Do Demographics Affect Monetary Policy Transmission

in Canada?∗

Jeremy Kronick Steve Ambler

C.D. Howe Institute ESG UQAM, C.D. Howe Institute
jkronick@cdhowe.org ambler.steven@uqam.ca

August 2018

Abstract

Using a panel of macroeconomic data for Canada and its ten provinces, we estimate
the dynamic effects of monetary policy shocks from the mid-1980s until the present. We
then relate the change in the impact of these shocks to macroeconomic factors including
demographics, specifically changes in the old-age dependency ratio. We find that the
inflation-targeting regime has had an ambiguous effect on the impact of monetary policy
shocks in Canada. On the other hand, changing demographics have unambiguously
reduced the impact of monetary policy shocks. This can help to explain tepid inflation
since the financial crisis and could eventually undermine the effectiveness of Canada’s
inflation targeting regime.

∗
We thank Rosalie Wyonch, John Crow, Pierre Duguay, Paul Jenkins, David R. Johnson, David Laidler,
David Longworth, Angelo Melino, John D. Murray, Angela Redish, Pierre Siklos, participants at the 2018
Canadian Economic Association annual meeting, and two anonymous referees for comments and suggestions.
Remaining errors are our responsibility.

1
1 Introduction

A large number of countries have adopted inflation targeting (IT) since the early 1990s.
Overall, the experiment with IT has been very successful,1 at least until the onset of the
financial crisis and the Great Recession in 2007-2008. However, since the crisis, inflation has
been systematically below target in many countries in spite of rock-bottom interest rates
and different forms of unconventional monetary policy such as quantitative easing. Canada
has not been spared this phenomenon. Since 2010, inflation in Canada has averaged 1.5
percent, well under the 2 percent target. This paper asks what role demographic changes,
in particular an aging population, have had on the transmission of monetary policy. While
demographics is an issue for both inflation and non-inflation targeting regimes, we focus on
the former as a result of the missing inflation post-crisis that has plagued countries whose
central banks have an explicit inflation mandate.2 Nevertheless, any country with a central
bank that uses a short-term interest rate as its primary monetary policy instrument would
also face the same effects of demographic change on monetary policy transmission.

The economic context of inflation targeting and the transmission of monetary policy has
changed gradually over the last 30 years. Most notably, real and nominal interest rates
have decreased gradually (with a sudden decrease at the onset of the financial crisis and
the Great Recession), and long-run neutral interest rates are now lower than they were.
This has important consequences for central banks, since the probability that policy rates
will be pushed by negative shocks to their effective lower bound has increased. This leaves
less room for reducing policy rates in order to stimulate demand.3

One of the main influences on long-term real interest rates is demographic change.
Population growth has been slowing in developed economies, putting a brake on the rate of
1
See Parkin (2016).
2
See, for example, Friedrich (2014) for more on the missing inflation puzzle.
3
See Ambler (2016) on the challenges facing monetary policy in a low interest rate environment.

2
increase of the size of active populations, and increasing the proportion of older people. As
the proportion of older people increases, the long-term equilibrium real interest rate falls as
demand for capital decreases.4 Furthermore, changing demographics can in principle affect
the impact of monetary policy because changes in interest rates have different effects on
different age cohorts: this means that the population age pyramid matters for the response
to changes in central banks’ policy rates. In turn, this occurs mainly because individuals’
and households’ wealth over the life cycle is not constant.

In Canada, the median age in 1971 was 26.2, and has steadily risen, reaching 40.6 in 2017.5
Additionally, for the first time there are now more Canadians over 65 than under 15
(Robson 2017). Offsetting some of this increase in median age is the fact that life
expectancy has increased, which has led some to delay retirement.6 Regardless, Canada’s
aging population, and the resulting fall in demand for capital, has pushed down the neutral
rate of interest over the past 10 years, and now sits, according to the Bank of Canada,
somewhere between 2.5 and 3.5 percent.

Additionally, the relationship between the level of economic activity and inflation has
weakened gradually over the last 30 years. This has often been referred to as a flattening of
the Phillips curve, and has been shown to be true in most industrialized economies. The
implications of this phenomenon are mixed. On the one hand, if long-run inflation
expectations are tied to the inflation target, imbalances between demand and supply are
less likely to move inflation from target. On the other hand, it can also mean that central
banks, in order to alter inflation rates, must affect aggregate demand and output even
more than before.7 What all this means is that the Bank of Canada is starting from a
4
See Baker, DeLong and Krugman (2005) and Ambler and Alexander (2016) on the theoretical link
between population growth and real interest rates.
5
CANSIM Table 051-0001.
6
Year-by-year graphs of Canada’s age pyramid can be found at https://www.populationpyramid.net/
canada/.
7
Carney (2017) notes that economies have become more open and the nature of international trade has
changed, with global value chains boosting trade in intermediary goods. Carney claims that inflation has
become a much more globalized phenomenon since the financial crisis.

3
lower neutral interest rate with less room to decrease the overnight rate, in an environment
where they may actually need to stimulate more than before in order to generate any
desired change in prices.

The goal of this paper is to examine the link between demographics and the impact of
Canadian monetary policy shocks over time. We use a two-stage estimation procedure
similar to that of Imam (2015). In the first stage, we use a Bayesian time-varying
structural vector autoregression (TVC-BVAR), where we take advantage of a new data set
from Champagne and Sekkel (2017) for a more appropriate identification of Canadian
monetary policy shocks. Indeed, this new shock series helps to alleviate the price puzzle
that arises from standard SVAR setups. We use this TVC-BVAR to estimate the impact of
monetary policy shocks over the 1985:Q2 to 2015:Q4 period. We measure the impact of
monetary policy shocks by both the maximum and the cumulative increase/decrease in
inflation and unemployment in response to a shock of the same size.

In the second stage we exploit a database of macroeconomic variables for a cross section of
Canadian provinces and use dynamic panel estimation to evaluate the link between
demographics and the impact of Canadian monetary policy. By exploiting these provincial
data sets we provide a more accurate assessment of the impact of monetary policy in
Canada, in comparison to Imam (2015) who focuses on a cross section of different countries.

Our primary finding is that demographic change has reduced the impact of monetary
policy shocks. Since monetary policy has been expansionary for the bulk of the period
under analysis, the implication is that lower interest rates have not, all other things being
equal, led to as much increased borrowing and demand, and in turn the impact on inflation
and the real economy has been weaker than it would have been in the absence of
demographic change.8 We find that the interest rate and credit channels are important
8
Juselius and Takats (2015) find that demographics accounts for one-third of the variation in inflation
across a set of countries over the 1955-2010 time period. They find a U-shaped pattern implying that higher
inflation is more likely in populations with more dependents (i.e. more young and old). They also show that

4
transmission mechanisms highlighting the fact that an aging population tends to have
lower debt levels — as the need for significant borrowing decreases — and are therefore less
sensitive to movements in interest rates. Our conclusion is that demographic change is a
plausible partial explanation for why inflation has been unable to reach higher levels
post-crisis, despite considerable monetary policy stimulus.

