Review of Economic Dynamics 5, 45–72 (2002)
doi:10.1006/redy.2001.0141, available online at http://www.idealibrary.com on
The Great Depression in Canada and
the United States: A Neoclassical Perspective1
Pedro S. Amaral
Department of Economics, University of Minnesota, Minneapolis, Minnesota 55455
E-mail: pamaral@econ.umn.edu
and
James C. MacGee
Department of Economics, University of Minnesota, Minneapolis, Minnesota 55455;
and Research Department, Federal Reserve Bank of Minneapolis,
Minneapolis, Minnesota 55480
E-mail: macgee@econ.umn.edu
Received August 15, 2001
Canada suffered a major depression from 1929 to 1939. In terms of output it
was similar to the Great Depression in the United States. However, total factor
productivity (TFP) in Canada did not recover relative to trend, while in the United
States TFP had recovered by 1937. We find that the neoclassical growth model,
with TFP treated as exogenous, can account for over half of the decline in output
relative to trend in Canada. In contrast, we find that conventional explanations for
the Great Depression—monetary shocks, terms of trade shocks, and labor market
and competition policies—do not work for Canada. Journal of Economic Literature
Classification Numbers: E30, N12, N42. 2002 Elsevier Science
Key Words: Great Depression; Canada; productivity; terms of trade; deflation.
1
We are especially grateful to Tim Kehoe, Ed Prescott, Lee Ohanian, and Hal Cole for
their comments and suggestions. Comments by Ben Bridgman, Ron Leung, Igor Livshits, and
Manuel Santos as well as seminar participants at the SED 2001 meeting in Stockholm and the
Macro Workshop at the University of Minnesota are also much appreciated. Amaral acknowledges financial support from Fundação para a Ciência e Tecnologia. The views expressed
herein are those of the authors and not necessarily those of the Federal Reserve Bank of
Minneapolis or the Federal Reserve System.
45
1094-2025/02 $35.00
2002 Elsevier Science
All rights reserved.
46
amaral and macgee
1. INTRODUCTION
Canada suffered a major depression from 1929 to 1939. In terms of output, it was similar in both timing and magnitude to the Great Depression
in the United States. This has led some to conclude that the two episodes
were essentially identical and share a common explanation (Betts et al.,
1996; Siklos, 2000).
The declines in output, productivity, and employment were very similar.
However, the recoveries, though very similar in terms of output, were different in two important respects. In Canada, productivity did not return to
trend as it did in the United States, while employment recovered more.
The recovery in U.S. productivity led Cole and Ohanian (1999) to conclude that the slow recovery of output per adult in the United States was a
puzzle. Cole and Ohanian (2000b) argue that cartelization and labor market policies can resolve this puzzle. However, in Canada there is no puzzle
because productivity did not return to trend. We found that Canada did not
follow the policies that Cole and Ohanian argue gave rise to the incomplete
recovery in the United States. Our conclusion is that the main reason that
Canadian output per adult was still 30% below trend in 1938 was that productivity failed to return to trend.
Trade accounted for roughly half of Canadian output. A conventional
view is that Canada imported the Great Depression via a collapse in its
terms of trade. We find that this terms of trade shock has a negligible
effect on output in standard models.
A voluminous body of research has developed on the role of deflation in
the Great Depression. We consider four standard transmission mechanisms
that operate through either the credit market or the labor market. We find
that these stories fail to account for the 10-year Canadian depression. These
stories are not consistent with the 1920–1922 deflation, which was similar in
magnitude to the 1929–1933 deflation. Also, these stories are inconsistent
with the slow recovery.
Given our findings we conclude that any successful theory of the Canadian 10-year depression should explain why productivity was so far below
trend for so long. Any explanation should also be consistent with the fact
that productivity recovered in the United States.
2. DATA ON THE GREAT DEPRESSION IN CANADA AND
THE UNITED STATES
This section presents some macroeconomic data on the Canadian and
U.S. economies during the Great Depression. We establish two main points
in this section. First, Canada experienced a decline in output between 1929
great depression in canada
47
and 1939 that was large and quantitatively very similar to that of the United
States.2 Second, in contrast to the United States, Canadian total factor
productivity (TFP) was well below trend throughout 1929–1939.
We use the neoclassical growth model to organize the data. As a result,
we look at per adult variables. Unless otherwise stated, all data are divided
by the number of people older than 14 for Canada and older than 16 for
the United States.
We detrend all variables that grow at the same rate as output in a balanced growth path at a 2% yearly rate. This trend rate is close to the
long-term average growth rate for both the United States and Canada. In
detrending, we have taken the view that the growth in production efficiency
due to increases in the stock of usable knowledge is smooth. Other things
being equal, this gives rise to a yearly growth rate of GDP per adult of 2%.3
Real Data
As we can see from Table I, the behavior of real output in the two
countries was very similar. By 1933 both countries were roughly 40% below
trend. The recovery was very protracted in both countries, with the United
States recovering slightly faster than Canada. By the end of the decade,
U.S. output was still 25% below trend while Canada’s was almost 30%
below trend.
Relative to trend, consumption fell more in Canada and remained below
that of the United States throughout the 1930s. Investment in Canada fell to
15% of its trend value by 1933 and recovered very slowly in both countries
(remaining roughly 50% below trend in 1939). Government purchases in
the two countries followed a similar pattern during the downturn, before
diverging in the late 1930s when U.S. government spending remained above
trend, while in Canada it fluctuated about trend.
Having looked at the product side, we now turn to the input side. We
first calculate TFP, the part of output growth that cannot be attributed to
input growth. We do this using the production function,
Yt = At Ktθ Ht1−θ
(1)
Henceforth, capital letters denote aggregate variables, while lowercase
letters denote household variables. Y is aggregate output, K is aggregate
capital, H are aggregate hours, and A is the TFP factor.
1939
Given values for Yt Kt Ht 1939
t=1929 and θ we can compute At t=1929 .
The parameter θ is the share of product that accrues to factor payments
2
We look at 1929–1939; however, Canadian GDP per adult peaks in 1928.
Average per capita GDP growth in Canada over the twentieth century is actually slightly
higher than that in the United States.
3
48
amaral and macgee
TABLE I
Detrended Levels of Real Output and Its Components: 1929–1939
Year
Can.
GNP
U.S.
GNP
Can.
cons.
U.S.
cons.
Can.
inv.
U.S.
inv.
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
100
91.6
77.0
66.5
59.6
64.5
67.1
67.5
71.8
69.7
72.4
100
87.7
79.7
65.9
62.0
65.3
71.5
76.4
80.0
73.2
76.1
100
92.6
81.9
73.4
66.9
67.7
67.9
67.1
68.8
66.2
65.7
100
89.7
83.8
74.2
70.4
70.5
71.9
76.2
76.5
72.7
73.8
100
85
50.5
24.8
15.2
28.5
32.9
28.2
44
42.8
52
100
69.2
46.1
22.2
21.8
27.9
41.7
52.6
59.5
38.6
49
Year
Can.
govt.
U.S.
govt.
Can.
exp.
U.S.
exp.
Can.
imp.
U.S.
imp.
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
100
104.8
104.6
97
81
101
101.1
94.6
98.1
108.9
102.8
100
105.1
105.3
97.2
91.5
100.8
99.8
113.5
105.8
111.5
112.3
100
77
58.7
51.8
51.9
60.7
65.2
75.9
79.2
64.9
67.4
100
85.2
70.5
54.4
52.7
52.7
53.6
55
64.1
62.5
61.4
100
81.6
58.6
49
44.1
48
49.3
53.4
59.6
50.9
52.3
100
84.9
72.4
58
60.7
58.1
69.1
71.7
78
58.3
61.3
Note. Cons. is consumption, inv. is investment, govt. is government purchases, exp. is
exports, and imp. is imports. The Canadian data are from Historical Statistics of Canada,
Series F1-13. The U.S. data for GNP are from Kendrick (1961), and those for the different
components are from Cole and Ohanian (1999).
to capital. From both countries’ national accounts we get θCAN = 03 and
θUS = 033.
Table II presents the computed series.4 Notice that TFP in the United
States (TFP U.S.) recovers much faster than TFP in Canada (TFP Can),
and it is back to trend by the end of the decade. This pattern is the same
for output per hour. Two questions emerge right away: Why did TFP fall so
much in both countries, and why did it not recover in Canada? We return
to these questions later in the paper.
4
Note that TFP is not detrended at a rate of 2% but at a rate equal to 1021−θ for each
country, a trend that is close to the historical averages (excluding war periods).
great depression in canada
49
TABLE II
Detrended Inputs
Year
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
TFP Can.
100
99.49 92.04 90.25 85.29 85.43 87.95 87.79 88.69 89.33
93.60
TFP U.S.
100
93.27 91.15 83.34 80.35 86.91 93.89 96.38 99.63 97.33 100.1
K Can.
100
96.93 92.43 86.68 80.95 76.21 72.38 68.87 66.2
63.85
62.04
K U.S.
100
98.55 96.16 92.23 87.27 82.64 79.26 76.61 74.97 73.11
71.05
H Can.
100
91.70 83.32 72.70 69.18 77.42 79.52 82.43 88.66 86.36
86.19
H U.S.
100
91.92 83.57 73.41 72.62 71.73 74.72 80.63 83.03 76.25
78.68
Note. Canadian capital data are from Brown (1965, p. 199, Series 5). The U.S. capital
data are from Kendrick (1961, Table A-XV). The U.S. hours data are from Kendrick (1961,
Table A-X). Canadian hours data are Series C-51 from Urquhart (1965) multiplied by average
hours worked in non-agriculture, Series D-409. We used GDP from the National Income and
Expenditure Accounts (1988). U.S. GDP is from Kendrick (1961). U.S. population is from U.S.
Bureau of the Census (1965).
We are aware that what is presented in Table II is not TFP, but measured
TFP. There are a number of reasons why measured TFP may differ from the
actual TFP. One major issue is factor mismeasurement. In terms of capital,
there is the issue of capacity utilization. In terms of labor, there is evidence
that the reduction in employment was much more severe for unskilled than
for skilled workers. We used Ohanian’s (2001) estimates for the United
States for the magnitude of these factors and recomputed TFP. We found
that these two factors roughly cancel each other, so that measured TFP is
almost unchanged.
Another measurement question relates to what Bernanke and Parkinson
(1991), among others, term labor hoarding. However, 10 years seems to be
too long a period for this argument to make sense.
Finally, there is the issue of sectoral compositional effects. We could only
compute TFP for manufacturing. Manufacturing TFP is similar to aggregate
TFP.
The measured TFP is completely determined by the path of the inputs.
So we now look at the inputs.
Table II reports the capital stock for both countries (K Can and K U.S.).
The most important feature regarding the capital stock is that it declines by
more in Canada than it does in the United States. This result is not due to
higher depreciation and is in contrast to the investment figures in Table I.
Thus, there are problems with the capital stock data. Do they affect the
qualitative results in terms of the measured TFP? We think they do not. If
anything, Canada’s capital stock declined by less than the value reported
in Table II. This suggests that adjusting for possible measurement errors
50
amaral and macgee
in the capital stock would imply that TFP in Canada was even lower than
reported.
Table II also compares total hours worked for the two countries (H Can
and H U.S.). Total hours are the product of the number of people employed
and average hours worked. The series for Canada was computed using
average hours for the nonagricultural sector, since a series for the whole
economy (or for agriculture) was not available. This is likely to lead us to
overestimate the fall in labor input in Canada, as agricultural hours in the
United States (and most probably in Canada) fell by less than nonagricultural hours.
