inflation is
exacerbated (not shown in the figure).
23 Recall that the yield spread rises (above its steady-state value) when the real interest rate rises
above its steady-state value.
24 The variances from the BVAR were calculated from a BVAR with four lags estimated with a
Normal-Wishart prior (Kadiyala and Karlsson 1997), with the prior variance hyperparameter
set at 0.012, a lag decay proportional to 1=k 2 , and 13 degrees of freedom (the minimum
permissible) on the Inverse Wishart prior on the error covariance matrix. Further, only draws
from the posterior that resulted in a stationary process (permitting a finite unconditional
variance) were used, further restricting the BVAR.
25 In an early example, Schorfheide (2000) evaluates the ability of two DSGE models to match the
correlation between inflation and output produced by a VAR.
Housing collateral and the business cycle 227
the properties of the forecast errors of each model at different horizons up to two
years.
Standard deviations. In table 2, we report the standard deviations of selected
variables.26 These are conditional statistics that show the standard deviation of
the forecast of variable X at a four-quarter horizon relative to the standard de-
viation of GDP. We focus on the four-quarter horizon as an illustration, but the
results are broadly similar for horizons up to eight quarters.
Overall, the standard deviations are quite similar for the model with collateral
constraints (FA) and the model estimated with those effects constrained to zero
(NoFA). Both models are able to generate a high degree of volatility in residential
investment and house prices, though still less than in the VAR. In addition, both
models under predict the relative volatility of imports. In contrast, in both models
the volatility of consumption is higher than what generated by the VAR, though
consumption is somewhat less volatile in the model with the collateral effects.
The relative volatilities of inflation, the real exchange rate and the yield spread
are all close to their values in the data.
Cross-correlations. Table 3 reports the cross-correlations for the same selected
variables. We compute the correlation in the forecast errors of X and Y at a
26 From an aggregate perspective there are a number of reasons to think that house prices could
influence consumption decisions in Canada. First, residential structures and land account for a
large share of Canadian household sector wealth. A total of 68% of Canadian households own a
home and for many it represents their largest asset. Second, house price growth is associated
with higher household borrowing. The positive correlation between consumption and house
prices may be related to housing’s role as collateral. Between 2000 and 2007, the real price of
existing homes increased by 52%. At the same time, the ratio of household debt to GDP rose
dramatically from 58% in 2000 to 76% in 2007. By 2007 roughly 80% of Canadian household
debt was secured by real estate.
228 I. Christensen, C. Mendicino and S.-I. Nishiyama
TABLE 2
Four-step ahead standard deviations, as ratio of stan-
dard deviation of GDP
VAR model FA model NoFA model
C 0.823 0.766 0.935
L 0.809 0.865 0.869
w 0.894 0.799 0.780
Ik 3.227 3.481 3.453
Ih 4.617 3.229 3.285
qh 2.894 2.322 2.346
¼ 0.213 0.248 0.252
Yx 2.475 2.379 2.355
Ym 2.624 2.023 1.874
s 2.830 2.509 2.425
Yield spr. 0.135 0.159 0.154
TABLE 3
Four-step-ahead correlations
VAR model FA model NoFA model
Correlation with GDP
C 0.672 0.643 0.598
L 0.452 0.319 0.365
Ik 0.408 0.544 0.508
Ih 0.715 0.284 0.207
¼ −0.113 −0.136 −0.140
Correlation with house prices
C 0.475 0.462 0.116
Ik 0.231 0.110 0.216
Ih 0.567 0.476 0.488
four-quarter horizon, rather than the infinite horizon correlation. The results are
qualitatively similar at the one-step and eight-step forecast horizons.
In the model, consumption and both types of investment are positively corre-
lated with GDP, as they are in the VAR. The model-implied correlations between
GDP and consumption are quite close to the correlation in the data and the cor-
relation of fixed capital investment implied by the models is slightly stronger.
Though the model with collateral effects generates a higher correlation between
residential investment and GDP it is still well below the correlation in the data.
Turning to cross correlations with real house prices, we see that the model with
collateral effects generates a correlation between consumption and real house
prices that is very close to the VAR. The collateral effects appear to be impor-
tant for this finding, since the model without those effects generates only a very
weak correlation. Both models produce a correlation between house prices and
residential investment that is close to that seen in the VAR.
Housing collateral and the business cycle 229
Overall, the model with collateral effects matches some key facts about con-
sumption, particularly its correlation with GDP and house prices. In addition, it
produces a plausible correlation between housing investment and house prices,
despite a highly stylized housing production sector. Though the model is cap-
turing the procyclical nature of residential investment, the correlation is weak
relative to the data.