The rest of the paper is structured as follows. In the following section, we review different
possible channels through which monetary policy operates, not all of which imply the same
sign for the relationship between the old-age dependency ratio and its impact. In the third
section, we summarize our econometric methodology and results. In the fourth section we
do some sensitivity analysis, including discussing the results with different lag timing. The
fifth section concludes.

2 Different Channels for Monetary Policy

Transmission

In this section we look at the three primary transmission channels discussed by Imam
(2015): the interest rate channel, the credit channel, and the wealth channel. One other
potential transmission channel for a small open economy is the exchange rate channel.9 Our
choice to not explicitly investigate this channel is due to two factors. First, the effect of the
channel is ambiguous. Aging populations tend to be characterized by a decline in savings,
but also a decline in investments. Similarly, public savings and investments fall with aging
populations due to declining revenues and higher expenditures. It is this combination of
private and public savings/investments that define the net effect on the current account,
monetary policy has both reinforced this relationship and mitigated it at different points over the course of
their sample period.
9
He also discusses two other “less studied and more difficult to discern” channels including the risk-taking
and expectation channels. Readers are referred to Imam (2015) for more details. See also Mishkin (1996)
for a general discussion on the transmission channels of monetary policy.

5
and therefore on the exchange rate. Since it is not clear ex ante which of the decline in
savings and investment outweighs the other, the effect of the channel is ambiguous.

The second reason is that the exchange rate itself, as we show in the next section, is
directly accounted for in the Bank of Canada’s reaction function in the development of the
monetary policy shock series. Therefore, the exogenous component of the monetary policy
shock, and the effect aging has on its transmission, already incorporates the role of the
exchange rate.

In our discussion of the transmission channels, we concentrate on the impact of interest

rate changes on private consumption spending. Ultimately monetary policy operates by
affecting overall aggregate demand and the output gap, thereby affecting firms’ price
setting behaviour and affecting unemployment via firms’ demand for labour.

2.1 The Interest Rate Channel

The interest rate channel follows from the life cycle hypothesis (Modigliani 1966).
Households accumulate debt as they go through the early stages of adulthood, but as they
age they are able to pay off this debt and become net creditors by the time they hit life’s
later stages. Younger households are therefore more sensitive to changes in interest rates
since they need more credit. In this case, monetary policy has a reduced impact on an
aging population. Wong (2018) uses household-level data from the US and finds that the
consumption spending of younger households is in fact much more sensitive to changes in
interest rates than that of the average household, due primarily to their level of mortgage
debt.10
10
Berja et al. (2018) also find that the consumption spending of households with higher levels of mortgage
debt is also more sensitive to interest rate changes in the US, and Cloyne, Ferreira and Surico (2017) find
the same thing for the UK and the US.

6
2.2 The Credit Channel

The credit channel is related to the interest rate channel. As households age, they have
higher net worth and possess significantly more collateral. It is therefore cheaper for them
to borrow should they need to (this is known as having a lower external finance premium).
The smaller this finance premium the less sensitive households are to changing interest
rates. Young adults have a higher external risk premium. Once again, the implication is
that monetary policy has a reduced impact on an aging population.

2.3 The Wealth Channel

While the interest rate and credit channels work in the same direction (aging leads to a
weaker impact of monetary policy), the wealth channel has the opposite impact. As
societies age, the population tends to own more assets, and have greater net worth. The
more assets one has the more sensitive one is to changes in interest rates. This is
exacerbated by the fact that these assets tend to be fixed-income products and are by their
very nature interest rate sensitive. Therefore, we expect monetary policy to generate larger
real effects through the wealth channel.

This argument ignores the possibility that some households are credit-constrained. Kaplan,
Moll and Violante (2018) build a model of households with infinite horizons who face
endogenous credit constraints. Credit constraints make household consumption much more
sensitive to current disposable income and less sensitive to wealth, weakening the
quantitative importance of the wealth channel.

Figure 1, using data from Canada’s Survey of Financial Security11 , indicates that, broadly
speaking, all channels are in play for Canada. Debt peaks in the 35-44 age group before
11
CANSIM Table 205-0002

7
falling over the rest of one’s life, while assets peak at 55-64 before being drawn down as
people enter retirement.

Figure 1: Lifetime debt and assets

In addition to assessing the net change in the impact of monetary policy shocks in Canada
— and the role an aging population plays — we test the role the interest and credit
channels play. Lack of data unfortunately prevents an evaluation of the wealth channel.

3 Methodology and Results

The methodology used in this paper is based on Imam (2015), which in turn is based on
the seminal paper of Primiceri (2005).12 Imam uses a panel of different countries to exploit
cross-country variations in demographics to identify the impact of demographic changes.
We use a panel of Canadian provinces to do the same for Canada. Canada’s provinces can
12
See also Boivin and Giannoni (2006) and Canova and Gambetti (2009).

8
be viewed as part of a currency union. In this respect our methodology is similar to that of
Georgiadis (2015), who uses a panel of euro area countries to identify the impact of
differing economic structures on the monetary transmission mechanism. In the first stage
of the empirical analysis we estimate the changing impact of monetary policy shocks over
time, and in the second stage this changing impact to demographic change while
controlling for other macroeconomic effects.

3.1 Stage One Methodology

In the first stage, we estimate a time-varying coefficient Bayesian vector autoregression

(TVC-BVAR), which establishes the impact of monetary policy shocks on inflation and
unemployment over time. Time-varying coefficients are necessary in order to allow the
impact of monetary policy shocks on inflation and unemployment to change over time.
Estimating a vector autoregression allows us to estimate the impact of one variable on
another when the direction of causation is unclear.

One major difference in this paper with respect to the methodology of Imam (2015) is the
assumptions we use to identify monetary policy shocks. Imam’s results for Canada show a
significant delay in the drop in prices following a contractionary monetary policy shock.
This so-called “price puzzle” is consistent with other studies (for example Grilli and
Roubini 1996) that estimate monetary policy shocks using a system of three equations for
inflation, the unemployment rate, and the central bank’s policy rate, in that order.13

One way of removing this price puzzle is the narrative approach pioneered by Romer and
Romer (2004). This method (in the US context) uses minutes of Federal Open Market
Committee (FOMC) meetings in order to generate a time series of intended changes to the
13
Variable ordering is a standard way of identifying structural shocks in the literature. Shocks to the
policy rate are assumed to have no contemporaneous effect on variables that precede the policy rate in the
model.

9
Federal Funds Rate (FFR). Using minutes in the US context was necessary for many years
since the FOMC did not always explicitly target the FFR. In Canada, it is easier as we
have always used an explicit target and can thus see the intended rate change simply by
evaluating how the overnight rate target differs from one meeting to the next.

However, the intended change itself is not exogenous. To extract the exogenous portion we
need to estimate the Bank of Canada’s reaction function, and subtract what amounts to an
expected change from the actual change. Champagne and Sekkel’s major departure from
the Romer and Romer reaction function is that they also control for lagged levels and
changes of the FFR and the lagged USD/CAD nominal exchange rate. Accounting for both
these variables has the added benefit of dealing with the fact that aging is a global
phenomenon that might impact monetary policy across the board, especially in advanced
countries like the US that have an outsized impact on a small open economy like Canada.

