Page 1
Jack Adons
In coordination with Charlie Fay and Gideon Ikpegaogu
Professor Murray
ECON 255
April 7, 2015
TERM PAPER
INTRODUCTION
In a report titled “New Evidence on Deficits and Interest Rates,” Gregory
Hoelscher analyzed the general assumption that increased government borrowing causes
an increase in interest rates. Hoelscher acknowledged that a number of previously
published papers failed to provide evidence of a strong relationship between government
deficits and short-term interest rates. His research, however, dealt with the connection
between deficits and long-term interest rates. He uncovered strong evidence to support an
association between high deficits and high long-term interest rates. Hoelscher used data
from the period of 1953 to 1984. In order to better observe the relationship between
deficits and interest rates, the periods of data needed to be extended beyond that included
in Hoelscher’s report. In this paper, Hoelscher’s study is replicated as closely as possible
to include data from 1953 to 2013. Consequently, Hoelscher’s evidence was confirmed,
and further investigation was conducted to better understand the connection between
government borrowing and interest rates.
1
Page 2
The association between federal borrowing and interest rates has a crucial impact
on the government’s fiscal decisions on private spending and changes to aspects of the
deficits. The correlation between deficits and long-term interest rates concerns the
question of crowding out. Short-run indirect crowding out, an issue Hoelscher addressed,
is defined as a reduction in private spending caused by a rise in interest rates as a result of
an increase in federal borrowing and spending. When the government spending occurs in
a market of loanable funds, the demand for such funds rises, increasing the interest rates
as well. As a result, buyers would be driven away from private investment. A decrease in
investment would lower aggregate demand as well as the output of the economy.
Therefore, it is critical to analyze the affect deficits have on crowding out, as Hoelscher
does in his report.
However, the chief focus of the paper is on the link between the deficits and long-
term rates. Previous studies as noted in Hoelscher conveyed a fiscal and wealth impact of
bond-financed deficits on output and interest, with both effects contributing to higher
interest rates. The long-term rates are closely related to consumer and business spending
decisions. It was generally theorized by scholars that the long-term interest rate transmits
the effect of deficits to the real side of the economy. This is due to the fact that interest-
sensitive components of private spending were most sensitive to variability in long-term
interest rates.
Data used in this report was gathered from the DRI database and contained a
number of economically relevant variables. The EViews program was utilized to search
for variables that would represent best, the variables Hoelscher used in his analyses. In
the instance that the exact Hoelscher variable did not appear in the database,
2
Page 3
combinations of variables were utilized to reconstruct necessary variables. The variables
that were unavailable in the DRI database, such as expected rate of inflation, were
provided by outside sources.
THE VARIABLES
The research Hoelscher conducted utilized a number of variables for regression.
The first task of the assignment required the reconstruction of such variables using the
DRI database. Some of the variables did not exist directly in the database, so it was
necessary to combine variables to correspond to those appearing in Hoelscher’s
regressions. These variables included the expected inflation rate, p; the expected real
short-term interest rate, rs; the change in national income, y; a deficit variable, d;
government spending, GOVSP; transfer payments, TP; and the nominal one-year
Treasury bond rate, is.
In his report, Hoelscher described in detail first the deficit variables he used
represented by the letter d. Hoelscher noted in his paper that there was disagreement
about the proper measure of the government deficit. There was dispute concerning the
relation between inflation and real value of outstanding government debt. Another cause
of controversy was the uncertainty as to how to measure the impact of federal loan
guarantees and unfounded liabilities on the credit markets.
