July 1997
Note: This is a draft working paper circulated for technical
consultation. It reflects the personal analysis of the author,
and it should not be construed as in any way representing the
official policy of the Ontario Government. Comments would be
most welcome.
Introduction
The developed economies of the world are characterized by
instantaneous communications. Nowhere is this more evident than
in financial markets, where events in the United States are immediately
reflected in the pricing of assets in other countries. Observation
of these co-movements gives rise to a view that financial markets
have become highly integrated. Sometimes, it is claimed that
the economic and financial policies of smaller countries like
Canada are no longer important in determining their own economic
destinies.
The degree of financial market integration is an important
question that can only be answered by empirical analysis. There
is no doubt that there is near-perfect integration among countries
and regions that use a common currency. However, integration
is likely to be less than perfect where different currencies are
involved.
Canadian interest rates usually move in the same direction
as U.S. interest rates when the latter change substantially.
However, there is also a considerable amount of independent variation
in Canadian interest rates. This differential can vary widely
from year to year.
The purpose of this paper will be to estimate how much this
differential is influenced by changes in Canada's current account
balance. The current account represents the total amount that
Canada must finance through capital inflows (either in the form
of debt or equity). It is the best summary indicator of the total
pressure on financial markets from the demand for capital in Canada.
The Relationship between the Current Account Deficit and Government
Deficits
The current account balance is the outcome of all saving,
borrowing and investment decisions in the Canadian economy. Hence,
it is influenced by monetary policy, by total government borrowing,
and by private sector saving and investment.
It overcomes some of the difficulties that have been experienced
in past attempts to measure the effects of government deficits
on interest rates. Other things equal, higher demand for credit
due to a higher government deficit leads to higher interest rates.
However, on many occasions this is obscured by offsetting changes
in private sector borrowing, or changes in monetary policy that
increase or decrease the quantity of credit.
This ambiguity of the effect of government deficits on interest
rates was highlighted by Evans (1985), who found that, for the
United States, there was a historical tendency for higher deficits
to be associated with lower interest rates. Contrary to Evans,
Spiro (1990) found that interest rates rose in response to a higher
debt to GDP ratio. However, a similar analysis by Ford and Laxton
(1995) using a shorter, more recent sample period found that the
debt to GDP ratio is statistically insignificant even for a large
country like the United States. Siklos (1988) and Darrat and
Suliman (1991) found no effect of Canadian deficits on interest
rates. Other recent studies, such as Orr, Edey and Kennedy (1995)
and Fillion (1996) reported some statistically significant results,
but only by imposing constraints on their models, some of which
may be unrealistic. Ford and Laxton (1995) found that
the combined debt to GDP ratio of all the industrial countries
raises the "world real interest rate." For each individual
country they looked at, they found that its debt ratio was not
significant at explaining its interest rates. They concluded
that this is an indication of a high degree of international capital
market integration. However, Ford and Laxton used data on domestic
debt, so their findings may not address the issue of capital market
integration. Changes in the total debt of a country need not
be highly correlated with changes in its foreign debt. Only if
interest rates are insensitive to changes in foreign debt can
one conclude that international capital flows are perfectly elastic.
The relationship between government deficits and the current
account balance is sometimes misunderstood. In some cases, an
increase in the government deficit is quickly matched by the current
account deficit. This appears to have been the case in the United
States in the 1980s, and gave rise to the "twin deficits"
hypothesis. It is, of course, also possible for a country to
have a large current account deficit while having a government
budget surplus (as was the case in Mexico in 1995, prior to its
financial crisis); or a current account surplus with a large government
deficit, as has occurred in some years for Belgium and Italy.
Looking at the historical Canadian data, there have been periods
when there was relatively little correlation between government
deficits and the current account balance, as noted by Poloz (1992).
These examples illustrate that the effects of high private
sector borrowing (as in the case of Mexico) or saving (as in the
case of Belgium and Italy) can offset the government deficit.
In Canada's case, volatile monetary policy has often had large
impacts on the current account deficit. In the period following
1987, tighter monetary policy caused a sharp appreciation in the
Canadian dollar's exchange rate and an increase in the current
account deficit.
This historical experience tells us that government deficits
are not the only cause of changes in the current account deficit.
However, that does not negate the hypothesis that they are one
of the leading causes of changes in the current account deficit.
At a point in time, there is a fixed pool of savings in Canada.
If the government increases its deficit, this might be offset
by an increased supply of money from the Bank of Canada, but this
is only likely to happen when the economy is depressed and inflation
is falling. Alternatively, there could be increased saving or
less borrowing from the private sector.