The main difference in measured TFP lies in the employment data. During the recovery period, total hours in Canada recovered more than did
total hours in the United States. The question about the lack of recovery
of TFP in Canada relative to the United States can now be posed as: Why
did total hours recover faster in Canada than in the United States?
We now compare the private nonagricultural sectors in the two countries.
This is an interesting disaggregation for several reasons. First, aggregate
employment and output figures were influenced by different government
policies toward public works and relief spending.5 Second, agriculture was
hit by identifiable weather shocks in both countries. Also, the agricultural
sector is a relatively small6 fraction of GDP.
As Table I documents, U.S. government output increased more relative
to trend than Canadian government output. A large part of the difference
in government expenditure can be attributed to different government policies toward providing unemployment relief. In the United States, the government relied much more heavily upon make-work projects (government
relief projects) than in Canada. The fraction of the workforce employed by
the government doubled in the United States, while increasing by less than
50% in Canada. The increase in U.S. government employment was mainly
due to public works, as nearly 7% of U.S. employment in the late 1930s was
in relief projects. Relief workers were never more than 1.5% of the total
number of employed people in Canada. (See Amaral and MacGee, 2001,
for more details.)
Table III reports TFP for the private nonagricultural sector. The calculation method and the shares used were the same as those for aggregate TFP.
We also use the same series for capital as before. We assume that the private nonagricultural sector benefits from the services of government-owned
capital. Total hours in Canada equal the product of employment in the
5
Government enterprises are included in the private sector.
Agricultural GDP as a fraction of total GDP averaged 6.2% in the United States and
10.4% in Canada from 1929 to 1939.
6
great depression in canada
51
TABLE III
Detrended Private Nonagricultural TFP
Year
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
Can.
100
99.32
92.43
92.36
88.34
85.23
87.03
88.6
88.3
88.75
93.52
U.S.
100
93.56
88.72
82.17
79.79
87.75
90
97.43
94.59
94.26
96.49
private nonagricultural sector and average hours in nonagriculture. Total
hours for the United States are from Kendrick (1961).
In Canada, TFP in this sector is very similar to aggregate TFP. U.S. TFP
in this sector behaved differently from aggregate TFP during the recovery
period, since it stopped recovering in 1936, while aggregate TFP recovered
continuously and was back to trend by 1937. This, we claim, is a major difference between the Canadian and U.S. experiences in the Great Depression.
The comparison of the private nonagricultural sectors reinforces our earlier conclusion that the two countries look very similar during the downturn
(1929–1933). However, this sectoral breakdown provides new insights into
the recovery period. It suggests that in the United States, something happened around 1936 that induced a decrease in productivity. In Canada,
the data reinforce the aggregate data—namely, that productivity did not
recover relative to trend during the Great Depression.
Nominal Data
Given that much research on the Great Depression has focused on the
role of monetary shocks, we present data on nominal variables that are
central to monetary business cycle theory. As Tables IV and V show, the
onset of the Great Depression coincided with a decline in money supply
and price levels of approximately 20% in both Canada and the United
States. This deflation was accompanied by a decline in nominal interest
rates, although real ex post rates were high by historical standards.
A cross-country comparison of interest rates is limited by the fact that
a market in short-term government securities in Canada did not exist
before 1934. The available data suggest that nominal interest rate spreads
increased from 1930 to 1932, before narrowing. Short-term interest rates
on government bonds did not fall as quickly as U.S. short-term rates did.
This increase in the interest rate spread from 1930 to 1932 appears to be
linked to differences in monetary policy.
Canada was the first country to leave the gold standard, suspending gold
shipments in January 1929 (Bordo and Redish, 1990). Despite the suspension of convertibility, the Canadian government took steps to prevent
depreciation of the dollar, motivated in part by a wish to maintain access
52
amaral and macgee
TABLE IV
Nominal Money, Prices, and Interest Rates in 1929–1939 in Canada
Year
Monetary
base
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
100
88.48
79.34
78.06
77.50
78.10
85.85
93.04
101.96
108.09
116.59
Price
level
M1
100
89.65
83.16
72.44
70.95
73.19
80.00
84.62
93.63
92.77
99.08
100
97.52
91.46
82.92
81.54
82.64
82.92
85.67
87.88
87.88
87.05
3-month
T-bill
Short-term
dom. bonds
Comm.
paper
—
—
—
—
—
2.83
1.249
0.753
0.763
0.676
0.808
5.34
4.87
4.43
5.08
4.15
2.91
2.29
1.61
1.93
—
—
5.31
5.28
5.64
6.6
6.49
5.27
4.76
4.12
3.95
—
—
Note. The monetary data are from Metcalfe et al. (1996), and the GNE deflator (price level)
is from Historical Statistics of Canada. The 3-month T-bill rate is from Historical Statistics of
Canada (H588–603). The short-term Dominion bonds and the corporate paper rate are from
Nixon (1937).
to U.S. capital markets to refinance Dominion debt (Shearer and Clark,
1984). As a result, the government maintained the advance rate at its 1928
level throughout 1930, despite the fall in world rates. This policy was ultimately abandoned in 1931. Despite this, the Canadian dollar did depreciate
relative to the U.S. dollar by approximately 15% between 1929 and 1931,
before recovering to its 1929 level in 1935.
TABLE V
Nominal Money, Prices, and Interest Rates in 1929–1939 in the United States
Year
Monetary
base
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
100
95.9
98.7
104.3
108.9
119.8
139.2
157.2
168.5
181.5
215.5
M1
Price
level
100
94.4
85.6
74.5
69.9
78.0
91.0
102.1
102.9
102.2
113.7
100
97.0
88.1
78.4
76.7
83.2
84.8
85.2
89.4
87.2
86.6
Note. Data are from Cole and Ohanian (1999, Table 8).
3-month
T-bill
Comm.
paper
4.4
2.2
1.2
0.8
0.3
0.3
0.2
0.1
0.5
0.1
0.0
6.1
4.3
2.6
2.7
1.7
2.0
0.8
0.8
0.9
0.8
0.8
great depression in canada
53
We suspect that the data reported dramatically overstate the difference
in commercial paper rates in the late 1930s. Neufeld (1972) presents the
commercial paper rate reported by Moody’s for Canada and the United
States. His data suggest that while the spread between Canadian and U.S.
corporate paper widened from 1930 to 1932; it then dramatically narrowed
and remained under 1% from 1934 to 1939.
The data suggest that while monetary shocks may help explain the 1929–
1933 downturn, it is unlikely that they played a significant role in the protracted recovery from 1934 to 1939. In both countries, the deflation ended
by 1933 and both nominal and real interest rates remained low by historical
standards.
Summary
We view the different behavior of TFP (together with the behavior of the
labor input discussed below) as the main difference in the two countries’
experiences during the recovery period. Both Canadian and U.S. TFP fell
roughly 20% relative to trend from 1929 to 1933, but unlike the U.S. TFP,
Canadian TFP was not back to trend by the end of the decade.
This contrasts with the similarities in per capita GNP. In both countries
it fell 40% relative to trend from 1929 to 1933, and its recovery was very
protracted. Finally, both prices and money aggregates fell considerably from
1929 to 1933 but had recovered by the end of the decade.
In the remainder of the paper, we use theory and data to assess different
explanations of the Great Depression in Canada.
3. HOW IMPORTANT WERE TFP SHOCKS?
In this section, we ask how much of the Great Depression can be
explained by measured TFP. In undertaking this experiment, we take measured productivity as exogenous and feed this series into the standard
stochastic growth model.
We conclude that the decline in measured TFP in Canada can account
for over half of the decline and does a very good job of accounting for the
protracted recovery. The TFP story can also account for 70% of the decline
in U.S. per adult output, but cannot account for the slow recovery.
Model
The production function is Eq. (1). To complete the description of technology, the law of motion of capital is
Kt+1 = 1 − δKt + Xt
(2)
54
amaral and macgee
TABLE VI
Calibration
Parameters
θ
β
χ
δ
γ
η
ψ
σ
Canada
0.3
0.96
1.7
0.05
0.02
0.02
0.9
0.017
U.S.
0.33
0.96
1.66
0.05
0.02
0.01
0.9
0.017
where δ is the depreciation rate and X is investment. The process for the
stochastic technological shock is
At+1 = 1 − ψ + ψAt + εt+1
(3)
where the stochastic components are independently and identically distributed with mean zero and variance σ 2 .
Preferences are defined over consumption and leisure and are represented by
E0
∞
t=0
βt 1 + ηt logct + χ log1 − ht
(4)
where h is time devoted to market activities and η is the population growth
rate.
We calibrate the model economy by choosing parameters in such a way
that the balanced growth path of the model economy matches certain steady
state features of the measured economies (see Cooley and Prescott, 1995).
All parameter values are reported in Table VI.
The depreciation rate is set to match an investment-to-capital ratio of
0.09 for Canada and 0.08 for the United States. β is chosen to match a
capital–output ratio of 2.6 in Canada and 2.8 in the United States. We
choose χ to match the fact that households dedicate one-third of their time
to market activities. γ is the growth rate of real per capita output, which
we take to be 2% for both countries. The population growth rate η is 2%
for Canada and 1% for the United States. Given the long-run similarities
between measured TFP in the two countries, we follow Cole and Ohanian
(1999) and set ψCAN = ψUS = 09 and σCAN = σUS = 0017.
The optimal decision functions are computed using a linear quadratic
solution method. Given the capital stock in 1929, which we assume to be
on a balanced growth path, we feed in the measured TFP series for each
country from Table II and obtain paths for all the variables in the model.
Findings
The results are in Figs. 1 and 2. The fall in measured TFP can account
for over half of the fall in output up to 1933 in Canada and approximately
great depression in canada
FIG. 1.
55
The effect of TFP shocks on Canadian output.
70% of that in the United States. The technology shock story qualitatively
matches the Canadian experience. The model predicts a faster recovery in
the United States than actually occurred. Given the fast recovery in TFP,
the model predicts an equally fast recovery in inputs. In reality, this recovery
in inputs is very protracted, as can be seen from Table II. The model also
predicts a faster recovery for market hours in the United States than in
Canada.
From this section we obtain one finding and two puzzles. The finding is
that measured TFP in Canada can account for over half of the decline and
does a very good job of accounting for the protracted recovery. This leads
FIG. 2.
The effect of TFP shocks on U.S. output.
56
amaral and macgee
us to conclude that any explanation for the Great Depression in Canada
should be consistent with the TFP behavior. Two puzzles remain. The first
puzzle is why did TFP decline so much in both countries? The second is
why was there no recovery in TFP in Canada while it recovered in the
United States?
4. HOW IMPORTANT WAS DEFLATION?
Many economists have argued that money and banking shocks caused
the Great Depression. In this section, we evaluate the effect of monetary
shocks in Canada. We follow recent work by Cole and Ohanian (2000a) for
the United States and focus our attention on four monetary transmission
mechanisms. The first two are the most common stories of financial factors:
(i) debt deflation and (ii) bank failures. The remaining two work via labor
markets and are (iii) surprise deflation and (iv) imperfectly flexible wages.
Our conclusion is that money and banking shocks can account for a small
part of the downturn and play an insignificant role in the slow recovery. The
insignificant role played by financial factors is not surprising, as Haubrich
(1990) provides strong evidence that they did not matter. Our results lend
further weight to his conclusion and provide some interesting parallels with
the analysis of Cole and Ohanian (2000a) for the United States.