All of the measures of the deficit were expressed in per capita 1972 dollars. The
first variable for the deficit that Hoelscher used was denoted as USDEF. This was defined
as the national income accounts version of the federal deficit. In the DRI database the
3
Page 4
variable ggfnet was discovered to be the national accounts version of the government
deficit. Because USDEF was measured as a per capita real value, it was necessary to
divide by the population, labeled gpop. Since ggfnet and gpog were measured in billions
and millions respectively the ratio was multiplied by 1000. Finally, it was multiplied by
the CPI-U from 1972 over the CPI-U from the corresponding year t, denoted pzunew in
the DRI database. This resulted in the following equation:
USDEF = [(ggfnet*1000)/(gpop)] * [(pzunew1972)/(pzunew)]
The second variable used to measure the government deficit was labeled
GOVDEF. Similar to USDEF, it was defined as the national income accounts version of
the federal deficit, however, GOVDEF also included borrowing from state and local
governments. In the DRI database the variable ggnet was discovered to be the sum of
both the state and local government deficits in addition to the federal deficit. Because
GOVDEF was measured as a per capita real value, it was necessary to divide by the
population, labeled gpop. Since ggnet and gpog were measured in billions and millions
respectively the ratio was multiplied by 1000. Finally, it was multiplied by the CPI-U
from 1972 over the CPI-U from the corresponding year t, denoted pzunew in the DRI
database. This resulted in the following equation:
GOVDEF = [(ggnet*1000)/(gpop)] * [(pzunew1972)/(pzunew)]
4
Page 5
The final variable Hoelscher utilized to measure the federal deficit was expressed
as CFD or the change in real par value of publically held federal debt. In the DRI
database the variable fbdp was discovered to be the federal debt securities held by the
public. Because CFD was measured as a per capita real value, it was necessary to divide
by the population, labeled gpop. Since fbdp and gpog were measured in billions and
millions respectively the ratio was multiplied by 1000. Finally, it was multiplied by the
CPI-U from 1972 over the CPI-U from the corresponding year t, denoted pzunew in the
DRI database. However, in order to achieve the change in real par value of publicly held
federal debts it was key to take the difference of the fbdp in per capita 1972 dollars from
one year and the fbdp in per capita 1972 dollars from the year previous. This resulted in
the following equation:
CFD = ([(fbdp*1000)/(gpop)] * [(pzunew1972)/(pzunew)]t) – ([(fbdp*1000)/(gpop)]
* [(pzunew1972)/(pzunew)]t-1)
The loanable funds model included a number of variables. Hoelscher denoted the
expected inflation rate as p. The DRI database did not include a variable that
corresponded well enough with Hoelscher’s version of p. Therefore, the variable
representative of p was obtained from the Livingston Survey.
Hoelscher labeled the expected real short-term interest rate as rs. It was defined as
the annual average rate on one-year Treasury bonds minus the expected rate of inflation.
In the DRI database the variable fygt1 was discovered to be the average annual rate on a
ten-year treasury bond. However, in order to obtain the expected short-term interest rate
5
Page 6
in real terms it was necessary to subtract the expected rate of inflation, p. The long-term
interest rate denoted, i (or fygt10 in the DRI database), was defined as the annual average
rate on ten-year Treasury bonds.
In addition, Hoelscher defined y as the change in national income or the change in
per capita GNP. In the DRI database the variable gnpq was discovered to be the gross
national product. Because y was measured as a per capita real value, it was necessary to
divide by the population, labeled gpop. Since gndp and gpog were measured in billions
and millions respectively the ratio was multiplied by 1000. Finally, it was multiplied by
the CPI-U from 1972 over the CPI-U from the corresponding year t, denoted pzunew in
the DRI database. Since y is the change in per capita GNP, the variable was expressed as
the following:
y = ([(gnpq*1000)/(gpop)] * [(pzunew1972)/(pzunew)])t - ([(gnpq*1000)/(gpop)] *
[(pzunew1972)/(pzunew)])t-1
The government spending from the National Income Accounts for all government
levels was represented in Hoelscher by GOVSP. In the DRI database the variable ggex
was discovered to be the federal spending in all levels of government. Because GOVSP
was measured as a per capita real value, it was necessary to divide by the population,
labeled gpop. Since ggex and gpog were measured in billions and millions respectively
the ratio was multiplied by 1000. Finally, it was multiplied by the CPI-U from 1972 over
the CPI-U from the corresponding year t, denoted pzunew in the DRI database. The result
was the following equation:
6
Page 7
GOVSP = [(ggex*1000)/(gpop)] * [(pzunew1972)/(pzunew)]
The transfer payments from the National Income Accounts for all government
levels were symbolized in Hoelscher as TR. In the DRI database the variable ggt was
discovered to be the federal spending in all levels of government. Because TR was
measured as a per capita real value, it was necessary to divide by the population, labeled
gpop. Since ggt and gpog were measured in billions and millions respectively the ratio
was multiplied by 1000. Finally, it was multiplied by the CPI-U from 1972 over the CPI-
U from the corresponding year t, denoted pzunew in the DRI database. The result was the
following equation:
TR = [(ggt*1000)/(gpop)] * [(pzunew1972)/(pzunew)]
In Hoelscher the deficits are at times reported as percentages of the gross national
product. Consequently, each DRI database variable representative of the deficit (i.e.