To control for change in the regime of the Bank of Canada, the authors break their sample
in two: 1974–1991, and 1992–2015. If the market understands this reaction function,
regressing the actual change in the policy rate on these variables leaves a set of residuals,
which represents the true set of exogenous monetary policy shocks. They are spliced
together across the sub-samples. Champagne and Sekkel show that this splicing over the
two sub-samples is crucial for eliminating the price puzzle.

Using the Champagne and Sekkel monetary policy shock series, we can move to the system
of equations mentioned earlier (i.e. inflation, unemployment, and the Bank of Canada’s
policy rate) and replace the bank rate with this new series. We sum the three months
making up a given quarter to create a quarterly series from these authors’ monthly series.
Generally, variables in structural vector autoregressions are non-stationary, so the shock
series — which is stationary by construction — is cumulated in order to make it consistent
with the other series.

10
We depart from the authors in the variables which we do believe impact the monetary
policy shock variable contemporaneously. Like Champagne and Sekkel we add the Bank of
Canada’s commodity price index (ordered first). However, we also add FFR ahead of the
shock. We do this because, while the reaction function incorporates lagged FFR, at a
quarterly frequency14 we believe there are contemporaneous impacts of US monetary policy
on the Bank’s monetary policy decisions.15

h i
Formally, we define our vector of variables for each province as yt = pt , iUt S , it , πt , ut , ,
where pt is the commodity price index, iUt S is the Federal Funds Rate, it is the Champagne
and Sekkel monetary policy shock series, πt is Canadian or provincial inflation, and ut is
the Canadian or provincial unemployment rate. Therefore, the time-varying coefficient
model for each province is as follows16 :

yt = µt + δ1,t yt−1 + δ2,t yt−2 + ... + δp,t yt−p + β0,t −1 t , (1)

where δi,t , i = 1, ....p is a k × k matrix of time-varying coefficients, t is a k × 1 vector of

heteroskedastic shocks with covariance matrix Ωt . For our purposes we use two lags, i.e.
p = 2.17 In vectorized form, we can rewrite equation (1) as:

0 −1
yt = Xt δt + β0,t Σt et , (2)
14
While feasible to run the TVC-BVAR as a monthly analysis, it would be computationally very bur-
densome. Furthermore, it would require converting the monthly inflation and unemployment responses to
quarterly time series in order to run the second stage, as the provincial variables we use are only available
at a quarterly frequency. We did, however, rerun the results at a monthly frequency for Canada to compare.
We find that there is very little difference between the monthly and quarterly results. Results from the
monthly TVC-BVAR are available upon request.
15
Our results, available upon request, indicate that including the contemporaneous FFR matters both in
terms of direction and significance of the impact on inflation.
16
We also order the shock ahead of inflation and unemployment. Champagne and Sekkel put the shock
after these two variables, though their robustness checks show that this ordering is irrelevant. As provincial
inflation and unemployment does not tend to factor directly into monetary policy decisions, we order them
after the shock
17
We follow Imam (2015) in our use of two lags in order to protect degrees of freedom.

11
with
0 0 0 0
h i
Xt = In ⊗ (1, , yt−1 yt−2 , ..., yt−p

and
V [et ] = Ik .

The covariance matrix has the following structure:

−1 −1 0
Ωt = β0,t Σt Σt β0,t , (3)

where with five variables we define β0,t and Σt as:

 
 1 0 0 0 0 
 
 

 β1,t 1 0 0 0 

 
 
β0,t = 
 β2,t β3,t 1 0 0 ,

 
 

 β4,t β5,t β6,t 1 0 

 
 
β7,t β8,t β9,t β10,t 1

and  
 σ1t 0 0 0 0 
 
 

 0 σ2t 0 0 0 

 
 
Σt = 
 0 0 σ3t 0 0 .
 (4)
 
 

 0 0 0 σ4t 0 

 
 
0 0 0 0 σ5t

These definitions are standard in the literature and allow for both time-varying coefficients
and stochastic volatility (Koop and Korobilis 2009).

We assume that the parameters of interest δt , βt = [β1,r , β2,t , ...β10,t ] and

log(σt ) = log [σ1t , σ2t , σ3t , σ4t , σ5t ] follow random walks, implying that the covariance matrix

12
follows a random walk:
δt = δt−1 + eδt ; (5)

βt = βt−1 + eβt ; (6)

log(σt ) = log(σt−1 ) + eσt . (7)

All shocks are jointly normally distributed with a covariance matrix as in Imam (2015):

   
 et   Ik 0 0 0 
   
   

 eδt 


 0 V δ
0 0 

V =V   =  . (8)
eβt
   
β





 0 0 V 0 

   
   
eσt 0 0 0 Vσ

We estimate (2) subject to (5) through (8). The estimation is run using a Gibbs Sampler
where we perform 10,000 runs with a burn-in length of 2,000 in order to get closer to the
desired ergodic distribution.

3.1.1 Data

Outside of the monetary policy shock series from Champagne and Sekkel (2017) described
above, our data sources include Statistics Canada, the Bank of Canada, and the Federal
Reserve Economic Data database. All-items CPI data in CANSIM Table 326 0020 are used
to calculate year-over-year inflation rates for Canada and the ten provinces. We use the
labour force survey in CANSIM Table 282 0087 to get seasonally-adjusted unemployment
data for Canada and by province. We use the commodity price index from the Bank of
Canada, while we gather Federal Funds Rate data from the Federal Reserve Economic
Data database. Our sample period covers 1985:Q2 to 2015:Q4, and all variables are

13
measured in percentages.18

3.2 Stage One Impulse Response Function Results

The results are summarized for Canada and the provinces by the impulse responses and
peak impact figures in Appendix A. The impulse response functions indicate the response
of inflation and unemployment to a one-percentage point innovation in the Bank’s policy
rate. The confidence bands and the red point estimate line are for the first quarter in the
sample: 1985Q2. The remaining point estimates show the evolution of the impact of a
monetary policy shock at 5-year intervals. The peak impact graphs show the largest fall
(rise) in inflation (unemployment) from a contractionary monetary policy shock (with
confidence interval) across our time period. The estimated VARs are linear so the effects of
an expansionary monetary policy shock are symmetric by construction.

We measure the change in the impact of monetary policy shocks over time by changes in
the maximum or peak effect of the monetary policy shock on inflation and unemployment
or by the cumulative effect measured by the integral of the response over time. Maximum
inflation is defined as the peak change in inflation for a given monetary policy shock. For
an increase in the overnight rate this will be the largest (negative) change in inflation when
we calculate the impulse response function. A positive shock to the overnight rate leads to
an increase in unemployment, so the maximum response is the largest (positive) change in
the unemployment rate when we calculate the impulse response function. The cumulative
effects are calculated by summing the responses of either inflation or unemployment to the
policy shock over time until they die away.

We first discuss the results for Canada and then summarize the results across Canadian
18
Again, since we use provincial time series data in the second stage and are limited by the lack of monthly
time series data at the provincial level, we must use quarterly data. It is, therefore, appropriate to generate
estimates of the impact of monetary policy shocks in stage one at a quarterly frequency.

14
provinces.

Figures 3 and 4 give the results for Canada as a whole. The peak impacts of a monetary
policy shock on both inflation and unemployment are significant across time. However, the
magnitude of the declines in both cases mildly fall over the period under analysis. The
peaks for the two variables occur approximately when we would expect from a transmission
perspective, anywhere from 6-11 quarters after the monetary policy shock occurs.