In fact, while higher interest rates would crowd out some
private investment, and may generate some additional saving, empirical
studies have rejected the contention that higher higher government
deficits are fully offset by such changes, as found by Johnson
(1995). While the effect may take place with a time lag, and
may be less than one to one, logic dictates that changes in government
deficits lead to changes in the current account deficit under
equilibrium conditions. The key to this is the exchange rate:
it must be allowed to adjust downwards in response to a reduced
fiscal deficit, in order to induce an increase in net exports.
The central bank needs to accept this and accommodate it. This
outcome is contrary to the prediction frequently made by financial
journalists, that reduced deficits lead to an appreciating exchange
rate as a result of improved confidence.
The Effect of Monetary Policy on Interest Rates and the Current
Account
In an open economy the central bank's monetary policy is
only one of two sources of liquidity. International borrowing
is an alternative source. In the extreme case, where international
capital flows are perfectly elastic, the simplest form of the
Mundell theory predicts that there will be no increase in interest
rates due to a tighter monetary policy. The tightening occurs
entirely through the appreciation of the exchange rate caused
by higher capital inflows.
If a monetary policy change the exchange rate, it inevitably
also moves the current account. In an open economy, monetary
policy cannot be effectively tightened without inducing an increase
in capital inflows and a corresponding increase in the current
account deficit. The elasticity of capital inflows determines
how much the current account deficit will have to increase in
order to bring about the magnitude of the increase in interest
rates that the central bank desires. This is one reason for the
very strong inverse relationship between interest rates and the
current account balance seen in Figure 1.
An Empirical Model of the Canada-U.S. Short-Term Interest Differential
1. Factors Determining Canadian Short-Term Interest Rates
Canadian financial markets are closely linked to those of
the United States, and the U.S. T-bill rate explains 76 percent
of the variance of the Canadian T-bill rate even over the recent
history from 1988 to 1996. It is the differential between Canadian
and U.S. interest rates which is the focus of discussion in Canada,
and this is what will be used as the dependent variable in the
econometric analysis discussed below.
The unemployment rate can be used to represent the economic
environment which determines the Bank of Canada's monetary policy
decisions. The unemployment rate is a better variable in this
regard than the inflation rate. The Bank of Canada targets the
inflation rate, but it does so in a forward looking way. It does
not wait for the inflation rate to rise before it tightens monetary
policy. Rather, it looks at the unemployment rate as a predictor
of how inflation will behave in the future.
The current account balance represents marginal net borrowing
by the Canadian economy. An increase in the current account deficit
represents an upward shift in the demand curve for funds in Canada
without any corresponding upward shift in the supply curve. Unless
that foreign supply of funds to Canada is perfectly elastic, this
increase in demand will cause an increase in Canadian interest
rates. As noted above, when government borrowing rises in Canada,
there are often offsetting reductions in private sector borrowing,
effectively shifting up the net supply curve.
The Canada-U.S. interest differential was regressed on polynomial
distributed lags of the current account to GDP ratio, the Canada-U.S.
unemployment rate differentials, and the Canada- U.S. inflation
rate differentials. All of these explanatory variables were found
to have significant t-statistics. Regressing over the 1977 to
1996 sample period, the sum of coefficients on the current account
balance (lagged from t-2 to t-6) was -0.60, with a t-statistic
of -4.1. This implies that the Canada-U.S. three month T-bill
differential would eventually rise by 60 basis points for every
1 percentage point increase in the current account balance as
a percent of GDP.
One interesting result that emerges from these regressions
is that the Canada-U.S. inflation differential affects the T-bill
differential with a very long lag, and has little short-term impact.
Values of the inflation differential from 3 years earlier are
significant, with a sum of coefficients of 0.3. This is consistent
with the fact that purchasing power parity does not hold in the
short run. The Canadian inflation rate only matters to foreign
investors if it gets reflected in the exchange rate, and this
does not happen in the short run.
2. Ordinary Least Squares May Not be Reliable
The strong inverse relationship between the interest differential
and the current account ratio can be seen in Figure 1. It is not
obvious that one of them leads the other, and economic theory
tells us that causality is likely to run in both directions. A
larger current account deficit will require higher interest rates
to attract the foreign capital needed to finance it. Conversely,
a higher interest rate, caused by tighter monetary policy or other
domestic factors, will attract capital, boosting the exchange
rate, and reducing net exports and the current account balance.