ggfnet for USDEF, etc.) was divided by the GNP, and multiplied by 100 because it was
measured as a percentage. In the regressions these are labeled usdefg, govdefg and cfdg
respectively.
THE MODEL
7
Page 8
In his analyses, Hoelscher utilizes a loanable funds model to study the
relationship between deficits and long-term interest rates, as adapted from models used
by Sargent (1969) and Echols and Elliot (1976). The use of this model enabled Hoelscher
to analyze a variety of explanatory variables and observe their effects on long-term
interest rates. It originated from the need for economists to study interest rates instead of
the price of bonds. Therefore, this model is similar to a supply and demand diagram for
loanable funds in the economy where the x-axis is occupied by the quantity of bonds and
the y-axis is occupied by the interest rate. The point of intersection on the diagram, or
where the supply of loanable funds equals the demand for loanable funds, indicates the
interest rate. As a result, the equation (1) in Hoelscher is explained:
S(i, rs, p) – D(i, rs, p, y, d) = 0.
In this equation, the supply curve is determined by i, rs, and p, or 10-year
Treasury bond rates, expected real short-term interest rate, and the expected inflation rate.
8
Page 9
The demand curve is determined by the same components i, rs, p, in addition to y and d,
or the change in per capita GNP and some measure of the deficit. Hoelscher used the
equilibrium equation to solve for the long-term interest rate, denoted i, which produced a
reduced form function. An equation involving long-term interest rates equaled to the
outstanding variables included in the supply and demand functions apart of the
equilibrium equation can be observed:
i = α0 + α1p + α2rs + α3y + α4d.
As a result, Hoelscher was able to regress long-term interest rates on the
explanatory variables and display them in tables. These regressions determined the values
of the coefficients α0, α1, α2, α3, and α4, or the impact each independent variable had on
the dependent variable, 10-year Treasury bond rates. The correlation between long-term
interest rates and the variables could then be evaluated. In all of the regressions that
Hoelscher ran, he employed the Cochrane-Orcutt procedure to adjust for serial
correlation. Serial correlation is defined as a similarity between variables due to a time
lag between them.
The observed original Durbin-Watson statistic (1.614) signified that there did not
appear to be any serial correlation because its value was less than that of the critical value
(1.739). However, the concept of serial correlation could not be dismissed due to the
first-order autoregressive nature of the data. In addition, the t-statistic for rho that was
found in Table 1 to be .83, which was less than the critical value (1.96) at the 99 percent
significance level. Normally, this would indicate that the null hypothesis should not be
9
Page 10
rejected, however this could be inaccurate. The use of this estimation procedure
demonstrated that Hoelscher was concerned with similarities of observations within
variables from year to year.
THE REPLICATED TESTS
Replicated Table 1: Corc Estimates Of Ten-Year Bond Yield Determinants, 1953-84.