We devote the greatest attention to the labor market transmission mechanisms, particularly the imperfectly flexible wages story, for two reasons.
First, to our knowledge, this story has not been subjected to a careful evaluation for Canada. Second, this transmission mechanism has figured prominently in recent papers arguing that money and banking shocks played a
key role in the Great Depression (i.e., Bordo et al., 1997; Bernanke, 1995).
Our conclusion is that imperfectly flexible wages may play a small role in
the downturn but play no role in explaining the slow recovery.
A key argument against the money and banking stories that we emphasize
is a consistency requirement. Cole and Ohanian (2000a) point out that
monetary explanations of the Great Depression in the United States face
the problem of explaining why the deflation of 1920–1922 was associated
with a short depression. This leads them to argue that any transmission
mechanism must be consistent with both the deflation of 1929–1933 and
the (comparable) deflation of 1920–1922.
Table VII shows that Canada and the United States experienced similar deflations in 1920–1922 and 1929–1933. In both countries, output
fell much more between 1929 and 1933 than between 1920 and 1922.
Paraphrasing Cole and Ohanian (2000a): If the 19% deflation caused the
great depression in canada
57
TABLE VII
Comparing 1920–1922 and 1929–1935
Year
Y Can.
P Can.
Y U.S.
P U.S.
1920
1921
1922
100
85.71
97.77
100
84.58
73.12
100
93.9
96.2
100
85.2
80.6
1929
1930
1931
1932
1933
100
91.6
77
66.5
59.6
100
97.5
91.5
82.9
81.5
100
86.9
77.6
64
60.9
100
97.5
88.5
79.5
77.5
Note. Y is GNP per capita, and P is the GNP deflator. The U.S. data are from Cole
and Ohanian (2000a). The Canadian data are from Urquhart (1965, 1993). All real data are
detrended.
Great Depression in Canada, why didn’t the 27% deflation of 1920–1922
also cause a major depression?7
Credit Markets
We consider two alternative channels via which deflation could have
helped cause the Great Depression through credit market disruption. The
first is debt deflation, and the second is a financial crisis that may have
disrupted intermediation.
The “debt deflation” view of the Great Depression asserts that deflation
and high private debt levels contributed to the Great Depression by reducing borrower wealth and constraining lending. Haubrich (1990) argues that
the debt crisis was much less severe in Canada than in the United States.
He argues that there is little evidence to suggest that the debt crisis caused
the Great Depression in Canada.
Comparing the 1920s with the 1930s supports Haubrich’s (1990) conclusion. If Canada experienced a debt deflation crisis, then business failures
should increase. Table VIII reports commercial failures in both countries.
The Canadian series includes bankruptcies, insolvencies under provincial
company acts, and proceedings such as bulk sales and tariff sales, which
led to loss to creditors. The U.S. failure data include any business that was
involved in court procedures or voluntary action, which probably ended in
a loss to creditors.
7
The two depressions were also similar in that Canada faced deteriorating terms of trade
and the Canadian dollar depreciated relative to the U.S. dollar by a similar amount during
both depressions.
58
amaral and macgee
TABLE VIII
Commercial Failures
Canada
United States
Year
Number of
failures
Liability
(thousands)
Number of
failures
Liability
(millions)
1920
1921
1922
1923
1924
1925
1078
2451
3695
3247
2474
2371
26,494
73,299
78,069
65,810
64,531
45,768
—
19,652
23,676
18,718
20,615
21,214
—
627
624
539
543
444
1929
1930
1931
1932
1933
1934
1935
2310
2741
2563
2938
2344
1627
1402
44,441
57,191
52,987
56,631
29,251
20,728
14,542
22,909
26,355
28,285
31,822
19,859
12,091
12,244
483
668
736
928
458
334
311
Note. The Canadian data are from Historical Statistics of Canada (Y215–216). The U.S.
data are from U.S. Department of Commerce (1975), (V20–30). All liabilities are in current
dollars.
What is striking is that the number of commercial failures is not that high
during the Great Depression. Indeed, while commercial failures in Canada
more than tripled between 1920 and 1922, they increased by less than 20%
between 1929 and 1932, before plunging to their lowest levels since 1920 in
1934. The U.S. data also show a similar pattern. However, failures in the
United States increase more than those in Canada between 1929 and 1932.
This suggests that the debt crisis story is not a good candidate to explain
the Great Depression in Canada.
A variation on the debt crisis story that may apply to Canada is the role
of external debt. Canada had borrowed considerably from abroad prior to
the Great Depression. There are two problems with this story. First, as
noted above, there was a decline in failures during this period. Second, the
risk premium on Dominion bonds sold abroad—primarily in the United
States—did not significantly increase during the 1930s. This suggests that
investors did not view Canada as likely to default and lends further credence
to the view that there was no external debt crisis.
A common view is that banking crises played a significant role in transforming the 1929 downturn into the Great Depression. For example,
Bernanke (1983) states that “the financial crisis of 1930–33 affected the
macroeconomy by reducing the quantity of financial services, primarily
credit intermediation” (p. 262). As has been pointed out by numerous
authors, however, Canada did not experience any bank failures. While
great depression in canada
59
the number of branches did fall, Haubrich (1990) finds no evidence that
this impacted the level of economic activity. Indeed, Haubrich (1990) concluded that if monetary shocks mattered in the United States, it must have
been because of the financial crisis.
Labor Markets
We consider two alternative channels via which deflation could have
helped cause the Great Depression through labor market disruption: the
surprise deflation story and the sticky wage story.
The surprise deflation story of Lucas and Rapping (1969) argues that the
Great Depression was severe because it was unexpected. Cole and Ohanian
(2000a) point out that for this story to work, we should observe low nominal
interest rates in the 1920s and high nominal and ex post real interest rates
in the 1930s.
Interest rate data for both countries are reported in Table IX. Real interest rates are the nominal interest rates minus the percentage change in the
annual GNP deflator. We report long-term Dominion bond yields as shortterm Treasury bill yields are not available until 1934. The Canadian data
match the U.S. data—and suggest that the 1930s deflation was more predictable than the 1920s deflation. This leads us to conclude that the surprise
deflation story cannot explain the Great Depression in Canada.
The last monetary story we consider is that imperfectly flexible nominal
wages and deflation led to high real wages. This story assumes that the
short side of the labor market dominates, so that high real wages cause
firms to lower their demand for labor, leading to lower employment and
output.
TABLE IX
Nominal and ex-Post Real Interest Rates in Canada and the United States:
1920–1922 and 1930–1933
Years
U.S. Treasury notes
(nom.)
U.S. Treasury bills
(real)
Long-term
Dominion (nom.)
Long-term
Dominion (real)
1921
1922
Avg.
1930
1931
1932
1933
Avg.
4.83
3.47
4.35
2.23
1.15
0.78
0.26
1.10
19.63
8.87
14.25
4.73
10.38
10.95
2.78
7.21
5.99
5.41
5.7
4.73
4.55
5.12
4.6
4.75
15.36
7.23
10.76
14.46
6.26
9.68
16.01
14.7
Note. U.S. data are from Cole and Ohanian (2000a, Table 4). Canadian data are from
Historical Statistics of Canada (H604–618).
60
amaral and macgee
We find that high real wages can account for no more than a 7% decline
in output and that predicted output is above trend by 1933. The imperfectly
flexible wage story also has a consistency problem, as changes in measured
real wages during the Great Depression are similar to changes during the
1920–1922 depression.
There are a number of critical issues involved in this story. As McGrattan
(1999, 2001) has pointed out, the relevant variable from the point of view
of the firm is the ratio of the product price to the nominal wage rate. She
finds that a key theoretical problem with standard sticky wage models is
that the spread between the output price and the real wage does not vary
much.
McGrattan’s work also points to an important empirical question: the
price index one uses to deflate wages matters. We use the GDP deflator.
There were large changes in the relative prices of different types of goods.
The prices of agricultural products and other commodities fell substantially
relative to those of other goods. Both the wholesale price index and the
consumer price index overweight commodities and agricultural products.
This means that using either one as a deflator would overestimate the real
wage.
Figure 3 shows undetrended real wages for the industrial sector and agriculture. The nominal wage index for the industrial sector is based on the
weighted average of eight nonagricultural industries (one of which is manufacturing). These nominal wage indices are for wage earners and are based
on surveys conducted by the Dominion Bureau of Statistics of employers.
Wage earners comprised approximately 70% of the workforce (with most
of the remaining workers being farmers). The farm wage series is com-
FIG. 3.
Real wages in Canada.
great depression in canada
61
puted using indices of farm wages reported in various issues of the Labour
Gazette.
The figure shows an important fact: real wages differed substantially
across sectors of the economy. For the industrial workers the undetrended
real wage increased by only 7% during the decline.
The industrial real wage reported in Fig. 3 may be biased by compositional effects. The reduction in employment affected unskilled workers the
most. Cole and Ohanian (2000a) argue that for the United States, compositional effects could cause the reported real wage to be overstated by up to
15%. Given the similarities between Canada and the United States, it could
be argued that a similar figure also applies to Canada. This adjustment
would imply that real wages were actually low during the Great Depression.
The real wage story also faces a consistency problem. Nominal wage
inflexibility in 1920–1922 appears very similar to that of 1929–1939. For
the United States, this fact has been established by both Cole and Ohanian
(2000a) and Dighe (1997). Table X reports detrended real wage movements
in Canada and the United States. The real wage movements are very similar. Real wages in the nonagricultural sector increased slightly more over
1929–1931 than they did over the 1920–1922 period. Conversely, real wages
in the farm sector fell more in the Great Depression than during the 1920s.
Even if one rejects the arguments presented above, the imperfectly flexible nominal wage story is quantitatively unable to explain the Great Depression. To show this, we undertake the following experiment. We modify the
model economy from Section 3 by assuming that the wage rate is determined exogenously and is given by the real wage in the industrial sector.
The labor input is determined by the firms’ first-order condition. Since we
are taking a real wage series for the industrial sector, we compare the predictions of the model to the nonagricultural sector.
Figure 4 shows that the model fails to replicate the magnitude of the
initial fall in output. It also completely fails to explain the lack of recovery
as it predicts that output should be above trend by 1933.
TABLE X
Detrended Real Wages
Year
1920
1921
1922
1929
1930
1931
1932
1933
Can. farm
U.S. farm
Can. manuf
U.S. manuf.
100
100
100
100
85.7
71.9
103.3
101.5
81.3
73.1
100.1
101.2
100
100
100
100
89.9
93
101.1
102.1
71.9
76.8
103.7
106.8
58.9
64.7
105.4
106.5
55
60.2
100.6
104.2
Note. Date are from Cole and Ohanian (2000a), Historical Statistics of Canada (D1–11),
and Labour Gazette. The wage series are deflated using a 1.4% linear trend.
62
amaral and macgee
FIG. 4.
The effect of imperfectly flexible nominal wages on Canadian output.
We also repeated this experiment combining the drop in measured TFP
for the nonagricultural sector and the reported real wage series. In this
case, the model can account for most of the decline, but predicted output
is back to trend by 1939. Comparing this to the experiment where we take
only measured TFP as exogenous, we can explain 70% of the decline, as
opposed to 60%. However, the predicted recovery is much faster.
We conclude that the imperfectly flexible nominal wage story cannot
explain the Great Depression in Canada. Indeed, given the questions about
the data, it is an open question as to whether the real wage was actually
“high” during the Great Depression.
Summary
Our conclusion is that money and banking shocks are unable to explain
the Great Depression in Canada. Furthermore, none of these monetary
explanations provides a direct channel for explaining either the observed
drop in productivity or its lack of recovery.