The cumulative impact is harder to tell from the graphs but it turns out that we have seen
a bit of an improvement in the size of the response of inflation to a contractionary
monetary policy shock. These contradictory results for inflation are consistent with the
fact that on the one hand inflation targeting central banks have been more successful by
many metrics, including lower variability of inflation, and lower inflation levels, but also
firmer expectations have made it more difficult for policy changes to have the same impact
on inflation.

When we narrow the scope to Canada’s low-growth post-crisis economy, the impact of
monetary policy is weaker as measured either by the maximum or by the cumulative
change in inflation. This confirms the results other studies have found for a wide range of
countries: see for example Borio and Hofmann (2017).

The cumulative impact on unemployment falls from the beginning of our sample to the end
- consistent with the peak unemployment impact. However, it should be mentioned that
both cumulative and peak unemployment impact experience increases during the early
years (up until the end of 1992), fall for approximately three years, and flatten from the
mid-1990s on including through the crisis and after.

The remaining figures in Appendix A are for the different Canadian provinces.
Interestingly, the provincial results highlight the idiosyncratic nature of the impact of
monetary policy across the country. Across the board, contractionary monetary policy

15
shocks lead to the expected negative impact on inflation and positive impact on
unemployment. However, the evolution of the impact of monetary policy shocks on these
macroeconomic variables varies across the provinces. For example, a handful of provinces
experience no significant impact on unemployment, e.g. BC, Saskatchewan, PEI, New
Brunswick, and Alberta (post-crisis). We also see cases where the impact of a monetary
policy shock on inflation and/or unemployment has become larger (in absolute value
terms) over time, such as the case of Quebec (both) and BC (inflation). Ontario,
Manitoba, and Newfoundland all share similar stories to the overall Canadian experience.
What this all means is that monetary policy does not have uniform impacts across
provinces. This is important for central bankers to understand and make use of in their
modeling. This also makes for an interesting line of future research.

To summarize, whether the impact of monetary policy has increased or not over time (as
measured by maximum or cumulative changes to inflation and unemployment) is somewhat
unclear. However, while this is an important question to continue to evaluate, the primary
question we answer in this paper is what role has demographic change played on monetary
policy’s impact on key macroeconomic variables.

3.3 Stage-Two Methodology

To look at this question we use a dynamic ordinary least squares (D-OLS) panel regression
analysis. We take our measures of the impact of monetary policy shocks by province from
the previous section as dependent variables in a series of regressions.19 We test our results,
as in Imam (2015), on our four different dependent variables: maximum inflation,
maximum unemployment, cumulative inflation, and cumulative unemployment. For
notational purposes we label this variable ef f ectjit where i represents the different
19
Using generated variables as dependent variables raises no significant econometric issues. See Lewis and
Linzer (2005).

16
provinces in the panel, t is time, and j separates when we are looking at inflation versus
when we are looking at unemployment (and for that matter maximum vs. cumulative).

The explanatory variables we use (similar to Imam 2015) are the old-age dependency ratio,
manufacturing output relative to total output, a measure of the economy’s openness to the
rest of the world, and a measure of the importance of private-sector credit. We are
interested primarily in the effects of demographic change, but we use the other variables in
the regressions as controls in order to avoid biased results from omitted variables. Imam
also includes the share of smaller firms to total firms within an economy, which we are
unable to do as a result of data availability at the provincial level.

The use of the old-age dependency ratio as our demographic variable reflects an assumption
that the behaviour of different age cohorts is relatively constant over time. We checked this
assumption by analyzing data on the cross-cohort behaviour of assets and debts in Canada
in 1999, 2005, 2012 and 2016. On all four dates, debts peaked for the cohort aged 35–44
while assets peaked for the cohort aged 55–64 (Figure 1 above). Our interpretation is that
it is in fact the change in the age distribution of the population that matters, in other
words what is important is the changing relative sizes of different cohorts over time.

Since the purpose of this paper is to determine what impact aging has had on the impact
of monetary policy shocks, we should drill down further on this measure of changing
demographics. We begin by using CANSIM Table 282 0087 and subtract those aged 15–64
from those aged 15 and over to back out those aged 65 and over. At this point we are
consistent with Imam’s old age dependency ratio. However, because this arbitrarily defines
65 to be “old,” we subtract those aged 65 and older who remain in the labour force from
the total number of dependents.

This new ratio has steadily increased on balance over the period under analysis across all
provinces (Figure 2) except for Saskatchewan. Quebec and the Atlantic provinces have

17
experienced particularly large increases. Figure 2 is an index that normalizes all of the
series to 100 at the start of the sample. In the regressions we use the dependency ratios
without normalization.

Figure 2: Old Age Ratios — Provinces

3.3.1 Other control variables

There are several reasons to think that controlling for manufacturing, openness, and credit
is important. Manufacturing has been shown to have different price rigidities and
unemployment sensitivities compared with other industries (Dias et al. 2016 and Carlino
and Defina 1998). Next, theory suggests that a more open economy is more sensitive to an
increase in interest rates since the resulting appreciation in the exchange rate will have
greater impacts on net exports. Finally, the higher is private sector credit, the more
sensitive households and businesses are to movements in the interest rate.

As discussed, due to data availability at the provincial level, we are not able to create a
variable for the small firm ratio. For this reason, we end up with three control variables.
Furthermore, data availability forces some adjustments to the variables we do have. Both

18
provincial GDP and trade are available at annual frequencies only. We interpolate the
annual data to create quarterly data sets for each. To do this, we employ the Denton
method (see Bloem et al. 2001). To use this method we need an appropriate high-frequency
variable, and choose quarterly Canadian GDP. As shown in Table 1 the correlation
between annual Canadian GDP growth and annual provincial GDP growth is high. A
similar relationship holds between annual Canadian GDP growth and provincial trade
growth (New Brunswick is an exception). We can test how effective this method is by
comparing actual quarterly Canadian total trade growth to our estimated interpolated
quarterly Canada total trade growth. The correlation is once again quite high (0.823). We
are therefore confident that our interpolation method produces reliable measures of
quarterly provincial GDP and total trade.

CDA GDP - Prov GDP CDA GDP - Prov Trade

Alberta 0.731 0.737
British Columbia 0.720 0.817
Manitoba 0.474 0.567
New Brunswick 0.453 -0.157
Newfoundland 0.623 0.541
Nova Scotia 0.372 0.398
Ontario 0.906 0.841
PEI 0.446 0.519
Quebec 0.793 0.703
Saskatchewan 0.533 0.491

Table 1: Correlations between Canadian GDP and Provincial Data — Annual

Next, we generate the intensity of manufacturing variable. Using quarterly data from
CANSIM Table 304 0015 we focus on manufacturing sales by province, which we deflate by
dividing by province-specific consumer price indices. To determine the manufacturing
intensity we divide real manufacturing sales by real GDP, creating the variable mf git .

For openness, we use provincial export and import data to/from other countries from
CANSIM Table 384 0038. By adding exports and imports we create a measure of real total

19
trade which we divide by real GDP to create our openness measure, openit .