Coefficient Estimates
Deficit variable Constant p rs y d R2 D-W rho
(t-stat)
USDEF 1.2 0.745 0.785 -0.000841 0.00807 0.978 1.61 0.161
(0.83)
GOVDEF 1.29 0.762 0.788 -0.000276 0.00744 0.9820 1.742 0.697
CFD 1.47 0.870 0.811 -0.000932 0.00691 0.9857 1.70 0.0674
USDEF --- 0.533 0.6868 -0.0010074 0.00386 0.8173 --- ---
Table 1 in Hoelscher described the regression results for estimating the long-term
interest rate equation using data from 1953-1984. The first three tests were run using a
Cochrane-Orcutt Two-Step procedure that adjusted for serial correlation. The ten-year
Treasury bond yield was regressed on the expected inflation, the expected real short-term
interest rate, the change in per capita GNP, and three separate deficit variables. The
10
Page 11
fourth line of Table 1 utilized the first difference estimation method with no constant. In
his report, Hoelscher cited that other studies recommended that time series regression
models be constructed with both levels and changes in levels of variables as a precaution
against spurious regression. Line 4 in Table 1 utilized this method to eliminate correlation
between a variable for a given year and the same variable from the year before.
The replicated version of Hoelscher’s Table 1 is observed above. The Cochrane-
Orcutt Two-Step procedure was again utilized to approximate the best linear unbiased
estimator of the model with first-order autoregressive disturbances. In order to replicate
Table 1 line 4 lag variables of one year were generated for each variable in the USDEF
regression. This lag variable was then included in generated variables that expressed the
change in one year from the previous year for each variable in the USDEF regression.
The coefficients produced in the replicated Table 1 closely resembled those
reported in Hoelscher, however, were not exact due to differences in data. Nonetheless, it
was possible to imply the same relationship between the explanatory variables and the
long-term interest rate as in Hoelscher. The estimated coefficients in all three regressions
had expected signs, and were all statistically significant at the at the 99 percent level,
except for the term y. This was the case because the p-value for each significant variable
was below .01. It was concluded that all the regressions fit the data very well and almost
all of the variation in ten-year Treasury bond rate had been explained. This was
determined because the values for R2 were .85 or greater, signifying a good fit.
Replicated Table 2: Additional Corc Estimates Of Ten-Year Treasury Bond Yield
Determinants, 1953-1984.
11
Page 12
Coefficient Estimates
Deficit variable Constant p rs y d R2 D-W rho
USDEF/GNP 1.05 0.785 0.804 -0.000584 0.456 0.973 1.60 0.149
CDF/GNP 1.22 0.888 0.828 -0.000792 0.397 0.979 1.57 0.151
It is often the case that deficits are recorded as a percentage of GNP as opposed to
a per capita basis. As a result, the regressions in Table 1 lines 1 and 3 were reconstructed.
The respective deficit variables were rescaled to be a percentage of GNP. Hoelscher
again regressed ten-year Treasury bond yield on the expected inflation, the expected real
short-term interest rate, the change in per capita GNP, and the rescaled deficit variables
and reported the results in Table 2 lines 2 and 4.
The replicated version of Table 2 lines 2 and 4 for years 1953-1984 was carried
out in similar fashion, but with different data. All three of the deficit variables maintained
positive and significant coefficients like Table 2 of Hoelscher. This indicated that deficits
are important regardless of the scaling mechanism used (population or GNP). The other
terms of the regression also maintained similar values to that in Hoelscher. However, the
sign for y became negative compared to the regressions ran by Hoelscher. Nonetheless,
the change and resulting effect on interest rates was small and insignificant.
Replicated Table 3: Corc Estimate Of Ten-Year Treasury Bond Yield Determinants With
Additional Regressors, 1953-1984.
Coefficient Estimates
12
Page 13
Deficit variable Constant p rs y d GOVSP TR R2 D-W rho
GOVDEF 0.803 0.521 0.791 -0.000578 0.00424 -0.000538 .0061604 0.985 1.67 0.103
It was noted in Hoelscher’s paper that some studies expressed concern about the
correlation between government deficits and government spending or transfer payments.
If this did exist, the coefficient on the deficit variables may have factored in the effects of
government spending and transfer payments and result in false estimates. Consequently,
he regressed long-term interest rates on previously used variables, with the addition of
GOVSP and TR variables. In Table 3 line 1, Hoelscher used the GOVDEF as the variable
for the deficit.