6. FA model
TABLE 4
Variance decomposition – FA
Hous. Hous. Disc. Techn. Inv. Exch. Exp. Other
sup. pref. fact. spec. rate dem.
House inv. Ah j b z Ak s x shocks
1 64.00 30.47 0.06 1.81 0.14 1.64 0.55 1.33
4 62.25 32.25 0.11 2.02 0.13 1.67 0.57 1.01
8 59.39 35.45 0.09 2.06 0.09 1.47 0.61 0.85
Uncond. 31.17 46.55 0.05 4.96 0.05 0.98 14.86 1.38
Inv. spec. Exch. rate Exp. dem. Other
House prices Ak s x shocks
1 2.49 82.54 0.17 4.90 0.39 4.43 1.50 3.57
4 2.73 83.07 0.29 5.19 0.35 4.31 1.47 2.60
8 3.03 84.59 0.22 4.93 0.21 3.52 1.45 2.06
Uncond. 8.72 68.23 0.08 6.26 0.13 2.12 12.53 1.93
Infl. obj. Exch. rate Markup Other
Inflation IO s p shocks
1 0.02 1.26 0.75 9.62 44.58 7.48 29.80 6.49
4 0.01 1.24 0.61 8.94 61.64 6.38 15.63 5.55
8 0.02 0.88 0.41 7.60 72.28 4.80 9.75 4.27
Unconditional 0.01 0.12 0.06 3.95 92.28 0.99 1.26 1.32
Infl. obj. Exch. rate Exp. dem. Other
Consumption IO s x shocks
1 0.93 5.11 70.00 6.43 1.25 4.87 5.79 3.50
4 1.83 5.93 32.25 17.49 1.41 11.78 16.51 8.26
8 1.86 3.17 15.50 25.54 0.90 13.89 25.33 7.87
Uncond. 0.38 1.22 1.25 21.33 0.08 3.71 66.71 0.89
Invest. spec. Imp. dem. Exp. dem. Other
GDP Ak M x shocks
1 1.63 4.02 9.72 16.93 8.56 6.73 12.43 11.17
4 0.66 3.63 4.95 39.00 10.32 3.10 6.27 11.68
8 0.36 2.43 2.53 55.14 7.25 1.79 5.50 9.04
Uncond. 0.08 0.71 0.53 81.74 1.63 1.02 7.57 2.64
GDP variation at eight quarters.27 Overall, demand shocks in general play a more
important role at short horizons, particularly import demand, m , and discount
factor shocks, b . At long horizons the neutral productivity shock accounts for
the bulk of the fluctuations in GDP.
Meanwhile, housing demand shocks explain less than 1% of GDP and con-
sumption in the NoFA model (table 5). The drivers of house prices and residential
investment in NoFA are almost identical to the results reported for FA: hous-
ing demand accounts for above 80% of short horizon variance of house prices
and housing demand and investment-efficiency shocks account for over 90% of
the variance of residential investment. Much of the variance of consumption at-
tributed to the housing demand shock under FA is attributed to the discount
factor shock in the model where collateral effects are constrained to be zero. The
discount factor shocks explain 50% of consumption variance at the four-quarter
horizon under NoFA but only 30% under FA. The greater importance of the
housing demand shock in the FA model and lesser importance of the consump-
27 Studies of the US have also found relatively small impacts of housing demand shocks on
aggregate output (see Jarocinski and Smets 2008 and Iacoviello and Neri 2010).
Housing collateral and the business cycle 231
TABLE 5
Variance decomposition – NOFA
Hous. Hous. Disc. Techn. Inv. Exch. Exp. Other
sup. pref fact spec. rate dem.