Vector autoregression is a useful technique for forecasting where
each variable in a group helps to forecast the others.
In addition, these variables are nonstationary, and this
violates a key assumption of the OLS model. The interest rate
differential, unemployment rate and the current account ratio
are nonstationary variables, as seen from the numbers in Table
1. These three variables have unit roots. An OLS specification
with nonstationary data may appear to show a statistically significant
relationship which is actually a spurious correlation. A more
reliable estimate of the impact can be obtained with a vector
error correction, which can deal with the interdependence of cointegrated
variables.
3. Vector Error Correction Estimates Significant
cointegration was found among the interest rate, current account,
and unemployment rate, and between just the interest rate and
the current account. In some VEC's the unemployment rate was
not used, resulting in slightly smaller impacts of the current
account innovation on the interest differential.
The Canadian inflation rate and the Canadian T-bill rate
are not cointegrated, according to the Jorgenson test. Even the
Canada-U.S. inflation differential and the T-bill differential
are not cointegrated over the shorter sample from 1983 to 1996,
although they are cointegrated over the longer sample. The inflation
differential was included as an endogenous variable in some VEC's,
and this produced the largest long-run impact of the current account
on the T-bill differential. In other VEC's, only a distant lagged
value of the inflation differential was included. Taking a cue
from the OLS result, the 3 year earlier lagged value of the inflation
differential was used as an exogenous variable.
Two U.S. variables were included as exogenous variables:
the current account balance and the unemployment rate. These
serve as proxies for changes in the world supply curve of capital
to Canada. The fluctuations in the Canadian current account balance
can be viewed as the demand curve which shifts relative to this
world supply curve.
VEC's were also estimated for longer and shorter sample periods,
from 1963 to 1996 and 1983 to 1996, to check for sensitivity.
The cointegrating equation remained significant over the alternative
sample periods. The coefficient on the current account balance
from the longer sample period from 1963 to 1996 was not too different
from the 1977 to 1996 sample, with a value of 0.61 (and a t-statistic
of 7.1). By contrast, the coefficient on the current account
balance jumps substantially in the shorter sample period, to 1.17
(compared to 0.7 in the 1977 to 1996 period). It is likely that
the short sample period is strongly influenced by the rapid, dramatic
deterioration in the current account deficit in the late 1980s.
This may indicate an increasing inelasticity of capital inflows
as the deficit gets especially large. Given the likelihood of
asymmetric responses, one should not extrapolate from this type
of estimate to predict how low interest rates could drop if Canada
achieves a large surplus in the future.
In the two VEC's summarized in Tables 3 and 4, the lags on
the endogenous variables are from t-3 to t-6, as this was found
to be the range which maximized the likelihood ratio in the Jorgenson
cointegration tests. Lags from t-1 to t-6 were also tried, and
this produced slightly smaller impacts of the current account
on the T-bill differential.
4. Predicted Impacts of Changes in the Current Account Balance
The estimated VECs can be shocked with an exogenous change
in the value of the current account balance. This sets off a
chain reaction in which the current account balance and interest
rates mutually change each other's values, since both are endogenous
to the model. The implied impact from the VECs is generally larger
than from the OLS model for this reason. A lower current account
deficit leads to lower interest rates. The VEC model, appropriately,
reflects the virtuous circle aspect of this, which the OLS model
cannot do. It tells us, correctly, that lower interest rates
will lead to a lower current account deficit. This, in turn,
will allow a further reduction in interest rates.
Figure 2 shows the impact through time of a 1 percentage
point reduction of the current account deficit as a percentage
of GDP. It contrasts two VEC models with the results of the OLS
regression shown above.
VEC no. 1, which incorporates the unemployment rate, implies
that the mature impact (which occurs about 22 quarters after the
beginning of the shock) is a large 80 basis point reduction in
the Canadian three month T-bill rate. The peak impact occurs
after 12 quarters, and exceeds 90 basis points. This perhaps
reflects the monetary policy response to the drop in the current
account balance. It is noteworthy that the VEC which excludes
the unemployment rate shows a smaller impact. It is possible
that part of the initial reaction to a current account deficit
reduction is a cyclical one, since declines in Canada's deficit
often occur due to a growth slowdown in which Canada's imports
drop more than its exports. This would prompt an easing of monetary
policy, but in such a situation the drop in the current account
balance would be an indicator of cyclical conditions rather than
the actual cause of the decline in interest rates. Therefore,
it is possible that the result from VEC no. 1 exaggerates the
true impact, and it should be viewed as the upper limit of a potential
range of impacts.