Coefficients listed in the replicated Table 3 line 1 were comparable to that of
Hoelscher’s. The replicated Table 3 line 1 and the Hoelscher Table 3 line 1 both indicated
that GOVSP was not significant, but the coefficient on TR was positive and significant.
However, GOVDEF also had a positive and significant effect on the long-term interest
rate. Therefore, Hoelscher was able to prove that the government deficit was not in
correlation with government spending or transfer payments, so it was likely that the
added variables did not change the significance of the deficit’s effect on long-term
interest rate.
Replicated Table 4: Corc Estimate Of Ten-Year Treasury Bond Yield Determinants For
Subperiods.
Coefficient Estimates
13
Page 14
Period Constant P rs y d R2 D-W rho
1953-66 1.93 0.685 0.534 -0.000863 0.00379 0.904 1.61 0.187
1967-84 1.63 0.680 0.808 -0.000753 0.00770 0.958 1.55 0.186
In Table 4 the long-term interest rate was regressed using USDEF for two sub
periods in order to test for temporal stability, or retesting. It was important to test this to
see if it explained variation in the ten-year Treasury bond rate for each sub period.
Hoelscher determined the cutoff for the sub periods by noting that after 1966, ten-year
Treasury bond rates rose above five percent and continued on a gradual increase for the
rest of the time period.
The replicated Table 4 displayed estimated coefficients and R2 values comparable
to that of Hoelscher. The USDEF measure is only used because the other two measures of
the deficit estimated similar results. In both tables USDEF exerted a positive and
significant effect on the long-term interest rate despite a lower deficit in the earlier
period. As Hoelscher expected, the coefficient estimates for the independent variables
varied more in the latter sub period. In fact, the coefficient on USDEF doubled.
Replicated Table 5: Illustration Of Estimated Effects Of Deficits On Long-Term Rate.
Variable Average (1980-84) Estimated Effect on i (basis point increase)
USDEF 249.8 202
14
Page 15
GOVDEF 240.0 179
CFD 117.3 81
Hoelscher created Table 5 lines 1-3 to add perspective on the varying estimated
magnitude of the deficit coefficient. He summarized the estimated effect that the average
annual deficit over the last five years of the period would have on the ten-year Treasury
bond rate. The deficits were averaged over a five-year period to smooth transitory effects
and measured in per capita 1972 dollars.
The deficit variables were summarized to observe the mean for each. This is
represented in the first column of replicated Table 5. In order to find the estimated effects
on I each average value was multiplied by the corresponding estimated coefficient from
Table 1 and then multiplied by 100 to be measured in basis points. According to
Hoelscher’s findings, USDEF was responsible for a 1.95 percent increase in the long-
term interest rate. The average values and resulting basis point increase in the replicated
Table 5 varied in comparison to Hoelscher. The USDEF and CFD values were smaller
than that recorded in Hoelscher, whereas the GOVDEF values were larger.
THE UPDATED TESTS
Further tests were run in order to re-estimate the relationships of variables using
data from 1953 to 2013. The regression commands for the 1953-1984 data were repeated
for the updated data, but for the period 1953-2013. In the process of expanding the data
through the year 2013 it was necessary to decide the best procedure for estimation. The
15
Page 16
options provided were OLS with ordinary standard errors, OLS with Newey-West
standard errors, or Cochrane-Orcutt. Because serial correlation was present in each of the
earlier regressions and therefore would no longer be BLUE, OLS with ordinary standard
errors was ruled out. The Newey-West procedure estimated coefficients with OLS with
heteroskedasticity and first-order auto regression robust standard errors. The Cochrane-
Orcutt procedure for iterated estimates is only robust for first-order auto regression
standard errors. However, both procedures produced similar coefficients when tested for
Table 1 line 1.