Hous. inv. Ah j b z Ak s x shocks
1 62.26 31.91 0.32 1.69 0.05 2.32 0.28 1.18
4 60.75 33.34 0.42 1.86 0.04 2.46 0.31 0.81
8 59.15 35.14 0.36 1.94 0.03 2.42 0.39 0.57
Uncond. 42.07 36.47 0.22 3.98 0.05 3.47 13.06 0.7
Inv. spec. Exch. rate Exp. dem. Other
House prices Ak s x shocks
1 2.43 82.49 0.82 4.36 0.14 6.01 0.73 3.03
4 2.77 82.58 1.04 4.61 0.11 6.10 0.77 2.03
8 3.22 83.21 0.85 4.59 0.07 5.74 0.92 1.40
Uncond. 12.11 61.45 0.41 6.21 0.11 6.77 11.90 1.05
Infl. obj. Exch. rate Markup Other
Inflation IO s p shocks
1 0.06 0.02 3.96 8.70 49.13 6.38 27.90 3.82
4 0.06 0.02 3.73 7.77 64.74 5.14 15.61 2.93
8 0.05 0.01 2.67 6.59 75.05 3.86 9.66 2.09
Uncond. 0.02 0.01 0.47 3.78 91.42 1.73 1.63 0.94
Infl. obj. Exch. rate Export dem. Other
Consumptions IO s x shocks
1 0.41 0.31 72.09 8.98 0.52 10.39 2.74 4.55
4 0.76 0.58 50.81 17.19 0.23 19.03 5.38 6.01
8 1.08 0.87 32.68 24.94 0.13 25.71 9.10 5.49
Uncond. 0.50 0.55 4.31 17.84 0.02 15.99 57.88 2.90
Inv. spec Imp. dem. Exp. dem. Other
GDP Ak m x shocks
1 1.83 0.91 10.88 16.43 7.02 38.70 16.68 7.55
4 0.91 0.47 9.92 40.07 8.32 20.98 9.69 9.63
8 0.56 0.31 5.59 57.47 6.19 13.28 8.21 8.39
Uncond. 0.17 0.12 1.57 79.98 1.85 3.78 8.22 4.32
tion preference shock, suggests that the borrowing constraints are helping the
model to capture some of the co-movement between consumption and house
prices.
First, much of the growth in consumption in the early part of the housing boom
of the late 1980s is attributed to collateral effects. This positive effect peaked in late
1986 but continued to have a positive effect on consumption growth until 1989.
Further, after 1989 the collateral effects begin to negatively affect consumption
growth and have a pronounced negative impact on consumption in 1991. The
sharp collateral effect on consumption also occurs in 1982, suggesting that it an
important role in accounting for consumption dynamics in recessions.
Finally, we consider the post-2000 period, during which many OECD coun-
tries, including Canada, have seen sharp increases in house prices and consump-
tion. Our model suggests that collateral effects contributed as much as 1% to
yearly consumption growth in 2000 and had a positive effect for most of the
remainder of the sample. This contribution is less than in the period from 1986
to 1990. One reason might be that the house price increases since 2000 have been
more gradual than in the late 1980s when they achieved a comparable peak in
half the time, three years instead of six. See figure 1.
Explaining the differences between the 1980s housing boom and the recent
one is thus partly due the fact that the housing demand shocks were more
important in the earlier period. The main differences among the two booms
are related to the behaviour of residential investment and external factors. In
fact, unlike the real house price, residential investment does not continue to rise
after 2005 (though well above trend). To explain the fact that prices continue to
rise but housing construction does not, the model uses a decline in the housing-
investment efficiency shock. Thus, residential investment becomes more costly
and the housing stock more difficult to adjust. In the housing sector, this might
be the result of bottlenecks in production, including the difficulty in finding and
0.04
0.02
0.01
–0.01
–0.02
–0.03
–0.04
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
2007
keeping skilled trades and reduced availability of land with access to public util-
ities and desired amenities. Of course this is speculative and worthy of attention
in future research.
External developments are also a major difference between the late 1980s and
the period after 2000. A decomposition of consumption over this period shows
a rising contribution from the country-risk premium. This shock turns out to
be very important for capturing the pronounced rise in the real exchange rate
vis-à-vis the United States. A key factor driving the exchange rate movement in
the data is a large improvement in Canada’s terms of trade. The result in the
model is a sharply increasing net foreign asset position. The long-lived effects
on wealth lead households to consume more and work less. One key distinction
between a positive housing demand and a negative risk premium shock is that
the latter implies a reduction in hours worked. Together these two shocks are able
to generate rising consumption and house prices and relatively little movement
in hours worked consistent with the data from 2005 onward.
Summarizing, housing collateral-induced spillovers accounted for a large share
of consumption growth during the housing boom of the late 1980s and the sharp
declines in consumption growth in the early 1990s. They also somewhat con-
tributed to boost consumption in the early part of the 2000s housing boom.
7. Conclusion
References
Adolfson, M., S. Laséen, J. Lindé, and M. Villani (2007) “Evaluating an estimated New
Keynesian small open economy model,” Sveriges Riksbank working paper series, no.