Last, we need a measure for private credit. This is important as credit levels have soared
due to cheap credit, which in theory should cause more sensitivity to monetary policy. We
note, though, that debt service ratios in Canada have remained fairly flat over the period
under analysis (Kronick 2017), which would negate some of this sensitivity. That said,
while a complete private credit story by province is difficult to get, asset and liability data
broken down by province is available from chartered banks in CANSIM Table 176 0074.
Since chartered banks make up the bulk of the lending in the economy, we believe we have
an appropriate proxy measure. We include the following in private credit: 1) non-mortgage
loans to individuals for non-business purposes (total personal loans); 2) non-mortgage loans
to individuals and others for business purposes (total loans to Canadian resident
non-financial business); 3) residential mortgages; 4) non-residential mortgages; and 5)
customers’ liabilities under acceptances. We again divide this private credit data by real
GDP by province to create the final control variable, creditit .

We use D-OLS to run the following regression equation with two lags and one lead:20

ef f ectjit = α0 + α1 oldageit + α2 mf git + α3 creditit + α4 openit + ηit . (9)

Due to data availability the sample period in this second stage runs from 1992:Q1 to
2015:Q4. There are a few different benefits to using D-OLS. First, the use of leads and lags
of the first differences of all the independent variables helps mitigate endogeneity concerns.
That said, our dependent variables are the reactions of inflation and unemployment to an
exogenous monetary policy shock, and while it is clear why an aging population would
impact these variables, it is not clear how they would impact the old age dependency ratio,
which is our variable of interest. In other words, we do not see a reverse causality concern.
20
The advantage of using levels when your variables are cointegrated is that one does not need to then
include an error correction term of a possibly unknown cointegration structure. See Lütkepohl (2013) for
more details.

20
While we choose two lags and one lead as our default, our robustness checks in Section 4
confirm our results for different lead and lag choices.

Another benefit of D-OLS, as indicated in Imam (2015), is that it ensures the stochastic
error term is independent of past shocks, and the panel estimator is more efficient and
robust. Lastly, the D-OLS method controls for any fixed effects across our panels.

3.3.2 Stationary tests

Before running equation (9) in levels, we check the stationarity of the variables by using
panel unit root tests, which differ from some of the standard time series unit root tests
such as Dickey-Fuller. We run two tests. First, the Im-Pesaran-Shin (2003) test, with a
null hypothesis that all panels are unit root and an alternative hypothesis that at least one
panel is stationary. The advantage with the Im-Pesaran-Shin test is it allows for different
unit root coefficients across panels. The second test is the Dickey-Fuller version of the
Fisher-type tests, which perform unit root tests on each individual panel and then combine
the p-values to produce an overall test. The null is that all panels contain unit roots, while
the alternative is that at least one panel is stationary.21 Table 2 shows that our dependent
variables all contain unit roots.

Im-Pesaran-Shin ADF – Fisher Chi-square

Cumulative Inf 0.353 0.480
Cumulative Unemp 0.129 0.208
Max Inf 0.314 0.366
Max Unemp 0.531 0.004

Table 2: Dependent Variable Unit Root Tests — p-values

For our independent variables, old age and credit clearly contain unit roots, and we can
21
Imam (2015) also uses the Levin-Liu-Chu (2002) test, with a null that all panels are unit root, and the
alternative that all panels are stationary. The disadvantage of this test is it assumes a homogeneous unit
root coefficient across all panels.

21
only weakly reject a unit root for manufacturing (Table 3). Further, for manufacturing,
when we use the Hadri (2000) test — not shown here — with a null of stationarity of all
panels, it can also be rejected in favour of the alternative that some panels contain unit
roots. Openness results suggest the variable is already stationary. Table 4 indicates that all
variables are stationary in first differences.

Im-Pesaran-Shin ADF – Fisher Chi-square

Old Age 1.000 0.995
Manufacturing 0.055 0.089
Credit 1.000 1.000
Openness 0.000 0.000

Table 3: Control Variable Unit Root Tests — p-values

Im-Pesaran-Shin ADF – Fisher Chi-square

Cumulative Inf 0.000 0.000
Cumulative Unemp 0.000 0.000
Max Inf 0.000 0.000
Max Unemp 0.000 0.000
Old Age 0.000 0.000
Manufacturing 0.000 0.000
Openness 0.000 0.000
Credit 0.000 0.000

Table 4: Unit Root Tests (First Differences) — p-values

We then check for cointegration to determine the appropriate way to run equation (9). In
particular, if there is cointegration across our variables then it is possible to still run
equation (9) without differencing. As indicated in Table 5, the Kao (1999) test for
cointegration suggests that all of our dependent variables, when paired with all our
independent variables, are cointegrated (the null hypothesis is no cointegration). We
therefore estimate equation (9) in levels as our primary empirical test.

22
 Augmented Dickey-Fuller t
Cumulative Inf 0.001
Cumulative Unemp 0.003
Max Inf 0.004
Max Unemp 0.005

Table 5: Cointegration Test — p-values

3.4 Stage-Two Results

Table 6 has the empirical results for each of our four dependent variables. The monetary
shock that we model in stage one is a one percentage point increase in the Bank’s policy
rate, which has a negative effect on inflation and a positive effect on unemployment. This
means that a positive (negative) sign in the table means that the relevant variable reduces
(increases) the impact of monetary policy shocks on inflation and increases (reduces) their
impact on unemployment.

(1) (2) (3) (4)

cum_inf cum_unemp max_inf max_unemp
oldageit -0.0291 -0.0590∗∗ 0.00341∗∗ -0.00605∗∗∗
(-1.31) (-2.06) (2.00) (-2.64)

mf git 0.0924∗∗∗ 0.218∗∗∗ 0.00782∗∗∗ 0.0141∗∗∗

(5.53) (10.10) (6.08) (8.17)

creditit -0.0393∗∗∗ -0.0153∗∗∗ -0.00239∗∗∗ -0.000103

(-13.63) (-4.10) (-10.81) (-0.35)

openit 0.00174 -0.00840∗ 0.000399 -0.000131

(0.52) (-1.94) (1.55) (-0.38)
Observations 920 920 920 920
t statistics in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 6: Primary Regression Results

The results for all four dependent variables are broadly consistent with Imam (2015) in

23
direction and significance, though with smaller magnitudes. The key result for our
purposes is the fact that an aging population does appear to reduce the impact of
monetary policy shocks on both inflation and unemployment. The effect is stronger on
unemployment. The results are consistent for unemployment whether we use cumulative or
maximum as the dependent variable. With inflation, the effect is only statistically
significant when we look at maximum inflation. The more mixed results for inflation are
consistent with the anchored inflation expectations that have become more widespread
since the Bank of Canada became an inflation-targeting central bank.

We also note that, while the standard 1 percent increase in old age is an appropriate way of
analyzing the question at hand, perhaps more interesting given the slow moving nature of
demographics, is the impact of the entire increase in our aging population on the impact of
monetary policy shocks over the period we study. In Canada as a whole, the old age
dependency ratio increased 33.8 percent from Q1 1992 to Q4 2016. Therefore, the impact
of policy shocks, as judged by maximum inflation, fell by 0.115 percent (33.8 × 0.00341), or
approximately 12 percentage points. So if the average maximum fall in inflation from a
contractionary monetary policy shock was 0.40 percentage points across panels over the
period of our analysis, it would have fallen 0.52 percentage points if not for an aging
population.