5. HOW IMPORTANT WERE COMPETITION AND
LABOR MARKET POLICIES?
The role of government policies in the Great Depression—particularly
U.S. “New Deal” policies—has long been a subject of debate among
economists. Cole and Ohanian (2000b) present persuasive arguments that
U.S. government competition and labor market policy play a key role in
explaining the slow recovery from the Great Depression. In this section we
great depression in canada
63
ask two questions: What were the labor market and competition policies
in Canada during the Great Depression?8 What was their impact?
What we find is surprising. Government policies were very different in
Canada and the United States, particularly during the recovery period.
The United States pursued a policy of reducing domestic competition and
increasing wages (Cole and Ohanian, 2000b). There is no evidence to
suggest that policies limiting competition were implemented in Canada.
We also find important differences in labor market policies. Unlike the
United States, Canada did not restrict hours worked or strengthen unions.
Canadian policy consisted mainly of measures targeted at directly increasing wage rates. However, as we discussed in Section 4, high real wages can
explain a very small part of the slow recovery in Canada. This leads us to
conclude that neither competition nor labor market policies can explain
the Great Depression in Canada.
The Bennett “New Deal”
From the legislative record, one would conclude that Canada introduced
polices that were very similar to those put in place by the Roosevelt administration in the United States. The Bennett government introduced Canadian “New Deal” legislation in 1934 and 1935, which included the main
features of both the Roosevelt New Deal and British Unemployment Insurance schemes. Unlike in the United States however, this legislation was not
implemented before being struck down by the courts. These policies were
not implemented because the Bennett government was defeated in 1935 by
the Liberals, whose leader opposed these policies.9 Upon coming to power
the Liberals referred the Bennett “New Deal” legislation to the courts.
Most of the substantive elements of the New Deal legislation were ruled
ultra vires. As a result these policies were never implemented.
Competition Policy
There is considerable evidence that Canadian competition policy was
not relaxed during the 1930s. The number of cases dealt with under
the Combines Investigation Act (the antitrust law in Canada) increased
from approximately 50 during the 1923–1925 period to over 100 during
each of the 1926–1930 and 1931–1933 periods. In contrast, the Roosevelt
administration pursued an explicit policy of facilitating cartelization by
8
Amaral and MacGee (2001) provide a more detailed description of Canadian government
policy.
9
King’s view is reflected in his quote that Roosevelt’s “mad desire to bring about State
control and interference beyond all bounds made one shudder” (Struthers, 1983, p. 105).
64
amaral and macgee
not enforcing antitrust laws. This is reflected in the nearly 50% fall from
1925–1929 to 1930–193410 in antitrust cases filed by the U.S. government.
Price behavior during the 1930s also supports the view that competition
policy differed across the two countries. Wholesale prices in Canada and
the United States moved together during the interwar period, except for the
1933–1936 period, when U.S. prices rose much more quickly than Canadian
prices. Romer (1999) attributes this rise in U.S. prices to the effects of the
National Industrial Recovery Act (NIRA). This suggests that competition
in Canada was less restrained by government policy during this period.
Labor Market Policy
In both countries, labor market policy attempted to increase wages. In
Canada, these labor market policies primarily took the form of provincial
governments’ minimum wage schedules. These governments put very few
restrictions on hours worked per worker and did not significantly change
labor legislation. In sharp contrast, the Roosevelt government both limited
hours worked per worker and increased the bargaining position of unions
relative to management.
Most labor market intervention in Canada was done by provincial
governments. They introduced minimum wage laws after the 1920–1922
depression. These laws initially applied solely to female workers in the
nonagricultural sector. These minimum wage schedules were unchanged
until being superseded by other legislation in the late 1930s. In 1934, these
minimum wages were extended to male workers replacing female workers. From 1935 to 1937, legislation was passed which allowed provincial
governments to set minimum wage schedules by industry. Although this
legislation allowed for the regulation of hours, this provision was rarely
used.
Provincial government intervention probably increased nominal wages
during the late 1930s. Did this policy have a large impact on the recovery?
The answer is no. As we discussed in Section 4, high real wages are unable
to account for the slow recovery. Since labor market policy only increased
wages, it cannot explain the slow recovery.
The contrast with U.S. labor market policy is substantial. U.S. government policy not only increased nominal wages, but also attempted to
decrease hours worked. This policy was explicit under the NIRA (1933–
1935). After the NIRA was ruled unconstitutional in 1935, this policy was
implicitly implemented through the National Labor Relations Act (NLRA),
which strengthened the position of unions. The effect of this policy can be
10
The number of antitrust cases filed is contained in Cox (1981), who cites data compiled
by Posner.
great depression in canada
65
seen clearly in the rapid growth of union membership in the late 1930s.
The fraction of unionized employees in nonagricultural jobs nearly doubled, increasing from 14% in 1936 to 27% in 1938. There was also an
increase in strikes in the mid-1930s.
Canadian government policy was very different. The influence of unions
did not increase. This is reflected in the fact that the fraction of unionized
workers did not increase (nor was there a large increase in labor unrest).
The timing suggests that U.S. labor market policy may have slowed the
recovery by adversely impacting productivity. The growth accounting exercise for the nonagricultural sector in Section 2 suggests that the recovery of
TFP was abbreviated in 1936 in the United States, but continued unabated
in Canada. This change in U.S. TFP coincided with the strengthening of the
unions and the rise in the fraction of the unionized labor force. This conjecture is consistent with anecdotal evidence that the formation of unions
in the 1930s lowered productivity (see Chap. 5 of Brecher, 1997).
Summary
Canadian government competition and labor market policy cannot
account for the Great Depression. There is no evidence to suggest that
Canadian governments undertook policies to reduce domestic competition. The main effect of labor market policies was to increase nominal
wages, particularly during the late 1930s. However, high real wages cannot account for the slow recovery (see Section 4). Moreover, in Canada
hours worked recovered to a much greater extent than those in the United
States, which also suggests that labor market policies in Canada were less
restrictive than those in the United States.
Our analysis suggests that U.S. New Deal policies may have prolonged
the Great Depression by halting the recovery in TFP. The productivity
recovery in the private nonagricultural sector was arrested at precisely the
time that U.S. labor legislation strengthened unions. This suggests another
avenue via which the NLRA may have slowed the U.S. recovery.
6. HOW IMPORTANT WERE TERMS OF TRADE SHOCKS?
In contrast to the United States, Canada had a very large trade sector
with exports plus imports accounting for approximately 50% of GDP. In
this section, we quantify the contribution of terms of trade shocks to the
Great Depression in Canada. We first consider a simple partial equilibrium
argument and then undertake a dynamic analysis using an open economy
model. Our conclusion is surprising. Despite the fact that trade declined by
50% and only partially recovered (see Table I), we find that terms of trade
shocks can account for less than 5% of the decline in GDP.
66
amaral and macgee
FIG. 5.
Terms of trade in Canada.
Figure 511 shows that the onset of the Great Depression was associated with an adverse movement in Canada’s terms of trade. This can be
attributed to several factors. First, Canada was a net exporter of commodities. In particular, Canada was a major exporter of wheat, which experienced a large decline in price relative to that of other goods during the
early 1930s. Another factor was the large increase in tariffs both in Canada
and abroad. Canada increased tariff rates by 50% on average in 1930 in
retaliation to the Smoot–Hawley Act and imposed a number of nontariff trade barriers. These nontariff barriers were substantial, as the Ministry
of National Revenue made extensive use of its power to assign artificial
valuations to Canadian imports (Brecher, 1957).
One feature of Fig. 5 worth noting is that there was a steeper fall in the
terms of trade in 1920–1922, and this did not cause a protracted depression. This suggests that the terms of trade shock story faces a consistency
problem.
A simple back-of-the-envelope calculation suggests that trade cannot
explain more than a third of the Great Depression in Canada. Suppose
that a reduction in exports will lead to a one-for-one reduction in output.
Exports were roughly 25% of the Canadian GNP in 1929. By 1932 they had
fallen by slightly more than half their 1929 level. If factors used in the production of exports could not be reallocated, then this could account for a
decline of 13.5% in output at most. This is roughly one-third of the actual
decline in real GNP per capita. Moreover, the fall in output that can be
11
The terms of trade are the ratio of an index of Canadian prices of export goods divided
by an index of Canadian prices of imported goods.
great depression in canada
67
attributed to a decline in trade with the United States is less than half of
this figure—less than 6%.
This calculation also casts doubt on the common wisdom that the United
States transmitted the Great Depression to Canada via trade (Safarian,
1970). While the United States was Canada’s largest trading partner in
1929 (having surpassed the United Kingdom in 1927), the United States
received 35%–45% of Canadian exports. Furthermore, while total trade
with the United States fell by more than half between 1929 and 1933, net
exports to the United States increased.
The above exercise, although illustrative, abstracts from important issues,
namely the fact that domestic and imported goods can be imperfect substitutes. If this is the case, the domestic country is partially unable to substitute
away from imports as their relative price increases. This will lower investment, which in turn will lead to a fall in output.
We use a variation of the Backus et al. (1995) model to quantify the
effects of terms of trade shocks. Canada is modeled as a small, open economy that takes terms of trade as given. For the sake of consistency with the
Backus et al. language, we define the terms of trade as the price of imports
divided by the price of exports.
The economy is populated by an infinitely lived representative household
whose preferences can be represented by Eq. (3).
The home country, Canada, specializes in the production of a single good,
which we call a. The rest of the world specializes in the production of a single good b. Canada produces a using a constant returns to scale production
function that takes as inputs domestic labor and domestic capital,
act + af t = Yc t = At Ktθ Ht1−θ
(5)
where θ is capital’s share of product and A is total factor productivity. The
process for At is the same as that defined in Section 3.
Yc is GDP in Canada. This can be consumed either in Canada, ac , or
exported abroad, af . Domestic consumption C and investment X are composites of the imports bc and the domestic good ac ,
Ct + Xt = Gac t bc t
(6)
where G is an aggregator given by
1−ρ
1−ρ
Gac t bc t = ωac t + bc t
1
1−ρ
(7)
where ω is the relative weight of domestic goods and the elasticity of substitution between foreign and domestic goods is given by σ = 1/ρ. Capital
is a nontraded good, and its law of motion is given by Eq. (2).
68
amaral and macgee
We assume that the world markets for both goods are perfectly competitive. The price of the foreign good is denoted by qf while for the domestic
good we use qc . The trade balance is the value of exports minus the value
of imports and is given by
nxc t = qc t af t − qf t bc t
The terms of trade are given by pt =
qf t
qc f
(8)
, where the process for pt is
pt+1 = ϕ + φpt + εt+1
(9)
Trade is assumed to be balanced throughout, which implies that af t =
pt bc t .12
In calibrating this model, all parameters common to the model in
Section 3 have the same values as shown in Table VI. The parameter ω
was calibrated to match a 25% steady state share of imports in GNP. The
parameters defining the terms of trade process were estimated using ordinary least squares and are ϕ = 03 and φ = 066. Note that because we
use a linear quadratic approximation, the standard deviation of the error
term plays no role.
We assume that the economy is on its balanced growth path in 1929.
The 1929 capital stock is our initial capital stock in the model. We take the
terms of trade from the data (the reciprocal of Fig. 5) and feed these into
the computed decision functions. This gives us the predicted paths for all
the variables in the economy.