This indicated that both procedures would be appropriate to utilize because both
adjusted for the presence of serial correlation. The Cochrane-Orcutt for iterated estimates
or Cochrane-Orcutt One-Step was chosen as the procedure for estimation. The observed
original Durbin-Watson statistic (.89) signified that there did not appear to be any serial
correlation because its value was less than that of the critical value (1.727). However, the
concept of serial correlation could not be dismissed due to the first-order autoregressive
nature of the data. This was concluded by the fact that the transformed Durbin-Watson
statistic (1.964) was greater than the critical value, and therefore the null hypothesis of no
serial correlation was rejected. In addition, the t-statistic for rho that was found in Table 1
to be 4.98, which was greater than the critical value (2.00) at the 99 percent significance
level. Consequently, this would indicate that the null hypothesis of zero serial correlation
should be rejected. The Newey-West standard errors were consistent, but bias because the
standard errors were changed. Thus, the Cochrane-Orcutt procedure for iterated estimates
was the best linear unbiased estimator. Though this method discards the first observation
16
Page 17
compared to the Prais-Winsten procedure, it was utilized in order to remain as close to
Hoelscher as possible. The updated versions for each table are listed below.
Updated Table 1: Corc Estimates Of Ten-Year Bond Yield Determinants, 1953-2013.
Coefficient Estimates
Deficit variable Constant p rs y d R2 D-W rho
(t-sta)
USDEF 2.7 0.626 0.597 -0.000283 0.00144 0.698 0.896 0.872
(4.98)
GOVDEF 2.68 0.627 0.604 -0.000228 0.00137 0.698 0.929 0.880
CFD 2.75 0.667 0.694 -0.000274 0.00121 0.702 0.907 0.861
USDEF --- 0.585 0.575 -0.000284 0.00133 0.672 --- ---
Updated Table 2: Additional Corc Estimates Of Ten-Year Treasury Bond Yield
Determinants, 1953-2013.
Coefficient Estimates
Deficit variable Constant p rs y d R2 D-W rho
USDEF/GNP 2.60 0.646 0.611 -0.000355 0.133 0.713 0.926 0.851
CDF/GNP 2.42 0.732 0.632 -0.000337 0.126 0.733 0.981 0.816
17
Page 18
Updated Table 3: Corc Estimate Of Ten-Year Treasury Bond Yield Determinants With
Additional Regressors, 1953-2013.
Deficit variable Constant p rs y d GOVSP TR R2 D-W rho
GOVDEF 7.72 0.609 0.618 -0.000132 0.00191 -0.00280 .002621 0.715 0.919 0.874
Coefficient Estimates
Updated Table 4: Corc Estimate Of Ten-Year Treasury Bond Yield Determinants For
Subperiods.
Coefficient Estimates
Period Constant P rs y d R2 D-W rho
1953-66 2.92 0.469 0.438 -0.000454 0.00295 0.965 1.61 0.899
1967-84 1.76 0.672 0.815 -0.000814 0.00726 0.943 1.55 0.332
1985-1998 2.18 0.877 0.448 0.000566 0.00398 0.968 2.45 -0.337
1999-2013 9.85 -2.23 0.473 -0.00296 -0.00288 0.952 2.05 -0.462
Updated Table 5: Illustration Of Estimated Effects Of Deficits On Long-Term Rate.
Variable Average Estimated Effect on i (basis point increase)
USDEF 697.4 100
18
Page 19
GOVDEF 839.6 115
CFD 639.4 77.27
In order to evaluate the updated results, it was necessary to observe the
coefficients on the deficit variable in the regressions. In addition, it was sensible to
compare the replicated and updated Table 5 to see the effect of deficits in the final years
of the data between both versions. Finally, the R2 values were discerned with expanded
years relative to Hoelscher.
The coefficient of an independent variable infers the extent and method to which
the independent variable relates to the dependent variable. The independent deficit
variables included were USDEF, GOVDEF, CFD, USDEF / GNP, and CFD / GNP and
were regressed with a dependent variable i. Paralleled with Hoelscher, the deficit
variables were consistently positive and significant as indicated by their p-values, except
for USDEF in Table 4 for sub period 1999-2013. Because the data was extended to 2013,
two more sub periods had to be created. The positive relationship between government
deficits and long-term interest rates is indicated nonetheless. This meant that government
borrowing increased as the ten-year Treasury bond rate went up. The negative coefficient
of USDEF in Table 4 was most likely explained by the presence of enormous debt in the
economic recession during the sub period 2009 to 2013.