Given the insignificance of an aging population on cumulative inflation we do not perform

a similar analysis on this variable.

Looking at unemployment, multiplying 33.8 by the coefficients on cumulative and

maximum unemployment gives us -1.994 and -0.204 respectively. Therefore, an
expansionary monetary policy shock at the beginning of our sample that lowered the
unemployment rate by 1.994 percentage points would have no impact today. Similarly, if
the peak decrease in the unemployment rate from an expansionary monetary policy shock
in 1992 had been 0.204 percentage points, it would be zero now.

24
3.5 Control Variables

What about our control variables? We find that an economy with a higher concentration of
manufacturing causes greater unemployment sensitivity in response to monetary policy
shocks. This is consistent with the idea that the manufacturing sector tends to be more
interest rate sensitive (see, for example, Carlino and Defina 1998). This result does not
hold for inflation sensitivity, though this is consistent with Imam.

The results also indicate that a more open economy has no significant impact on the
impact of monetary policy shocks. Even in the one case where there is some weak
significance, the magnitude is so small we consider it negligible. The insignificance can be
explained by two opposing considerations. On the one hand, an increase in interest rates
will cause an appreciating exchange rate, which for a more open economy should lead to a
greater fall in inflation and a larger increase in unemployment. However, increased global
trade integration might actually weaken the linkages between monetary policy and key
macroeconomic variables (Poloz 2016). In large part this is because domestic inflation is
affected more by global demand and supply than changes in domestic behaviour. China’s
ascension in the global market and the trade collapse during the Great Recession are both
examples of significant international macroeconomic events that have had ripple effects
domestically, and the challenge is to figure out how inflation responds in this newer
environment and how central banks should react.

Finally, the coefficients on private credit tell a somewhat mixed story. For the regressions
with inflation as dependent variables we get consistently significant and negative
coefficients. This is consistent with the idea that more credit present in the Canadian
economy leads to increased sensitivity to changes in the interest rate, which then have a
greater impact on spending and inflation.

For unemployment, the results are either insignificant or indicate a mildly less sensitive real

25
economy. One possible explanation for this counter-intuitive result is that we have
experienced mostly expansionary monetary policy over our time period, and the more
credit-constrained individuals are the less space they have in their budgets to borrow and
spend when interest rates are lowered. If that occurs, there is less of an increase in
aggregate demand, less economic growth, and a more attenuated unemployment rate. This
explanation, however, should lead to lower inflation as a result of reduced spending.
However, if the spending that did occur went to parts of the economy that experienced
above-average inflation over our period (for example housing), then both effects could
apply simultaneously.

Overall, though, the key takeaway from the primary results is that an aging Canadian
population has acted as a drag on the impact of monetary policy changes. Below we will
look at the transmission mechanism(s) that are responsible for these results, but this
demographic drag is what we want the readers to focus on when we discuss the
implications for the Bank of Canada in the conclusion.

3.6 Transmission Mechanism

The next logical question to ask, once we know that aging acts like a drag on the impact of
monetary policy changes, is to look for explanations why. What is happening in the
economy that causes this phenomenon? To answer these questions, we use empirical
regression techniques to test the importance of the interest rate and credit channels.
Unfortunately, we do not have access to wealth data at the provincial level. Wealth data is
available at the national level, but as we have mentioned, cross-sectional and time series
variation in the data are both needed in order to properly evaluate very slow-moving
variables such as old age.

We use two techniques to test the impact of the interest rate and credit channels. Since

26
they are closely linked, we test them together. The first technique involves removing the
private credit control variable from the primary regression. The second technique involves
interacting the private credit variable with the old age variable.

Table 7 shows the results for the variable removal technique. For inflation, removing private
credit leads to the appearance of old age having a positive effect on the impact of monetary
policy shocks. This is precisely what we would expect since an increase in credit on its own,
i.e. not accounting for aging, should cause an increase in the impact of monetary policy
changes as higher debt loads mean more sensitivity to interest rate movements which
funnels down towards household spending, and, therefore, inflation. When we remove this
variable, its correlation with the age pyramid causes it to impact the coefficient on old age
in the direction of increased sensitivity. What this implies, though, is that, if not for aging,
monetary policy changes would have had a stronger impact via their effect on credit.

For unemployment we also see an impact from the removal of private credit, but this time
there is a mild decrease in impact of monetary policy shocks with aging. This is consistent
with the discussion above on the role credit appears to be playing on the real economy, as
it exacerbates the already decreasing impact from old age on unemployment.

Table 8 shows results for the variable interaction technique. The way to interpret the role
of old age is to add the coefficient on old age to the interaction coefficient multiplied by
some value of private credit (in our case we take the mean). When one does that one gets
coefficients matching the primary results across the board. What is relevant is the fact
that, with the exception of the cumulative inflation regression, the interaction terms are
significant in all cases, highlighting the link between credit and old age. The insignificance
of the interaction coefficient in the cumulative inflation regression is consistent with the
insignificance of old age for this same regression in our primary specification. The bottom
line is that these results are further evidence that the interest rate and credit channels play
an important role as transmission mechanisms.

27
 (1) (2) (3) (4)
cum_inf cum_unemp max_inf max_unemp
oldageit -0.0893∗∗∗ -0.0824∗∗∗ -0.000264 -0.00621∗∗∗
(-4.14) (-2.97) (-0.15) (-2.74)

mf git 0.0508∗∗∗ 0.202∗∗∗ 0.00529∗∗∗ 0.0140∗∗∗

(2.97) (9.19) (3.83) (7.80)

openit 0.00973∗∗∗ -0.00529 0.000885∗∗∗ -0.000111

(2.84) (-1.20) (3.20) (-0.31)
Observations 920 920 920 920
t statistics in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 7: Transmission Mechanism — Interest Rate/Credit: Removal Technique

(1) (2) (3) (4)

cum_inf cum_unemp max_inf max_unemp
oldageit -0.0343 -0.391∗∗∗ 0.0129∗∗∗ -0.0329∗∗∗
(-0.67) (-6.30) (3.71) (-7.07)

mf git 0.0934∗∗∗ 0.238∗∗∗ 0.00728∗∗∗ 0.0157∗∗∗

(5.70) (11.94) (6.54) (10.50)

creditit -0.0385∗∗∗ -0.156∗∗∗ 0.00171∗ -0.0115∗∗∗

(-2.71) (-9.00) (1.77) (-8.85)

openit 0.00167 -0.00810∗∗ 0.000387∗ -0.000105

(0.51) (-2.02) (1.73) (-0.35)

interactionit -0.000113 0.00755∗∗∗ -0.000223∗∗∗ 0.000613∗∗∗

(-0.15) (8.25) (-4.37) (8.93)
Observations 920 920 920 920
t statistics in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 8: Transmission Mechanism — Interest Rate/Credit: Integration Technique

28
4 Robustness Checks

A lack of provincial variables makes robustness checks difficult to perform. Therefore, we

focus primarily on the different combinations of leads and lags in our D-OLS panel
regression.22 Tables 9 and 10 show the results for 2 leads and lags and 3 leads and lags
respectively. The results indicate that our primary variable of interest oldageit does not
change from our primary results at all at two leads and lags. At three leads and lags, the
impact on inflation from aging is more ambiguous, while the results for unemployment do
not change. Therefore, the timing of the change of the different independent variables does
not significantly change the contemporaneous effect of aging on the impact of monetary
policy shocks.