We report the results for two different values of the elasticity of substitution (recall that ρ is the reciprocal of the elasticity of substitution). The
first case corresponds to an elasticity of substitution of 8 and the second
case of 0.8. If the terms of trade are going to have any effect on output, it
will be in the second case where the country cannot easily substitute away
from imports.
Figure 6 shows our results. The results indicate that terms of trade shocks
are unable to account for the Great Depression. In both the elastic and
inelastic cases, the model predicts a decline in output of around 3%. However, in the elastic case, the model predicts a slightly bigger fall in trade
than actually occurred, while in the inelastic case trade falls very little.
We have also repeated the above experiments using the TFP series from
the data. We find that combining the two is not significantly different from
using TFP shocks only.
We conclude that adding terms of trade shocks to the business cycle
model does not significantly add to its ability to explain the Great Depression. This result is subject to some caveats. Crucini and Kahn (1996) have
12
This is a reasonable approximation given the data.
great depression in canada
FIG. 6.
69
The effect of terms of trade shocks on Canadian output.
emphasized that a substantial part of the interwar trade was in intermediate
products. If the domestic and foreign intermediate products are imperfect
substitutes in production, an increase in the relative price of the imported
intermediate goods will affect capital accumulation and output. However,
a problem with this story is that it does not provide a link between trade
and the behavior of TFP, which we argue is key to understanding the Great
Depression in Canada.
7. CONCLUSION
Was the Great Depression in Canada similar to the Great Depression
in the United States? While the downturn (1929–1933) was very similar in
both countries, the recovery (1934–1939) was very different. In the United
States, the recovery in output was very slow despite the rapid recovery of
productivity. In Canada, productivity recovered much more slowly than in
the United States, while output recovered almost as quickly. Any explanation of the Great Depression must be able to account for this difference.
TFP shocks can account for a significant part of the Canadian 10-year
depression. This leads us to conclude that any successful explanation of the
Great Depression must be one that involves an initial decline and a very
protracted recovery in measured productivity. However, since we do not
have any theory for either the decline or the lack of recovery of TFP, we
view this TFP behavior as an unresolved puzzle.
Can the usual explanations of the Great Depression account for the
Great Depression in Canada? Our answer to this question is no. As we
show, money shocks, policy shocks, and terms of trade shocks cannot
account for the 10-year depression. Explanations based on these shocks
70
amaral and macgee
fail because their effects are quantitatively too small to explain the Great
Depression.
Our findings in this paper tell us where to go next. Future research into
the Great Depression in Canada should focus on models in which changes
in the level of trade affect the level of productivity. Such models are consistent with the fact that Canada’s TFP and trade both declined from 1929 to
1933. Beginning in 1934, trade began to slowly recover and so did TFP. This
also matches the fact that the only large shock that hit Canada but not the
United States was trade, while the main difference in macroperformance is
the behavior of productivity.
REFERENCES
Amaral, P., and MacGee, J. (2001). “Canadian Government Policy during the Great Depression,” University of Minnesota, mimeo.
Bailey, M. (1983). “The Labor Market in the 1930’s,” in Macro-Economics, Prices and Quantities (J. Tobin, Ed.), Washington, DC: Brookings Institution.
Backus, D., Kehoe, P., and Kydland, F. (1995). “International Business Cycles: Theory and
Evidence,” in Frontiers of Business Cycle Research (T. Cooley, Ed.), Princeton, NJ: Princeton
Univ. Press.
Bernanke, B. (1983). “Non-Monetary Effects of the Financial Crisis in the Propagation of the
Great Depression,” American Economic Review 73, 257–276.
Bernanke, B. (1995). “The Macroeconomics of the Great Depression: A Comparative
Approach,” Journal of Money, Credit and Banking 27, 1–28.
Bernanke, B., and Parkinson, M. (1991). “Procyclical Labor Productivity and Competing Theories of the Business Cycle: Some Evidence from Interwar U.S. Manufacturing Industries,”
Journal of Political Economy 99, 439–459.
Betts, C., Bordo, M., and Redish, A. (1996). “A Small Open Economy in Depression: Lessons
from Canada in the 1930s,” Canadian Journal of Economics 39, 1–36.
Bordo, M., Erceg, C., and Evans, C. (1997). “Money, Sticky Wages and The Great Depression,”
NBER Working Paper 6071.
Bordo, M., and Redish, A. (1990). “Credible Commitment and Exchange Rate Stability:
Canada’s Interwar Experience,” Canadian Journal of Economics 23, 357–380.
Brecher, I. (1957). Monetary and Fiscal Thought and Policy in Canada: 1919–1939, Toronto:
Univ. of Toronto Press.
Brecher, J. (1997). Strike. Boston: South End Press.
Brown, T. M. (1965). Canadian Economic Growth, Ottawa: Queen’s Printer.
Cole, H., and Ohanian, L. (1999). “The Great Depression in the United States from a Neoclassical Perspective,” Federal Reserve Bank of Minneapolis Quarterly Review 23, 1–24.
Cole, H., and Ohanian, L. (2000a). “Re-Examining the Contributions of Money and Banking
Shocks to the U.S. Great Depression,” Federal Reserve Bank of Minneapolis Research
Department Staff Report 270.
Cole, H., and Ohanian, L. (2000b). “New Deal Polices and the Persistence of the Great
Depression: A General Equilibrium Analysis,” Federal Reserve Bank of Minneapolis Working Paper 597.
great depression in canada
71
Cooley, T., and Prescott, E. (1995). “Economic Growth and Business Cycles,” in Frontiers of
Business Cycle Research (T. Cooley, Ed.), Princeton, NJ: Princeton Univ. Press.
Cox, C. (1981). “Monopoly Explanations of the Great Depression and Public Policy Towards
Business,” in The Great Depression Revisited (K. Brunner, Ed.), Boston: Martinus Nijhoff.
Crucini, M., and Kahn, J. (1996).“Tariffs and Aggregate Economic Activity: Lessons from the
Great Depression,” Journal of Monetary Economics 38, 427–467.
Department of Labor, Ontario. Annual Reports, 1929–36.
Dighe, R. (1997). “Wage Rigidity in the Great Depression: Truth? Consequences?” Research
in Economic History 17, 85–134.
Haubrich, J. (1990).“Nonmonetary Effects of Financial Crisis: Lessons from the Great Depression in Canada,” Journal of Monetary Economics 25, 223–252.
Kendrick, J. W. (1961). Productivity Trends in the United States, Princeton, NJ: Princeton Univ.
Press (for NBER).
Labour Gazette, 1920–1940. Department of Labour, Canada.
Lucas, R., and Rapping, L. (1969). “Real Wages, Employment, and Inflation,” Journal of
Political Economy 77, 721–754.
McGrattan, E. (1999). “Predicting the Effects of Federal Reserve Policy in a Sticky-Price
Model: An Analytical Approach,” Federal Reserve Bank of Minneapolis Working Paper
598.
McGrattan, E. (2001). “Predicting the Effects of Federal Reserve Policy in a Sticky-Price
Model: An Analytical Approach,” unpublished Appendix to Federal Reserve Bank of
Minneapolis Working Paper 598.
Metcalfe, C., Redish, A., and Shearer, R. (1996). “New Estimates of the Canadian Money
Stock: 1871–1967,” University of British Columbia Discussion Paper 96-17.
Neufeld, E. P. (1972). The Financial System of Canada, Toronto: Macmillan.
Nixon, S. (1937). “Interest Rates in Canada: The Course of Interest Rates, 1929–1937,” Canadian Journal of Economics and Political Science 3, 421–434.
Ohanian, L. (2001). “Why Did Productivity Fall So Much During the Great Depression?”
Federal Reserve Bank of Minneapolis Staff Report 285.
Romer, C. (1999). “Why Did Prices Rise in the 1930s?” Journal of Economic History 59,
167–199.
Safarian, A. E. (1970). The Canadian Economy in the Great Depression, Toronto: Univ. of
Toronto Press.
Shearer, R., and Clark, C. (1984). “Canada and the Interwar Gold Standard, 1920–35: Monetary Policy without a Central Bank,” in A Retrospective on the Classical Gold Standard
(M. D. Bordo and A. J. Schwartz, Eds.), pp. 277–310, Chicago: Univ. of Chicago Press.
Siklos, P. (2000). “Understanding the Great Depression in the United States versus Canada,”
in World Economy and National Economies in the Interwar Slump (T. Balderstone, Ed.),
Macmillan Co.
Statistics Canada. Historical Statistics of Canada, http://www.statcan.ca/english/freepub/11-516XIE/sectiona/toc.htm.
Statistics Canada. (1988). National Income and Expenditure Accounts. Annual Estimates 1926–
1986, Ottawa: Canadian Govt. Publishing Center.
Struthers, J. (1983). No Fault of Their Own: Unemployment and the Canadian Welfare State
1914–1941, Toronto: Univ. of Toronto Press.
Urquhart, M. C., Ed. (1965). Historical Statistics of Canada, Toronto: Macmillan Co.
72
amaral and macgee
Urquhart, M. (1993). Gross National Product, Canada, 1870–1926: The Derivation of the Estimates, Kingston/Montreal: McGill–Queen’s Univ. Press.
U.S. Bureau of the Census. (1965). “Current Population Reports,” Series P-25, No. 311,
Washington, DC: U.S. Govt. Printing Office.
U.S. Department of Commerce. (1975). Historical Statistics of the United States: Colonial Times
to 1970, Part 2, Washington, DC: U.S. Govt. Printing Office.
Journal
of Monetary
Economics
25 (1990) 223-252.
North-Holland
NONMONETARY EFFECTS OF FINANCIAL CRISES
Lessons from the Great Depression in Canada
Joseph
G. HAUBRICH*
University of Pennsylvania, Philadelphia, PA 19104, USA
Received
March
1988, final version
received
January
1990
This paper empirically
examines the nonmonetary
effects of financial crises. It exploits the
special nature of the Great Depression
in Canada to test propositions
linking financial structure
and aggregate economic activity. Canada’s large domestic debt and shrinking banking system had
little impact on the course of the Depression.
In comparison
with results for the United States,
this implicates bank failures as a prime factor in the real effects of financial crises.
1. Introduction
Recent work examining the link between the financial system and macroeconomic
activity maintains
that bank failures and financial crises do not
simply respond to a collapsing economy, but instead have direct aggregate
effects beyond any impact on the money supply. This work [Bernanke (198311
has emphasized
the nonmonetary
causes, specifically bank crises, especially
in the paradigm case of the Great Depression.
It not only addresses contemporary concerns about bank safety but also has stimulated
rigorous analytical
development
[Gertler (1988>].
Bernanke
presents an important
alternative
to the traditional
monetarist
and Keynesian
explanations
of the Depression.
Friedman
and Schwartz
(1983, p. 357) maintain that the wave of bank failures ‘would have been only
a regrettable
occurrence
and not a catastrophe
if it had not been accompanied by a major decline in the stock of money’. Temin (1976, pp. 9-lo),
arguing the Keynesian
position, claims that ‘the banking panics were a part
of a larger process that started with the decline in autonomous
spending..
.
had they not taken place, the market would have been equilibrated
in other
ways but the overall story of the Depression
would not have been much
*Let me here thank the many people who have contributed
to this paper, including Mike
Bordo, Judith Farnbach,
Eugene White, the Penn Macro Lunch Group, an anonymous
referee,
and Ronald Shearer and Nancy Leaman, who also shared some important
data. Lillian Ng
provided first-rate research assistance, and the University of Pennsylvania
Research Foundation
and PARSS seminar provided financial support. Any errors remain mine.