203
Ambler, S., A. Dib, and N. Rebei (2004) “Optimal Taylor rules in an estimated model of
a small open economy,” Bank of Canada working paper no. 2004-36
Amirault, D., C. Kwan, and G. Wilkinson (2006) “Survey of price-setting behaviour of
Canadian companies,” Bank of Canada working paper no. 2006-35
Aoki, K., J. Proudman, and G. Vlieghe (2002) “Houses as collateral: Has the link
between house prices and consumption in the UK changed?,” Economic Policy
Review May, 163–77
Armour, J., J. Atta-Mensah, W. Engert and S. Hendry (1996) “A distant-early-warning
model of inflation based on M1 disequilibra,” Bank of Canada working paper no.
1996-5
Atta-Mensah, J., and A. Dib (2008) “Bank lending, credit shocks, and the transmission
of Canadian monetary policy,” International Review of Economics & Finance 17,
159–76
Attanasio, O., L. Blow, R. Hamilton, and A. Leicester (2005) “Booms and busts:
Consumption, house prices and expectations,” Institute for Fiscal Studies working
paper no. W05/24
Benito, A. and H. Mumtaz (2006) “Consumption excess sensitivity, liquidity constraints
and the collateral role of housing,” Bank of England working paper no. 306
Bernanke, B., M. Gertler, and S. Gilchrist (1999) “The financial accelerator in a
quantitative business cycle framework.” In Handbook of Macroeconomics, Volume
1B, J. Taylor, and M. Woodford, eds. Amsterdam: Elsevier
Calvo, G. (1983) “Staggered prices in a utility-maximizing framework,” Journal of
Monetary Economics 12, 383–98
Campbell, J. Y., and J. F. Cocco (2007) “How do house prices affect consumption?
Evidence from micro data,” Journal of Monetary Economics 54(3), 591–621
Campbell, J., and Z. Hercowitz (2004) “The role of collateralized household debt in
macroeconomic stabilization,” NBER working paper no. 11330
Campbell, J. Y., and G. Mankiw (1990) “Permanent income, current income, and
consumption,” Journal of Business & Economic Statistics 8, 265–79
Carroll, C., and A. Samwick (1997) “The nature of precautionary wealth,” Journal of
Monetary Economics 40, 41–71
Case, K. E., J. M. Quigley, and R. J. Shiller (2005) “Comparing wealth effects: The
stock market versus the housing market” The B.E. Journal of Macroeconomics 5(1),
Article 1
Christensen, I., and A. Dib (2008) “The financial accelerator in an estimated New
Keynesian model,” Review of Economic Dynamics 11, 155–78
Christiano, L., M. Eichenbaum, and C. Evans (2005) “Nominal rigidities and the
dynamic effects of a shock to monetary policy,” Journal of Political Economy 113,
1–45
Christiano, L., R. Motto, and M. Rostagno (2003) “The Great Depression and the
Friedman-Schwartz hypothesis,” Journal of Money, Credit and Banking 35, 1119–97
Côté, D., J-P. Lam, J. Kuszczak, Y. Liu, and P. St-Amant (2004) “The performance and
robustness of simple monetary policy rules in models of the Canadian economy,”
Canadian Journal of Economics 37, 978–94
Covas, F., and Y. Zhang (2010) “Price-level versus inflation targeting with financial
market imperfections,” Canadian Journal of Economics 43(4), 1302–32
Housing collateral and the business cycle 235
Davis, M., and J. Heathcote (2005) “Housing and the business cycle,” International
Economic Review 46, 751–84
De Graeve, F. (2008) “The external finance premium and the macroeconomy: US
post-WWII evidence,” Journal of Economic Dynamics & Control 32, 3415–40
Dib, A. (2003) “An estimated Canadian DSGE model with nominal and real rigidities,”
Canadian Journal of Economics 36, 949–72
(2006) “Nominal rigidities and monetary policy in Canada,” Journal of
Macroeconomics 28, 303–25
(2008) “Welfare effects of commodity price and exchange rate volatilities in a
multi-sector small open economy model,” Bank of Canada working paper no.