(1) (2) (3) (4)

cum_inf cum_unemp max_inf max_unemp
oldageit -0.0318 -0.0658∗∗ 0.00346∗∗ -0.00655∗∗∗
(-1.42) (-2.27) (2.00) (-2.83)

mf git 0.0960∗∗∗ 0.221∗∗∗ 0.00801∗∗∗ 0.0142∗∗∗

(5.68) (10.14) (6.16) (8.14)

creditit -0.0403∗∗∗ -0.0171∗∗∗ -0.00243∗∗∗ -0.000199

(-13.85) (-4.54) (-10.86) (-0.66)

openit 0.000922 -0.00962∗∗ 0.000367 -0.000189

(0.27) (-2.20) (1.41) (-0.54)
Observations 910 910 910 910
t statistics in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 9: Two Leads and Lags

22
That said, we did find that using the old age dependency ratio as in Imam (2015), i.e. without removing
those above 65 who continue to work, did not change our findings. Additionally, removing outlier periods did
not change our results. Lastly, performing the transmission mechanism analysis whereby we stop pre-crisis
did not change the importance of the credit channel.

29
 (1) (2) (3) (4)
cum_inf cum_unemp max_inf max_unemp
oldageit -0.0454∗∗ -0.0846∗∗∗ 0.00308∗ -0.00778∗∗∗
(-1.98) (-2.85) (1.74) (-3.28)

mf git 0.107∗∗∗ 0.232∗∗∗ 0.00848∗∗∗ 0.0147∗∗∗

(6.20) (10.41) (6.38) (8.26)

creditit -0.0444∗∗∗ -0.0222∗∗∗ -0.00260∗∗∗ -0.000471

(-14.92) (-5.78) (-11.37) (-1.53)

openit -0.00143 -0.0123∗∗∗ 0.000271 -0.000313

(-0.41) (-2.76) (1.02) (-0.88)
Observations 890 890 890 890
t statistics in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 10: Three Leads and Lags

5 Conclusion

Our results show that the inflation targeting policy framework in Canada is in reasonably
good shape, but is not perfect. A lower neutral interest rate means the Bank of Canada
starts from a lower neutral position and therefore has less room to decrease rates before
hitting the zero (or small negative) lower bound. Future crises and/or recessions will likely
put the Bank in the position of having to resort to unconventional monetary policies
sooner. If further population aging forces the Bank to have to make larger changes to the
overnight rate to generate the same economic impacts, this could eventually undermine
Canada’s inflation targeting regime.

This paper first evaluated the impact of monetary policy shocks on economic variables
from the mid-1980s onward, and second looked at the link between an aging population
and the impact of policy shocks in the years since the Bank of Canada became an
inflation-targeting central bank. We improve on previous empirical work on the effect of

30
Canadian monetary policy shocks by using a shock series that removes the price puzzle
found in standard structural vector autoregression setups. Furthermore, to exploit the need
for greater variance given the slow-moving nature of old age, we use a panel data set at the
provincial level, giving us both cross-sectional and time series variation and leading to
more accurate results for Canada.

We find mixed results concerning the changing impact of monetary policy shocks over the
last 30 years. Some of this is might be considered a good thing from the Bank’s
perspective, as inflation expectations have become more and more well-anchored. Our
results are broadly consistent with the literature, including the weakening of the standard
relationship between inflation and unemployment.

Our main focus was the role aging played on changes in the impact of discretionary
monetary policy. We find that Canada’s aging population has acted as a drag and is likely
a leading cause of the systematic undershooting of inflation we have seen since the financial
crisis. We also find that the interest rate and credit channels help explain this result.
Specifically, an aging population which takes on less debt is less sensitive to changes in the
interest rate.

This means that hitting the Bank of Canada’s inflation target will require more significant
changes to the overnight rate target. While we acknowledge the role fiscal policy, and
immigration policy in particular, can play in slowing down the aging effect, our focus is on
monetary policy. In the case of expansionary monetary policy, making more significant
changes to the overnight rate target will be made more difficult by a lower neutral rate of
interest, which may result in a quicker move to unconventional monetary policy. The
results in this paper should help the Bank of Canada adjust their analysis accordingly.

31
References

Ambler, Steve (2016), “Putting Money to Work: Monetary Policy at the Zero Lower
Bound.” E-Brief, 249, C.D. Howe Institute
https://www.cdhowe.org/sites/default/files/attachments/research_papers/
mixed/E-brief_249.pdf

Ambler, Steve and Craig Alexander (2015), “One Percent? For Real? Insights from Modern
Growth Theory about Future Investment Returns.” E-Brief, 216, C.D. Howe Institute
https://www.cdhowe.org/public-policy-research/
one-percent-real-insights-modern-growth-theory-about-future-investment-returns

Baker, Dean, Bradford DeLong, and Paul Krugman (2002), “Asset Returns and Economic
Growth.” Brookings Papers on Economic Activity 36, 289–330.

Berja, Martin, Andreas Fuster, Erik Hurst, and Joe Vavra (2018), “Regional Heterogeneity
and the Refinancing Channel of Monetary Policy.” Staff Report 731, Federal Reserve
Bank of New York
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2619932

Bloem, Adriaan, Robert Dippelsman and Nils Maehle (2001), “Quarterly National
Accounts Manual: Concepts, Data Sources, and Compilation.” International Monetary
Fund, Washington, DC
http://www.imf.org/external/pubs/ft/qna/2000/Textbook/index.htm

Boivin, Jean and Marc Giannoni (2006), “Has Monetary Policy Become More Effective?”
Review of Economics and Statistics 88, 445–462

Boivin, Jean, Michael Kiley and Frederic Mishkin (2010), “How Has the Monetary
Transmission Mechanism Evolved Over Time?” Working Paper 15879, National Bureau
of Economic Research

Borio, Claudio and Boris Hofmann (2017), “Is Monetary Policy Less Effective when Interest
Rates are Persistently Low?” Working Paper 628, Bank for International Settlements
https://www.bis.org/publ/work628.htm

Canova, Fabio and Luca Gambetti (2009), “Structural Changes in the US Economy: Is
There a Role for Monetary Policy?” Journal of Economic Dynamics and Control 33,
477–490

Carlino, Gerald and Robert Defina (1998), “The Differential Regional Effects of Monetary
Policy.” Review of Economics and Statistics, 8, 572–587

Carney, Mark (2017), “[De]Globalisation and Inflation.” 2017 IMF Michel Camdessus
Central Banking Lecture, September 18

32
http://www.bankofengland.co.uk/speeches

Champagne, Julien and Rodrigo Sekkel (2017), “Changes in Monetary Regimes and the
Identification of Monetary Policy Shocks: Evidence from Canada.” Staff Working Paper
2017-39, Bank of Canada
http://www.bankofcanada.ca/2017/09/staff-working-paper-2017-39/