0304.3932/90/$3.50
0 1990, Elsevier
Science
Publishers
B.V. (North-Holland)
224
J.G. Haubrich, Great Depression in Canada
different’.
According
to Bernanke
(1983, p. 2631, ‘the financial
crises of
1930-33
affected the macroeconomy
by reducing
the quality of certain
financial
services, primarily
credit intermediation’.
That is, losing a large
fraction of intermediaries
produces real effects: the cost of credit rises and
the ability to obtain funding to continue
or establish businesses
and farms
falls.
Extending
Bernanke’s
analysis to the Canadian
Depression
allows us a
natural
test of the cost of a credit model. In the U.S. - a unit banking
system - over one fifth of all banks failed; with failures, voluntary
liquidations, and mergers, one third of all banks disappeared.
Although
Canada’s
branch banking system proved immune to runs and panics, the number of
branches
dropped
from 4049 to 3640 between
1929 and 1933, loans and
deposits fell, and bank-stock
prices dropped. Thus, comparing
Canada with
the United
States provides an experiment
about the macroeconomics
of
financial
organization.
In particular,
it allows us to isolate the separate
components
of actual bank failures and a shrinking banking sector.
Studying Canada also eliminates
the need to account for some of the
extreme economic and political factors occurring in another natural comparison point, Europe. Hyperinflation,
Fascism, and Communism
were absent on
a national
scale in Canada. Likewise, it experienced
real shocks similar to
those in the United States. Chief among these was the agricultural
crisis or
‘Dust Bowl’. The province
of Saskatchewan,
for example, produced
321
million bushels of wheat in 1928, but only 37 million in 1937 [Lower (1957)].
This takes on added significance because agricultural
problems probably lay
behind many American bank closings [Temin (19761, Friedman
and Schwartz
(196311.
Finally, other authors have used comparisons
with Canada to make their
point about the U.S. Friedman
and Schwartz maintain
that ‘bank failures
were important
not primarily in their own right, but because of their indirect
effect’ (p. 352) by pointing out that Canada had a large monetary contraction,
no bank failures, and a depression
almost as severe as that in the United
States. Their student Morrison (1966) uses a similar comparison with Canada
to show that panics induced American banks to increase their excess reserves
during the Depression.
Bernanke
suggests that the severe Canadian
debt
crisis played a role similar to bank failures
in the U.S., but does not
acknowledge
that this credit crunch may be a transmission
mechanism
of
monetary policy.
More precisely, considering
the Canadian
experience
lets us test three
hypotheses.
First, it lets us test whether
the decline in bank operations
influenced
the course of the Depression.
A comparison
with the American
record can then differentiate
between the macroeconomic
consequences
of a
decline in operations
and outright failures. Secondly, we test the importance
of the nonbanking
‘debt crisis’ by looking at the effects of debt and commer-
225
J. G. Haubrich, Great Depression in Canada
cial failures. Finally, we test for the effects of money and thus obtain some
cross-country evidence on the monetarist explanation.
This paper is organized as follows. I begin with a short description of the
interwar financial system. Then to assess the role of the banking system in
the Depression, I estimate several regressions following Bernanke’s techniques using data for the entire interwar period (1919-1939). A further
section examines one aspect - domestic and international debt - that requires more detail. A final section draws some conclusions.
2. The Canadian
economy between the wars
Several features make the Canadian economy a natural testing ground for
the real effects of financial crises.’ A depression occurred, with falling output,
prices, and money supply, but in contrast to the U.S. no banks failed. On the
other hand, several differences create a need for caution in interpreting
results. The lack of a central bank, abandonment of the gold standard, and
large external debt provide pitfalls in drawing comparisons with the U.S.
The Canadian economy from 1919 to 1939 followed roughly the same
course as the U.S. economy.* After a very inflationary boom during WWI, the
economy steadily grew during the 1920’s, mostly due to growth in mining and
manufacturing; agricultural income, though large, did not expand rapidly.
After the peak in 1929 a severe depression set in, in part precipitated by a
poor harvest and collapse of export prices and quantities. The cycle reached
its trough in early 1933, with wheat and exports recovering before manufacturing. The recovery started rapidly, but then slowed down in early 1935,
giving Canada overall a slower total recovery than that of the United States.
1938 saw Canada also enter a short but steep recession prior to the World
War II driven recovery.
Figs. 1, 2, and 3 show the course of output, money supply, and prices over
the interwar period. The output measure, industrial production, distorts the
picture somewhat in that it excludes agriculture. Income from agriculture
accounted for almost a fifth of Canadian national income, and while national
income fell by 50% in the Depression, farm-sector income fell by 80%. It also
recovered more rapidly than investment or production.
Unlike the unit banking system of the United States, the branch system of
Canada proved relatively immune to runs or panics. The interwar period
showed a trend towards fewer and larger banks. In 1920 Canada had 18
banks, but by 1929 the number had fallen to 10. Acquisitions and mergers
accounted for most of the change. The only failure was that of the Home
‘This section
Clark (1984).
is based
2For more detail,
on discussions
consult
Safarian
in Beckhart
(1959).
(1929),
McIvor
(1958),
and Shearer
and
226
J.G. Haubrich, Great Depression in Canada
CIP
130 -
JJJJJJJJJJJJJJJJJJJJJJ
AAAA~AAAAAA~AARAAAAAAA
NNNNNNNNNNNNNNNNNNNNNN
1222222222233333333334
98123456789e123456789E
DRTE
Fig.1. Industrial
production,
Canada.
Bank in 1923. By the Depression,
numbers had settled down, with the only
merger occurring in 1931 and the chartered banks making their presence felt
by over 4000 branches throughout
the Dominion.
When failures did occur,
losses to note holders and depositors
were often small because bank stock
carried double liability.
Until 1935, Canada had no central bank. Despite this, the chartered banks
were heavily regulated. The Bank Act specified (among other things) audits,
capital requirements,
directors’ qualifications,
and loan restrictions,
including
a prohibition
against holding mortgages. These were held by loan companies,
which overall remained
small relative to the chartered
banks: in 1923 loan
companies
had assets of $188 million, compared
with bank assets of $2.5
billion [Beckhart (192911. Regulations
changed at regular intervals. Every ten
years the Bank Act was (and still is) revised and all banks had to renew their
charters. This was often the occasion for discussion, study, and debate about
the system, and amendments
frequently
changed the regulations.
Money in Canada took the form of chartered
bank notes (asset-backed
circulating
notes of the banks), Dominion
Notes put out by the government,
and most importantly,
deposits at the commercial
banks. Banks did not have
J.G. Haub~ch,
a0
Great depression in Canada
~,,,,,,,,,,,,,,,,,,,,,,,,,‘,,,‘,1,,,1,1,,,1
227
r
JJJJJJJJJJJJJJJJJJJJJJ
AAARRRARAAAAARRRA~AAAB
NNNNNNNNNNNNNNNNNNNNNN
1222222222233333333334
981234567B981234567B98
Fig. 2. Ml
Canada.
to hold reserves, but 40% of any reserves they chose to hold had to be
Dominion Notes.
Though lack of a central bank before 1935 left little room for discretionary
monetary policy, the government issued money and established important
rules for the moneta~ system. In theory, note issue (Dominion Notes) took
the form of a fixed fiduciary issue, above which all notes were backed 100%
by gold reserves. In practice this did not happen because the Finance Act of
1914 provided a major exception. Banks could obtain Dominion Notes
against collateral, which could be anything from Canadian Treasury bills to
feed stocks. The interest rate on this borrowing, known as the Advance Rate,
was set by Parliament, so while it was at the discretion of the government, it
changed only slowly. Banks requested a line of credit, above which they could
not borrow, but this was never a binding constraint. Banks once used 39% of
the line, but generally they used about 15% [Beckhart (19291, Shearer and
Clarke t 198411.
Other institutions somewhat compensated for the as yet nonexistent Bank
of Canada. The Canadian Bankers’ Association, a private group recognized
by the government, regulated clearing houses and created a fund to redeem
the notes of failed banks. The Dominion used the Bank of Montreal as its
228
J.G. Haubrich, Great Depression in Canada
PRICES
22E T
210 T
288 :
19u T
188 T
170 T
166 I
156 T
140 1
136 T
120 1
1lB T
100 1
90 1
se;,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
JJJJJJJJJJJJJJJJJJJJJJ
AAAAAAAAAAAAAAAAAAAAAA
NNNNNNNNNNNNNNNNNNNNNN
1222222222233333333334
9812345678961234567898
DATE
Fig. 3. Wholesale
price index.
fiscal agent. Canada also lacked a money market, but had easy access to New
York.
For most of the interwar period, except for July 1926 to January
1929,
Canada was not part of the international
gold standard. The rest of the time
the Dominion
was on either a de jure or de facto floating-exchange
regime.
For example,
in 1929, when still officially on gold, Canada
shifted the
gold-shipping
point from Montreal
to Ottawa, exported
only bullion and
Canadian coins (which had to be assayed before being accepted in the U.S.),
and instructed
banks to charge prohibitive
fees for handling gold. All pretense ended in 1931 when Canada left gold shortly after Britain.
3. Statistical
tests
This section tests the three competing
hypotheses
on the real effects of
financial structure. It empirically examines whether financial crises, either in
the banking
system or in the broader
debt market, help to explain the
Depression
after accounting
for the effects of money and other factors.
The first hypothesis is based on Bernanke
(1983) who argues that, in the
U.S., bank failures not only shrank the money supply, but also increased the
cost of credit by making the transfer
of funds from ultimate
lenders to
J. G. Haubtich, Great Depression in Canada
229
ultimate borrowers more difficult and more costly. Furthermore,
the extensive bankruptcies
among businesses made banks and others reluctant to lend
which further depressed the economy. Bernanke supports these claims empirically by showing that deposits of failed banks and liabilities
of failing
businesses
had large and statistically significant effects on industrial production, even after accounting for lagged values of income, unanticipated
money,
and unanticipated
prices.
The cost of credit hypothesis has two parts: increasing commercial
failures
crippled lending in the debt market and banking problems exacerbated
the
downturn.
Since no banks failed in Canada, explicitly comparing the banking
results requires a joint hypothesis:
the decline in bank operations
increases
the cost of credit as do bank failures and a higher cost of credit has real
effects. Rejecting the hypothesis may mean that bank failures have a different
impact from declining operations.
Empirically, we will look for significance of
commercial
failures and of the banking variable.
The monetarist
explanation
also has a straightforward
prediction.
Namely,
that the money supply will be significant. Furthermore,
the effects of banks
and debt should not be significant
once money is accounted
for in the
regressions.
BRANCHES
4800
I
4700
T
4600 1
4500 1
4400 1
4308 I
4200
T
4100
I
4600
1
3900
1
3900
T
3700 1
3600 1
3500
1
3466
1
3300 -/
,
/ , , ,,
,, /, ,
/ ,
, , ,, , / ,
JJJJJJJJJJJJJJJJJJJJJJ
AAAAAAAARAAAAAAAAAAAAA
NNNNNNNNNNNNNNNNNNNNNN
1222222222233333333334
98123456789B1234567898
DATE
Fig. 4. Number
of branches.
,, , , ,
,
, , , , , ;,
230
J.G. Haubrich, Great Depression in Canada
INDEX
1.6
1.5
1.4
1.3
1.2
1.1
1.B
8.9
E.FJ
0.7
8.6
6.5
0.4
0.3
11.2
U.1
11.8
JJJJJJJJJJJJJJJJJJJJJJ
AARRRARAARAAAAAAAAAAAA
NNNNNNNNNNNNNNNNNNNNNN
1222222222233333333334
9012345678981234567899
DATE
Fig. 5. Bank stock deflated
by stock index.