The difference between the Hoelscher and updated Table 5 was another indicator
of how the deficit variable affected the long-term interest rates. In addition, changes
beginning from the years of Hoelscher to the years of the updated data could be viewed.
The average value and estimated effects on i in basis point increase varied from that in
19
Page 20
Hoelscher. The average values for each deficit variable in the 2009-2013 period were
sizably larger than those in the 1980-1984 period. The estimated effects on i in terms of
basis point increase also differed from Hoelscher’s results. Unlike the average value, the
estimated effects on i were significantly smaller than that given by Hoelscher. The huge
increase in government deficits since the period used in Hoelscher produced decreased
deficit coefficients. However, the coefficients remained significant.
The last signal of difference between Hoelscher’s tables and the expanded tables
was the R2 value. The values corresponding to regressions for 1953 to 2013 were smaller
than those estimated by Hoelscher for 1953-1984. The R2 values for 1953 to 2013 were
large enough to conclude that the regression was a good fit of the data, however, not as
well as Hoelscher. This was due to a greater variability and range in values for the deficit,
as they have risen substantially since 1984. Therefore, the replicated and updated data
confirmed Hoelscher’s analyses that long-term interest rates increase as the government
deficit increases.
THE ADDITIONAL TESTS
In order to grasp the relationship between government deficits and long-term
interest rates further, additional tests were run. In order to better understand the
movement of long-term interest rates, variables were substituted and added into
regressions previously tested. This was similar to the procedure Hoelscher took in
creating Table 3 line 1. The condition of the U.S. economy has a major influence on
long-term interest rates. As the economy grows, consumers have jobs and savings to lend
20
Page 21
through banks; however, they must borrow for large capital items, such as a home. The
rise of interest rates performs as a ration for funds available, as the resulting increase in
demand for funds occurs. The opposite is true for a decrease in demand for funds. As a
result, data was collected from the Federal Reserve Bank of St. Louis FRED database on
the All-Transactions House Price Index for the United States for the years 1975 through
2013. This data was estimated using sales prices and appraisal data. Data did not exist
previous to the year 1975. The variable that represented this data was named house and
was measured with 1975 as the base year.
In addition to the condition of the economy, the dollar is a highly influential
factor of long-term interest rates. As the main currency in international trade and foreign
exchange markets, orderly fluctuations of the dollar are essential for stability. Major or
volatile exchange rate movements could force the United States Federal Reserve to affect
long-term interest rates and other monetary policy. As a result, data was collected from
the Federal Reserve Bank of St. Louis FRED database on the Real Trade Weighted U.S.
Dollar Index against multiple major currencies that circulate widely outside the United
States. Major currencies index included the Euro Zone, Canada, Japan, United Kingdom,
Switzerland, Australia, and Sweden. Data did not exist previous to the year 1975. The
variable that represented this data was named dollar and was measured with 1975 as the
base year.
The loanable funds model was again utilized to regress ten-year Treasury bond
rates. The dependent variable was regressed on the expected rate of inflation, the real
short-term interest rate, the house price index, the U.S. dollar price index and GOVDEF.
The GOVDEF deficit variable was utilized because it was the broadest measure, although
21
Page 22
the two other measures would presumably provide similar results. Also, it should be
noted that house was substituted in for y. The change in national income is a factor that
expressed the condition of the U.S. economy. However, in all of the previous regressions,
the coefficients on y were found to be insignificant. As a result, the house is utilized as an
alternative measure of the U.S. economy.
The increased variability in the Table 1 for the period 1953 to 2013 compared to
the 1953-1984 period of Hoelscher is expressed by the decreased R2 value. Since 1984,
more determinants of long-term interest rates have existed than just the four independent
variables originally used. This implied that the model utilized in Table 1 for the period
1953 to 2013 experienced some level of omitted variable bias. As a result, the dollar
variable was added to the regression to better explain how long-term interest rates move
in a more developed economy. Even though the R2 value decreased between the periods,
it was still important to add an explanatory variable to see if the absence of this variable
would affect the accuracy of the other coefficients estimated in the model.