Christiano, Lawrence, Martin Eichenbaum and Charles Evans (1996), “The Effects of
Monetary Policy Shocks: Evidence from the Flow of Funds.” Review of Economic
Statistics 78, 16–34

Cloyne, James, Clodomiro Ferreira, and Paolo Surico (2017), “Monetary Policy when
Households have Debt: New Evidence on the Transmission Mechanism.” Working
Paper 589, Bank of England
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2765415

Dias, Daniel, Carlos Marques and Christine Richmond (2016), “A Tale of Two Sectors:
Why is Misallocation Higher in Services than in Manufacturing?” Working Paper
16/220, International Monetary Fund
http://www.imf.org/en/Publications/WP/Issues/2016/12/31/
A-Tale-of-Two-Sectors-Why-is-Misallocation-Higher-in-Services-than-in-Manufacturing-443

Georgiadis, Georgios (2015), “Examining Asymmetries in the Transmission of Monetary

Policy in the Euro Area: Evidence from a Mixed Cross-section Global VAR Model.”
European Economic Review 75, 195–215

Grilli, Vittorio and Nouriel Roubini (1996), “Liquidity Models in Open Economies: Theory
and Empirical Evidence.” European Economic Review 40, 847–859

Hadri, Kaddour (2000), “Testing for stationarity in heterogeneous panel data.”

Econometrics Journal 3: 148–161.

Im, Kyung So, M. Hashem Pesaran, and Yongcheol Shin (2003), “Testing for unit roots in
heterogeneous panels.” Journal of Econometrics 115: 53–74.

Imam, Patrick (2015), “Shock from Graying: Is the Demographic Shift Weakening
Monetary Policy Effectiveness?” International Journal of Finance and Economics 20,
138–154

Kao, Chihwa (1999), “Spurious Regression and Residual-Based Tests for Cointegration in
Panel Data.” Journal of Econometrics 90, 1–44

Kaplan, Greg, Ben Moll, and Gianluca Violante (2018), “Monetary Policy According to
HANK.” American Economic Review 108, 697–743

Kim, Soyoung and Nouriel Roubini (2000), “Exchange Rate Anomalies in the Industrial
Countries: A Solution with a Structural VAR approach.” Journal of Monetary

33
 Economics 45, 561–586

Koop, Gary and Dimitris Korobilis (2009), “Bayesian Multivariate Time Series Methods for
Empirical Macroeconomics.” Foundations and Trends in Econometrics 3, 267–358

Kronick, Jeremy (2017), “Are Canadians Acting Like they are “House Poor” or “House
Rich”?” Commentary, 482, C.D. Howe Institute
https://www.cdhowe.org/public-policy-research/
spendthrifts-and-savers-are-canadians-acting-they-are-%E2%80%9Chouse-poor%
E2%80%9D-or-%E2%80%9Chouse-rich%E2%80%9D

Levin, Andrew, Chien-Fu Lin, and Chia-Shang James Chu (2002), “Unit root tests in panel
data: Asymptotic and finite-sample properties.” Journal of Econometrics 108, 1–24

Lütkepohl, Helmut (2013), “Vector Autoregressive Models.” Handbook of Research Methods

and Applications in Empirical Macroeconomics, chapter 6, pages 139-164 Edward Elgar
Publishing.

Lewis, Jeffrey and Drew Linzer (2005), “Estimating Regression Models in Which the
Dependent Variable Is Based on Estimates.” Political Analysis 13, 345–364

Mishkin, Frederic (1996), “The Channels of Monetary Transmission: Lessons for Monetary
Policy.” Working Paper 5464, National Bureau of Economic Research

Modigliani, Franco (1966), “The Life Cycle Hypothesis of Saving, the Demand for Wealth
and the Supply of Capital.” Social Research 33, 160–217

Parkin, Michael (2016) “Mounting Evidence: Findings from Natural Experiments in

Inflation Targeting.” E-Brief 233, C.D. Howe Institute
https://www.cdhowe.org/sites/default/files/attachments/research_papers/
mixed/e-brief_233.pdf

Poloz, Stephen (2016), “The Paul Storer Memorial Lecture — Cross-Border Trade
Integration and Monetary Policy.”
http://www.bankofcanada.ca/wp-content/uploads/2016/09/sdp2016-20.pdf

Primiceri, Georgio (2005), “Time Varying Structural Vector Autoregressions and Monetary
Policy,” Review of Economic Studies 72, 821–852

Rasche, Robert and Marcela Williams (2007), “The Effectiveness of Monetary Policy.”
Federal Reserve Bank of St. Louis Review September, 447–490

Robson, William (2017), “The Fiscal Implications of Canadians’ Working Longer.” E-Brief,
268, C.D. Howe Institute
https://www.cdhowe.org/sites/default/files/attachments/research_papers/
mixed/Final%20e-brief_268_web.pdf

34
Romer, Christina and David Romer (2004), “A New Measure of Monetary Shocks:
Derivation and Implications.” American Economic Review 94, 1055–1084

Wong, Arlene (2018), “Population Aging and the Transmission of Monetary Policy to
Consumption.” Draft, Princeton University
https://static1.squarespace.com/static/576576adbe659449f97e0d35/t/
5abcfec903ce64f01a071986/1522335434551/Paper_Jan2018.pdf

35
Appendix A

Figure 3: Monetary Policy Shock - Effect on Inflation and Unemployment (Canada)

Figure 4: Peak Impact - Inflation and Unemployment (Canada)

36
Figure 5: Monetary Policy Shock - Effect on Inflation and Unemployment (Saskatchewan)

Figure 6: Peak Impact - Inflation and Unemployment (Saskatchewan)

37
Figure 7: Monetary Policy Shock - Effect on Inflation and Unemployment (Quebec)

Figure 8: Peak Impact - Inflation and Unemployment (Quebec)

38
Figure 9: Monetary Policy Shock - Effect on Inflation and Unemployment (PEI)

Figure 10: Peak Impact - Inflation and Unemployment (PEI)

39
Figure 11: Monetary Policy Shock - Effect on Inflation and Unemployment (Ontario)

Figure 12: Peak Impact - Inflation and Unemployment (Ontario)

40
Figure 13: Monetary Policy Shock - Effect on Inflation and Unemployment (Nova Scotia)

Figure 14: Peak Impact - Inflation and Unemployment (Nova Scotia)

41
Figure 15: Monetary Policy Shock - Effect on Inflation and Unemployment (Newfoundland)

Figure 16: Peak Impact - Inflation and Unemployment (Newfoundland)

42
Figure 17: Monetary Policy Shock - Effect on Inflation and Unemployment (New Brunswick)

Figure 18: Peak Impact - Inflation and Unemployment (New Brunswick)

43
Figure 19: Monetary Policy Shock - Effect on Inflation and Unemployment (Manitoba)

Figure 20: Peak Impact - Inflation and Unemployment (Manitoba)

44
Figure 21: Monetary Policy Shock - Effect on Inflation and Unemployment (BC)

Figure 22: Peak Impact - Inflation and Unemployment (BC)

45
Figure 23: Monetary Policy Shock - Effect on Inflation and Unemployment (Alberta)

Figure 24: Peak Impact - Inflation and Unemployment (Alberta)

46