The Keynesian
hypothesis
favored by Temin suggests that banks and
commercial
failures merely reacted to the downturn
and denies importance
to money. It has other, more positive implications,
but those will not be
tested here, and the theory thus remains
one interpretation
of the null
hypothesis:
money, banks, and commercial
failures had no effect on the
course of the Depression.
Without
bank failures, I must use other proxies for the cost of credit.
Counting
branches
measures
one aspect of credit availability
and financial
services. Another aspect is bank-stock
prices which measure the health and
profitability
of the Canadian
banking sector. This measure has the flaw of
being indirect, but the advantage
of indicating
the perceived value of the
banks. Though neither measure
is ideal, the two together
should paint a
better picture than either one alone. Figs. 4 and 5 plot the number
of
branches and an index of bank-stock
prices.
Without a single perfect variable or agreed-upon
model of income determination, I must look at several different specifications
of the problem. Along
with using various proxies for the cost of bank credit, I include measures of
the money supply and U.S. production
to control for these important
effects.
Since this is a reduced-form
study, I specify the regressions first in levels and
231
J. G. Haubrich, Great Depression in Canada
then (following Bernanke more closely) in first differences. Later, Chow tests
and vector autoregressions
help check the robustness
of the results with
regard to stationarity
and structural biases3x4
3.1. Regressions in level form
The first set of regressions
are variations on a theme in level form. Each
regression, after controlling
for the impact on industrial production
of a time
trend, the money supply, and the single bank failure. I use three proxies for
the cost of credit: number
of branches,
price of bank stock, and spread
between
commercial
and Dominion
bond yields. A further variation
adds
U.S. industrial
production
to control for U.S. influence. Table 1 reports the
results.
The first regression regresses seasonally adjusted industrial
production
on
a time trend, four lags of industrial
production,
money and four lags, a
dummy for the Home Bank failure in August 1923, and the number
of
branches and four lags:
4
IP, = (Y + (YET+ C piIPt_i + 5 y;MOA’_i + diBANmAILt
i=l
+ 5
i= 0
6,BR,_,.
(1)
i=O
Serial
correlation
3As might
exponentially
is not a problem,
be expected,
the R2’s differ
detrended
variables.
measured
substantially
either
between
by the Durbin-Watson
levels and rates
of growth
of
4The data come from various sources. The income measure is monthly values for industrial
production,
from Brecher and Riessman (1957). As mentioned
above, these numbers unfortunately give a somewhat distorted view of the Great Depression
in Canada because they exclude
agricultural
production
[see Safarian (1959)]. Wholesale prices come from the same source. For
the money supply I use Ml, which is deposits, Dominion Notes in circulation,
and notes of other
banks in circulation.
The data were thoughtfully
provided by Ronald Shearer, who extended the
data beyond Curtis (1931), and from the Monthly Review of BusinessStatistics (assorted issues,
1931-39). The date for placing the dummy on the Home Bank failure in August 1923 comes
from Beckhart (1929). The monthly data on branches are from Houston’s Monthly Bank Directory
of Canada and Newfoundland. The three missing values were linearly interpolated.
The stockprice index is an index of the 8 largest banks, and it and the market index of 130 firms come from
various issues of the Dominion Bureau of Statistics’ Monthly Review of Business Statistics and its
November
1931 supplement
as do the data on real estate and mortgage
loans. The data for
liabilities of failed businesses come from the interwar editions of Dun’s Review. Data on interest
rates and yields come from Nixon (1937) and bond ratings from Moody’s Manual of Investments,
both Industrial Securities and Public UtilitySecurities. The regressions were run using SAS, with
seasonally adjusted data. Those variables with a seasonal component
were de-seasonalized
using
the SAS X11 procedure.
232
J.G. Haubrich, Great Depression in Canada
statistic or by direct estimations
of an autoregressive
error process.5 Thus,
even with lagged dependent
variables, the least-squares
estimates are consistent and asymptotically
efficient [Theil (1971, sect. 8.711. Four lags of industrial production
emerge as significant,
as does the contemporaneous
money
supply. Also notice that the bank-failure
dummy is insignificant.
Furthermore, total branches and its first and third lags have the wrong sign. Only the
second and fourth lags have the expected positive sign.
I also compute
F-statistics
testing whether
the sum of the coefficients
equals zero. These do not always support the t-tests. The tests concur on
industrial production:
the sum of the coefficients is very significantly different
from zero. However, the F-tests for money and branches
conflict with the
t-tests. The F-test on the sum of money coefficients
shows them insignificantly different from zero. The branch sum differs from zero at the 5% level.
Thus, the effects of both money and branches
are of somewhat ambiguous
significance.
Still, since eq. (1) includes four lags of industrial
production,
money and branches
do more than proxy for income. Causality is always
difficult to determine,
but the F-test on branches
may indicate something
beyond the normal pro-cyclical movement of the financial sector.
Regression
(21 adds lagged U.S. industrial production.
This specification
is
IPt=a+a,T+wUSIP,~,
+ d,BANKEAIL,
+ ~ piZPI_I + ~ y,MON,-;
i= I
i=O
+ 5 &BR,_,.
i=O
(2)
The U.S. index emerges as statistically and economically
very significant. The
branch coefficients
remain insignificant,
and the significant
money terms
remain, though with reduced significance. The sum of the money coefficients
and the sum of the branch coefficients
are not significantly
different from
zero. In a separate equation (not reported here), with 14 lags of money and 5
lags of U.S. industrial
production,
only the yearly (12th) lag on money was
significant, and only U.S. industrial production
was significant, so the results
appear robust to the lag specifications.
‘Since the Durbin-Watson
statistic means little when the regression
contains lagged dependent variables,
1 report
instead
Durbin’s
h [Durbin
(1970)]. Because
this indicated
some
autocorrelation,
the equations were re-estimated
using the AUTOREG
procedure
in SAS, which
uses a two-step full-transform
method of estimating serial correlation.
The estimated coefficients
and the standard errors changed little, and the estimates for the autoregressive
parameters
were
small and insignificant.
Those results are not reported here. Also, since the regressions
in levels
agree with the regressions
in differences, the model is most likely correctly specified [Plosser and
Schwert (1978)I.
Table
Regression
Dependent
variable
Sample period
Independent
(1)
IP
1919-39
TIME
(2)
IP
1919-39
(3)
IP
1919-39
(4)
IP
1929-37
variables
USIPI
CONST
l9
16.6060’
(2.063)
- 0.0086
(-0.867)
0.268gh
(3.490)
0.2593
(3.515)
4.593
(0.535)
1.499
(0.238)
0.0129
(1.123)
0.0169’
(2.254)
0.0249
(0.8811
0.4903b
(3.295)
- 9.773
(0.577)
IPI
0.5614h
(8.609)
0.4877’
(7.269)
0.4623’
(6.933)
0.3449h
(3.105)
IP2
0.1213“
(1.702)
0.0882
(1.2561
0.1073
(1.543)
0.0729
(0.671)
IP.?
0.3304b
(4.694)
0.3149h
(4.572)
0.3052’
(4.424)
0.3810h
(3.587)
IP4
-0.1127’
(1.770)
- 0.0995
(1.597)
- 0.0779
(1.243)
-0.1198
(1.2301
MONO
0.0327”
(2.195)
0.0281’
(1.923)
0.0190
(1.335)
0.0048
(0.250)
MONI
0.0029
(0.154)
0.0010
(0.0551
0.0041
(0.224)
0.0152
(0.673)
MON2
-0.0133
(0.6961
-0.0169
to.9031
-0.0173
(0.918)
- 0.0383
(1.627)
MON3
0.0148
(0.7851
0.0144
(0.7811
0.0135
(0.7271
0.0172
(0.736)
MON4
-0.0241
(1.6171
- 0.0189
(1.287)
- 0.0153
(I.0521
0.0115
to.0551
BA NKFAIL
- 2.056
(0.470)
-2.281
(0.5341
- 2.535
(0.6041
BRO
- 0.0002
(0.006)
BRI
- 0.040
(0.988)
- 0.0398
(1.003)
ABSI
6.145
(0.748)
SPRDl
1.768
(0.769)
BR2
0.0592
(1.455)
0.0541
(1.361)
ABS2
- 1.137
(0.138)
SPRDZ
- 1.133
(0.486)
BR3
- 0.0501
(1.2381
-0.0516
(1.305)
ABS3
2.063
(0.251)
SPRD3
3.351
(1.454)
BR4
0.027
(1.086)
ABS4
5.517
(0.892)
SPRD4
- 0.235
(0.127)
Adj. R2
Durbin h
0.95
n.d.d
0.0118
CO.4461
0.024 1
(0.9931
0.95
n.d.
ABS
- 14.25”
(2.394)
0.95
nd.
SPRD
- 1.402
(0.767)
0.94
n.d.
“All data are monthly. IPn is the nth lag of seasonally adjusted Index of Industrial Production,
Canada.
MONn denotes seasonally
adjusted
MI and its lags. BANKFAIL
is a dummy in the
month of the Home Bank failure. BRn denotes the number of branches and its lags. ABSi is
adjusted bank-stock
price and lags. SPRDi is the interest rate spread and lags. Absolute values
of I-statistics are given in parentheses.
‘Significant
at 5% level.
‘Significant
at 10% level.
dNot defined.
J.G. ~al~b~c~, Great Depression in Canada
234
Using the bank-stock price index broadly confirms these results. However,
this measure does not indicate causality, since the stock market is a leading
indicator of the business cycle, and bank-stock prices participate in this
movement. To control for this, the regressions use the bank-stock index
deflated by the DBS’s market index of 130 stocks obtained from the Monthly
Review of Business Statistics. In fact fig. 5 shows that, when deflated by the
market index, bank-stock prices appear counter-~clicai. Thus during a recession bank stocks (and thus perhaps banks) do well relative to other industries.
The regressions took the form of
IP, =
LY +
aT + wUSIP,_, + 5 piCIp,-i + i
i=l
+ diBANKEAIL,
Y;MON,-;
i=O
+ ; G,ABS,_i.
i=@
(3)
This adjustment does differ from the previous resuhs. Though the contemporaneous adjusted stock price is significant, it has an unexpected negative sign.
F-tests show neither stocks nor money significant.
To more directly test the effects of a debt crisis or the nonbank aspects of a
general financial crisis, regression (4) - not written out here - uses the yield
spread between an index’ of commercia1 bonds and Dominion bonds as a
measure of financial distress. Fig. 6 plots this spread for the avaiiable years
1929-1937. No financial or monetary indicators have any import, either singIy
or jointly.
3.2. Regressions using rates of change
Even tests that follow Bernanke more closely by using rates of change
provide littIe for the cost of credit or monetarist hypotheses. This method
also controIs for potential spurious correlation resulting from the time-series
properties of the variables. The regressions run here for Canada differ from
Bernanke’s in several ways. First, Bernanke uses deposits; I use differences of
the number of branches and stock prices. Second, Bernanke uses unanticipated money and prices; I use growth rates for several reasons. (a) The
equations predicting money growth did poorly; most money growth was
unanticipated anyway. (b) Estimating and interpreting two-stage models
involving expectations [see Mishkin (1983), Murphy and Topel (19851, Newey
(198411 adds many complications. (c) The current framework allows easier
comparison with other works not using the unanticipated-money
hypothesis.