The same model was tried using the differential between long-
term bond yields instead of the three month T-bill rate. The
likelihood ratio of the cointegrating equation declined considerably,
although it remained significant. The size of the impact was
about one-sixth as large as on the T-bill differential. This is
consistent with findings by Spiro (1994), Laidler and Robson (1995)
and Christiansen and Pigott (1997) on the relationship between
movements in short-term and long-term interest rates.
The cointegrating equations also indicated that there is
a substantial reverse impact, from changes in interest rates on
the current account ratio itself. A one percentage point increase
in the Canadian T-bill rate increases the current account deficit
to GDP ratio by about 1.25 percentage points. Tight monetary
policy in the 1987-93 period was a substantial contributor to
the increase in Canada's current account deficit.
It should be cautioned that the particular impacts estimated
here reflect the international capital market's demand curve for
Canadian debt as it existed over the historical time period covered
by the sample. We cannot state exactly what factors were important
in determining the degree of elasticity of demand over this time
period. It may have been primarily the response of risk averse
investors who prefer to be more diversified, and therefore insist
on a premium if they are asked to buy an unusually large amount
of a particular country's debt. There may be more complex factors,
specific to Canada, such as a perceived risk of a deteriorating
exchange rate in response to a high current account deficit.
The market's "taste" for Canadian debt may change,
either positively or negatively, in response to political or other
factors. For example, the proposed European Monetary Union may
lead to a marked change in the attitude of European investors
toward non-EMU debt. What we can say is that, over the past few
decades, there has been a consistent statistically significant
relationship, indicating that Canadian interest rates need to
rise substantially when Canadians taken together increase their
borrowing in foreign markets.
Conclusions
The analysis above suggests that large current account deficits
in the early 1990s contributed to significantly higher interest
rates in Canada. These current account deficits were the result
of a combination of large fiscal deficits and a very tight monetary
policy. Both of these adverse policy factors have been reversed
in the past two years, resulting in the current account deficit
being essentially eliminated in 1996. Many forecasters are predicting
a growing current account surplus for Canada in future years.
If this proves to be correct, it is likely that Canadian real
interest rates will remain considerably lower than they were in
the 1980s and early 1990s.
Data Used
DIF3 is the difference between the Canadian and U.S. three
month T-bill rates, Cansim B14007 and B54409.
CURR is the Canadian current account balance as a percent of GDP,
Cansim D72002 and D20011.
CURRUS is the U.S. current account balance as a percent of
GDP, Citibase BPCR and Cansim D51230.
DRP is the difference between the Canadian and U.S. year-
over-year percent change in CPI inflation, Cansim P700000
and D139105.
DUR is the difference between the Canadian and U.S.
unemployment rates, Cansim D767611 and B53106
References
Christiansen, Hans and Charles Pigott (1997), "Long-Term
Interest Rates in Globalized Markets." OECD Economics Department
Working Paper no. 175.
Darrat, Ali and M.O. Suliman (1991), "Have budget deficits
and money growth caused changes in interest rates and exchange
rates in Canada?" North American Review of Economics and
Finance 2, 69-82.
Evans, Paul (1985), "Do large deficits produce high interest
rates?" American Economic Review 75, no. 1 (March), 68-87.
Fillion, Jean Francois (1996), "L'endettement du Canada et
ses effets sur les taux d'interet reels de long terme."
Ottawa, Bank of Canada Working Paper 96-14.
Ford, Robert and Douglas Laxton (1995), "World Public Debt
and Real Interest Rates." Washington, International Monetary
Fund Working Paper 95/30.
Johnson, David R. (1994), "Ricardian equivalence: Assessing
the evidence for Canada." In W. Robson. and W. Scarth, Deficit
Reduction: What Gain, What Pain? (Toronto: C.D. Howe Research
Institute).
Laidler, David, and William Robson (1995), "Don't Break the
Bank! The Role of Monetary Policy in Deficit Reduction."
C.D. Howe Institute Commentary, no. 66.
Orr, Adrian, Malcolm Edey and Michael Kennedy (1995), "Real
long- term interest rates: the evidence from pooled time series."
OECD Economic Studies no. 25.
Poloz, Stephen S. (1992), "Fiscal policy and external balance
in the G7 countries." Ottawa, Bank of Canada Technical Paper
no. 60.
Siklos, Pierre L. (1988), "The deficit-interest rate link:
evidence for Canada." Applied Economics 20, 1563-1577.