2008-8
Dib, A., M. Gammoudi, and K. Moran (2008) “Forecasting Canadian time series with
the New Keynesian model,” Canadian Journal of Economics 41, 138–65
Dib, A., C. Mendicino, and Y. Zhang (2008) “Price-level targeting rules and financial
shocks: The case of Canada,” Economic Modelling 30, 941–53
Duarte, M., and A. Stockman (2005) “Rational speculation and exchange rates,”
Journal of Monetary Economics 52, 3–29
Erceg, C., D. Henderson, and A. Levin (2000) “Optimal monetary policy with staggered
wage and price contracts,” Journal of Monetary Economics 46, 281–313
Fung, B., and M. Yuan (1999) “Measuring the stance of monetary policy.” In Money,
Monetary Policy, and Transmission Mechanisms: Proceedings of a conference held by
the Bank of Canada, November 1999. Ottawa: Bank of Canada
Greenwood, J., Z. Hercowitz, and P. Krusell (2000) “The role of investment-specific
technological change in the business cycle,” European Economic Review 44, 91–
115
Hamilton, J. (1994) Time Series Analysis. Princeton, NJ: Princeton University Press
Hubrich, K., A. D’Agostino, M. Cervena, M. Ciccarelli, P. Guarda, M. Haavio,
P. Jeanfils, C. Mendicino, E. Ortega, M. Valderrama, and M.Valentinyiné (2013)
“Financial shocks and the macroeconomy: Heterogeneity and non-linearities,”
European Central Bank occasional paper series, no. 143
Iacoviello, M. (2005) “House prices, borrowing constraints and monetary policy in the
business cycle,” The American Economic Review 95, 739–64
Iacoviello, M., and R. Minetti (2006) “International business cycles with domestic and
foreign lenders,” Journal of Monetary Economics 53, 2267–82
Iacoviello, M., and S. Neri (2010) “Housing market spillovers: Evidence from an
estimated DSGE model,” American Economic Journal: Macroeconomics 2, 125–64
Jappelli, T. (1990) “Who is credit constrained in the US economy?,” The Quarterly
Journal of Economics 105, 219–34
Jarocinski, M., and F. Smets (2008) “House prices and the stance of monetary policy,”
European Central Bank working paper no. 891
Jeffreys, H. (1961) Theory of Probability. Oxford: Oxford University Press
Jermann, U., and V. Quadrini (2006) “Financial innovations and macroeconomic
volatility,” NBER working paper no. 12308
Justiniano, A., and B. Preston (2010a) “Can structural small open economy models
account for the influence of foreign shocks?,” Journal of International Economics 81,
61–74
(2010b) “Monetary policy and uncertainty in an empirical small open economy
model,” Journal of Applied Econometrics 25, 93–128
Justiniano, A., G. Primiceri, and A. Tambalotti (2010) “Investment shocks and business
cycles,” Journal of Monetary Economics 57, 132–45
Kadiyala, K., and S. Karlsson (1997) “Numerical methods for estimation and inference
in Bayesian VAR models,” Journal of Applied Econometrics 12, 99–132
236 I. Christensen, C. Mendicino and S.-I. Nishiyama
Kass, R., and A. Raftery (1995) “Bayes factors,” Journal of the American Statistical
Association 90, 773–95
Kiyotaki, N., and J. Moore (1997) “Credit cycles,” Journal of Political Economy 105,
211–48
Kostenbauer, K. (2001) “Housing depreciation in the Canadian CPI,” Statistics Canada
Analytical Series (Prices Division), no. 15
Lam, J-P., and G. Tkacz (2004) “Estimating policy-neutral interest rates for Canada
using a dynamic stochastic general equilibrium framework,” Swiss Journal of
Economics and Statistics 140, 89–126
Lawrance, E. (1991) “Poverty and the rate of time preference: Evidence from panel
data,” Journal of Political Economy 99, 54–77
Liu, Z., P. Wang, and T. Zha (2013) “Land–price dynamics and macroeconomic
fluctuations,” Econometrica 81, 1147–84
Ortega, E., and N. Rebei (2006) “The welfare implications of inflation versus price-level
targeting in a two-sector, small open economy,” Bank of Canada working paper no.
2006-12
Pichette, L. and D. Tremblay (2003) “Are wealth effects important for Canada?,” Bank
of Canada staff working paper no. 03-30
Schmitt-Grohe, S., and M. Uribe (2003) “Closing small open economy models,” Journal
of International Economics 61, 163–85
Schorfheide, F. (2000) “Loss function-based evaluation of DSGE models,” Journal of
Applied Econometrics 15(6), 645–70
Schwarz, G. (1978) “Estimating the dimension of a model,” The Annals of Statistics 6,
461–64
Smets, F., and R. Wouters (2003) “An estimated dynamic stochastic general equilibrium
model of the Euro area,” Journal of the European Economic Association 1, 1123–75
(2007) “Shocks and frictions in US business cycles: A Bayesian DSGE approach,”
The American Economic Review 97, 586–606
Zhang, Y. (2011) “Financial factors and labour market fluctuations,” Bank of Canada
working paper no. 2011-12