Similar to previously run regressions, the Cochrane-Orcutt procedure for iterated
estimates was utilized to adjust for serial correlation. The observed original Durbin-
Watson statistic (1.443) signified that there did not appear to be any serial correlation
because its value was less than that of the critical value (1.786). However, the concept of
serial correlation could not be dismissed due to the first-order autoregressive nature of the
data. This was concluded by the fact that the transformed Durbin-Watson statistic (1.98)
was greater than the critical value, and therefore the null hypothesis of no serial
correlation was rejected. The equation for the expected long-term interest rate when
regressed to substitute house for y, and include dollar is listed below:
22
Page 23
For 1975-2013: i = 1.230549 + .6492327p + .6869269rs – .0046901house
+ .031041dollar + .0017428GOVDEF
It can be observed that the coefficients on house and dollar were both statistically
significant due to very small p-values. The coefficient on house was negative, denoting
an inverse relationship between long-term interest rates and housing prices. Based on the
coefficient, a one-unit increase in the housing price index causes a .00469 percent
decrease in the long-term interest rate. This is a sensible conclusion because as housing
prices increase the demand for homes fall. In order to counter the decrease in demand, the
Federal Reserve would decrease interest rates in order to incentivize saving and
investment.
The coefficient on dollar was positive, denoting a direct relationship between
long-term interest rates and the exchange rate of the U.S. dollar. Based on the coefficient,
a one-unit increase in the housing price index caused a .031 percent increase in the long-
term interest rate. Again this is a sensible conclusion because as the exchange rate for the
U.S. dollar increase, the demand for loanable funds by foreign investment increases. In
order to cope with the increased demand, the Federal Reserve would raise interest rates to
discourage saving in U.S. banks.
The high computed R2 value of .9418 indicted that the substitution of house for y,
and the addition of dollar created a better fitting model than that in Table 1 and Table 3
line 1 for period 1953-2013. This was visible through the increase in the R2 term, and
consequently, the omitted variable bias present was reduced.
23
Page 24
Additional Table 5: Illustration Of Estimated Effects Of New Variables On Long-Term
Rate.
Variable Average Estimated Effect on i (basis point increase)
House 333.141 -156.2465
Dollar 97.3275 302.1143
GOVDEF 185.492
6
32.32765
In addition to regressing long-term interest rates on certain variables, the average
values and estimated effects on i, were discovered for each new variable. The results
were listed in the above table in order to discern the magnitude to which the new
variables effected i. The magnitude of the impact of housing prices and the dollar on the
ten-year Treasury bond rate could be compared to the magnitude of the impact of the
deficit on the ten-year Treasury bond rate. The magnitude of the estimated effect on i for
house and dollar were both greater than the value found for the deficit. As a result, we
can conclude that both housing prices and the exchange rate of the dollar have more of an
impact on long-term interest rates than any measure of the deficit. This was a significant
result to consider when looking back on Hoelscher’s research. It questioned the use of the
deficit in Hoelscher to determine the movements of ten-year Treasury bond rates.
CONCLUSION
24
Page 25
The empirical evidence of Hoelscher was ultimately confirmed by replicated and
updated regressions. Confirmation was given to the theoretical prediction that deficits
cause long-term interest rates to rise. In addition, the regression results indicated that the
relationship between government deficit and ten-year Treasury bond rates were strong,
robust and very significant for given periods.
The additional analyses carried out provided significant results that in turn
questioned the use of deficit in Hoelscher to determine the movement of long-term
interest rates. Because the effect of housing prices and the U.S. dollar exchange rate had
on i was of greater magnitude than the defict, it was concluded that Hoelscher chose not
to include the most impactful variables to regress against the ten-year Treasury bond rate.
However, this could mean that the housing price and exchange rate are more important
determinants today, than they were during the period of Hoelscher’s study. Further
analyses could be done to investigate deeper, the association of housing prices and
exchange rates on long-term interest rates versus the correlation of the deficit on long-
term interest rates.
25