Spiro, Peter S. (1990), "The effect of government debt on
short- term real interest rates." International Monetary
Fund Staff Papers 37 no. 4 (December).
Spiro, Peter S. (1994), "The differential between Canadian
and U.S. long-term bond yields." Canadian Business Economics
2, Winter, 16- 24.
Strauss, Jack and Dek Terrell (1995), "Cointegration tests
of the Fisher Hypothesis with variable trends in the world real
interest rate." Southern Economic Journal 61 no. 4 (April),
1047-1056.
Table 1. Unit Root Test ADF Statistics
Level First Difference
current account ratio -1.79 -5.64
interest rate differential -2.24 -5.73
exchange rate -2.96 -2.76
Cdn. T-bill rate -1.32 -4.45
Cdn. CPI inflation rate -1.15 -4.95
unemployment rate -2.54 -3.74
Sample from 1977.1 to 1996.4. The hypothesis of a unit root
cannot be rejected for the level of any variable at 5% confidence.
It can be rejected for the first difference at 1% confidence
for all the variables except the exchange rate, where it cannot
be rejected even at 5%.
Table 2. Tests for Cointegration among Interest Rate Differential,
Current Account Ratio, and Unemployment Rate
Sample Period Jorgenson Test with trend, lags from t-3 to t-6
ADF test with trend, 1 lag
None At most one
1977.1 to 1996.4 47.6*** 11.2 -3.59*
1963.1 to 1996.4 53.0*** 9.1 -4.15**
1963.1 to 1984.4 47.2*** 8.5 -4.04**
Hypothesis of no cointegration rejected at: 1% level (***), 5%
level (**) or 10% level (*)
Table 3. Summary Statistics of VECM No. 1
Sample: 1977:1 1996:4 t-statistics in parentheses
Cointegrating Eq: CointEq1 DIF3(-1)
1.000000 CURR(-1) 0.700729
(6.04011) UR(-1) 0.165385
(1.29753) C -2.091252
Error Correction: D(DIF3) D(CURR) D(UR)
CointEq1 -0.513346 -0.034814 0.097749
(-4.95825) (-0.40163) (2.78317)
D(CURR(-3)) 0.014219 -0.034403 -0.063461
(0.09299) (-0.26874) (-1.22349)
D(CURR(-4)) -0.163730 -0.196143 -0.117731
(-1.09466) (-1.56630) (-2.32034) D(CURR(-5))
-0.330852 -0.097605 -0.159170 (-2.16212)
(-0.76185) (-3.06632) D(CURR(-6)) -0.388194
0.108387 -0.109573 (-2.51995) (0.84037)
(-2.09680) R-squared 0.436178 0.233507
0.558669 Adj. R-squared 0.292986 0.038842 0.446585 Log
likelihood -98.23911 -84.02724 -11.75199 Akaike AIC
0.043108 -0.312189 -2.119070 Schwarz SC 0.549288 0.193992
-1.612890 Mean dependent -0.077083 0.014511 0.029822 S.D.
dependent 1.107220 0.795058 0.424533 Determinant
Residual Covariance 0.021688 Log Likelihood
-67.30490 Akaike Information Criteria
-2.481009 Schwarz Criteria
-0.873141
Table 4. Summary of VECM No. 2 (Excludes Unemployment Rate)
Sample: 1977:1 1996:4 t-statistics in parentheses
Cointegrating Eq: CointEq1 DIF3(-1) 1.000000
CURR(-1) 0.542429 (4.43273)
C -0.883309 Error Correction:
D(DIF3) D(CURR) CointEq1 -0.509682 -0.028947
(-5.08463) (-0.34231) D(CURR(-3))
-0.048640 -0.022131 (-0.32725) (-0.17650)
D(CURR(-4)) -0.146074 -0.230663 (-1.02040)
(-1.90997) D(CURR(-5)) -0.346265 -0.155412
(-2.36543) (-1.25846) D(CURR(-6))
-0.348477 0.056174 (-2.34006) (0.44714)
R-squared 0.393292 0.162583 Adj. R-squared
0.295148 0.027118 Log likelihood -101.1715 -87.56711
Akaike AIC -0.008582 -0.348692 Schwarz SC
0.348722 0.008612 Mean dependent -0.077083 0.014511
S.D. dependent 1.107220 0.795058 Determinant
Residual Covariance 0.349670 Log Likelihood
-104.9995 Akaike Information Criteria -0.400767 Schwarz
Criteria 0.373392