‘Nixon
(1937) constructed
this index. Moody’s
(1929-37)
rates these firms from Aaa to B.
J.G. Haubrich, Great Depression in Canada
235
PERCENT
31
B_,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
FEE38
NAY35
WC32
NOW29
FEE27
DATE
Fig. 0. Spread
between
commercial
and Dominion
bonds.
This is important
since the money-surprise
hypothesis
has come under
increasing attack [McCallum (1982), Cecchetti (198611.’ Third, my regressions
also include the levels of commercial
failures (liabilities of failed businesses),
not the difference,
because commercial
failures are already a difference.
Finally, the regressions
were estimated
separai.ely including
the growth of
industrial production
in the United States.
The regressions using rates follow regressions (l)-(4) in general form. The
results are reported
in tables 2 and 3. In the first specification,
industrial
production
is regressed
on two of its own lags, current
U.S. industrial
production
and one lag, growth rate of money and three lags, bank branches
and one lag, and contemporaneous
commercial
-failures and three lags:
If’G,
=
i
i=l
CY,IPG,_, +
5 p,kG_; +i
i=o
i=o
yiBRG,-; + i
I
hscF,-;.
=o
(5)
‘Earlier formulations
not correct the standard
than the following.
of these regressions which used unanticipated
money and prices but did
errors, showed less influence of money and more influence of branches
Table 2”
Regression
Dependent
variable
Sample period
(5)
IPG
1919-39
(6)
IPG
1929-39
(7)
IPG
1919-39
IPGl
-0.316b
(4.995)
- 0.3246h
(5.142)
- 0.0532
(0.545)
IPG2
-0.237h
(3.759)
-0.2198h
(3.534)
Independent
variables
-
0.0786
(0.821)
MG
0.0029’
(2.129)
0.0021
(1.468)
0.0037h
(2.463)
MCI
0.0025’
(1.786)
0.0023
(1.584)
0.0015
(1.017)
MG2
0.0025“
(1.753)
0.0026’
(1.823)
0.0021
(1.316)
MC3
0.003 lh
(2.168)
0.0027’
(1.903)
0.0039h
(2.744)
BRG
1.1217
(0.879)
0.0003
(1.245)
0.0005
(1.420)
BRGI
- 1.758
(1.481)
SCF
0.0003
(1.318)
SCFI
- 0.0004’
(1.657)
- 0.0004
(1.668)
SCF2
0.0003
(1.295)
0.0003
(1.442)
-0.0001
(0.330)
SCF3
- 0.0001
(0.582)
- 0.0001
(0.576)
0.0002
(0.757)
- 0.0007h
(1.985)
LNG
0.0896
(0.313)
LNGl
0.4452
(1.514)
LNG2
*
0.649
(0.228)
LNG3
0.0029
(0.010)
RABS
-0.2016h
RABSI
- 0.0991
(1.147)
(2.412)
Adj. R2
Durbin h
0.13
0.99
0.14
4.94
0.09
n.d.d
“All data are monthly. IPGi denotes the ith lag of the Index of Industrial Production,
growth
rate of seasonally
adjusted, exponentially
detrended.
MGi denotes money growth and its lags.
LNGi is loan growth and its lags. SCFi is the value of commercial
failures. BRGi denotes the
rate of growth of branches.
RABSi denotes the rate of growth of the adjusted bank-stock index.
f-statistics are given in parentheses,
‘Significant
at 10% level.
‘Significant
at 5% level.
“Not defined.
J. G. Haubrich, Great Depression in Canada
237
Table 3”
_~
Regression
Dependent
variable
Sample period
(6’)
(5’)
IPG
1919-39
IPG
1929-39
(7’)
IPG
1919-39
IPGI
- 0.3969h
(6.174)
- 0.4053h
(6.399)
-0.1195
(1.177)
IPG2
- 0.2920h
(4.676)
- 0.27gh
(4.592)
Independent
variables
0.0456
(0.476)
USIPI
0.461 lh
(4.113)
0.5044”
(4.476)
0.1 998h
(2.009)
MG
0.0025’
(1.870)
0.0018
(1.304)
0.0032h
(2.127)
MGI
0.0013
(0.955)
0.0012
(0.881)
0.0007
(0.450)
MC2
0.0010
(0.719)
0.0009
(0.659)
0.0009
(0.541)
MG3
0.0024’
(1.724)
0.002
(1.440)
0.0034”
cz.385)
BRG
1.339
(1.084)
RABS
- 0.2032h
(2.529)
BRGI
- 1.625
(1.415)
RABSI
- 0.0281
(0.332)
SCF
0.0004’
(1.741)
0.0004
(1.619)
0.0005
(1.592)
SCFl
- 0.0003
(1.234)
- o.ooll3
(1.147)
- 0.0006c
(1.808)
SCF2
0.0003
(1.078)
0.0003
(1.215)
- 0.0001
(0.417)
SCF3
- 0.0003
(1.219)
- 0.0003
(1.327)
0.0002
(0.493)
LNG
0.1036
(0.368)
LNGl
0.3749
(1.284)
LNG2
0.0573
(0.204)
LNG3
0.0189
(0.067)
Adj. R’
Durbin h
“See footnotes
0.23
n.d.d
to table 2. USIPI is 1st lag of growth
0.20
0.77
rate of USIP.
0.1 I
n.d.
238
J.G. Haubrich, Great Depression in Canada
This reconfirms
the earlier regressions.
Money growth and its three lags
appear significant. Branches and lagged branches are not significant, though
they show the same sign pattern
as before. Commercial
failures appear
insignificant
except for the first lag. F-tests on the sum of the coefficients for
branches and commercial failures cannot reject the null hypothesis that these
sums are zero. Money growth has a significant sum, however.
The second specification
replaces branch growth with the bank-stock index
adjusted by the market index:
IPG, = ; a,IPGr_,
i=l
+ i P,MG[_;
i-o
+ i
Y;SCF,_~ + i
i= 0
G;RABS,_;.
i=O
(6)
The basic results (including
the results of the F-tests) hold. As before, the
index is significant
at the 5% level, and several lags of money and prices
appear significant. The first lag of commercial
failures is also significant at
the 5% level when run with prices. Tests on the sum of coefficients
again
bear this out: the sum over bank stock differs significantly
from zero, as do
money and prices, but commercial
failures do not.
Regression (7) adds the growth in mortgages and real estate loans made in
1929- 1939:
IPG,=
&x,IPG,_,+
,=I
;p,MG,_,
i =o
+ &,SCF,
i=o
-; + ; y,LNG,_;.
i=o
(7)
Adding this direct measure of debt has little impact on the results. Monetary
variables remain significant, as does the first lag of commercial
failures. The
debt measure is not significant,
and an F-test on both commercial
failures
and debt suggests joint insignificance.
The regressions
in level form showed appreciable
changes when U.S.
industrial production
entered the equation. To account for the U.S. influence
on the growth-rate
equations,
I added the growth rate of exponentially
detrended
U.S. industrial
production
in regression
(5’1, (6’1, and (7’) (not
written out to conserve space). As table 3 shows, the differences
between
regressions
(5)-(7)
and regressions
(.5’)-(7’) echo but do not match the
differences between level-form regressions (1) and (2). Though U.S. industrial
production
again dominates
the equations,
money remains significant.
Regression (7’) has no individually
significant
monetary
terms, but their sum
differs from zero at the 10% level. In two of the three regressions, contempo-
J. G. Haubrich,
Great Depression in Canadu
239
raneous levels of commercial failures show significance. In no regression does
the F-test on coefficient sums reject a zero value for stock prices, debt, or
commercial
failures.
This extension of Bernanke’s
analysis to Canada points to several conclusions. Most importantly,
the cost of credit does not have a major impact on
the course of depression.
Neither bank variables nor debt variables emerge as
significant. Less definitely, money also seems to have little impact on output.
3.3. Robustness tests
The reduced-form
time-series
evidence presented
here is subject to two
serious complaints:
that the regressions
are not stable over time and that
simultaneity
problems
confound
any causality results. I examine these stationarity and structural
issues with Chow tests and vector autoregressions,
confirming the robustness of the above results. Some interpretations,
particularly on the role of money, must be modified in light of structural questions,
however.
To test for coefficient stability, I performed
Chow tests on the regressions.
This was particularly
important
for the branch variable which declined in the
1920’s, a period of high growth. In particular,
some authors feel the boom of
the 1920’s led to many bank mergers. The former competitors
then closed
down rival branches
in many towns. Since the merger process was largely
complete
by 1927, this might reverse the relation
between
branches
and
output. I checked for stability before and after January 1927.’ In the interests
of space, those regressions
are not reported
here. The results varied. In
general, the regressions in levels showed coefficient instability, while those in
growth rates did not. Splitting the sample did not appreciably
change the
results so far reported,
as the significance
of branches,
stock prices, and
commercial
failures did not change. In particular,
the huge build-up
and
decline in branches following World War I thus does not explain the lack of
correlation
between
branches
and economic
activity. Money tended to be
more important
in the 1927-1939 period than in 1919-1926 as regressions (6)
and (6’) suggest.
Like most work in this area, this paper relies heavily on reduced-form
analysis. But a long tradition
in economics makes us particularly
suspicious
about reduced-form
results involving money or the international
transmission
of the business cycle. Neither the money supply nor U.S. industrial
production is obviously exogenous in cl)-(7’). I test and find U.S. industrial produc-
‘F-tests
showed that
plausible the assumption
lO)l.
the split-sample
regressions
had the same error variance,
making
of homoscedasticity
needed for the Chow test [see Maddala (1977, ch.
J.G. Haubrich,
240
Great Depression in Canada
Table 4
&anger
causality
Money and Industrial
Specification:
F-stat.
F-stat.
constant,
all MON = 0
all IP = 0
U.S. and Canadian
Specification:
all CIP = 0
all USIP = 0
F-stat.
F-stat.
“Monthly
constant.
data,
1919-1939,
tests.”
Production
time trend,
four lags
2.50
2.03
Industrial
Prob. F > 0
Prob. F > 0
0.04
0.09
Production
time trend,
1.42
9.56
four lags
Prob. F > 0
Prob. F > 0
0.23
0.0001
four lags used
tion is exogenous,
but the endogeneity
of the money supply forces me to
qualify my conclusions
somewhat.
An endogenous
money supply probably lies behind the unexpectedly
low
significance
of money. 1 attribute
this to the monetary
system. Without
a
central bank, the money supply was demand-determined,
and high-powered
money responded
to banks’ advances under the Finance
Act [Courchene
(1969). Shearer and Clarke (198411. Table 4 supports this by reporting
the
results of a Granger causality test between money and income. The results
show bi-directional
causality, with both money and output influencing
each
other.
The small effect of money, though interesting,
should not be taken as a
counterfactual.
With the money supply demand-determined,
money should
not affect output. To make the point in a simpler context, consider the basic
IS-LM model. If the government
pegs the interest
rate, resulting
in a
horizontal LM curve, all variations in output come from shifts in the IS curve.
However, that does not imply money has no impact when a different regime
starts shifting the LM curve. Although monetary policy may not have caused
the Canadian
depression,
a counter-cyclical
monetary
policy might have
prevented
it.
A different structural
concern arises in an economy as open to the rest of
the world as Canada’s.
Accounting
for foreign influences
on the business
cycle becomes both necessary and difficult. The ideal structural model would
account for transmission
via exports (tourism,
forestry products),
imports,
investment,
technology
transfer,
and what...
Purchase answer to see full
attachment