it f \ *
UB&AItIB8
Vr«^
Digitized by the Internet Archive
in 2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/whywasstockmarkeOOvoth
DEWEY
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
Why Was Stock Market Volatility So High
During the Great Depression? Evidence
from 10 Countries during the Interwar Period
Hans-Joachim Voth
Working Paper 02-09
February 2002
Room E52-251
50 Memorial Drive
Cambridge, MA 02142
This paper can be downloaded witiiout charge from the
Social Science Research Network Paper Collection at
http://papers.ssm.com/paper.taf?abstract_id=xxxxx
^\
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
Why Was Stock Market Volatility So High
During the Great Depression? Evidence
from 10 Countries during the Interwar Period
Hans-Joachim Voth
Working Paper 02-09
February 2002
Room E52-251
50 Memorial Drive
Cambridge, MA 02142
This paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
http://papers.ssm.com/paper.taf?abstract_id=xxxxx
0FTECHM0L06Y
iVIAR 1 5 2002
LIBRARIES
Why Was Stock Market Volatility So High During the Great
Depression?
Evidence from 10 Countries during the Interwar Period
Hans-Joachim Voth
Economics Department, MIT
Departament d'Economia, UPF
12.2.2002
Abstract:
The extreme levels of stock price volatility found during the Great
Depression have often been attributed to political uncertainty. This
paper performs an exphcit test of the Merton/Schwert hypothesis
that doubts about the survival of the capitahst system were partly
responsible. It does so by using a panel data set on poHtical unrest,
demonstrations and other indicators of instability in a set of 10
developed countries during the interwar period. Fear of worker
mUitancy and a possible revolution can explain a substantial part of
the increase in stock market volatility during the Great Depression.
JEL: G12, G14, G18, E66, N22, N24, N12, N14
Keywords: Stock price volatility, political uncertainty, worker militancy.
Great Depression.
During the Great Depression, aggregate stock market volatility in a large
number of advanced economies reached levels not seen before or since.
Schwert (1989b) estimates that in the US, there was a two- to threefold
increase in variability. According to his measure, the monthly variation of
stock returns peaked at over 20 percent in 1932. Other developed countries
experienced similar increases in volatility. This is all the more puzzling since
macroeconomic series such as money growth and interest rates showed
markedly smaller increases in variability (Schwert 1989b). As a general rule,
neither wars nor periods of financial panic appear to lead to significantly
higher variability of equity returns over an extended period — despite the
highly unstable behavior of other macroeconomic series. Recessions, however,
are clearly associated with higher volatility (^chwert 1989a). Stock returns
and their volatility in general show only a tenuous link with fundamentals
(Cutler, Poterba and Summers 1989), even if uncertainty about these
fundamentals can in part explain variability (David and Veronesi 2001).
Why was stock market volatility in the US so much higher during the
Great Depression than at any time before or since? In his seminal paper,
Schwert (1989) concludes that there is a "volatility puzzle". Because all other
likely explanations are insufficient, the most likely one is that the very
survival of the capitalist system, even in the United States, was seen to be at
risk. As Robert Merton has pointed out, the Russian Revolution occurred
little more than a decade earlier. In the case of a communist take-over, for
example, private ownership of the means of production would have come to
an end. Even relatively small changes in the probability of a momentous
shock like this might lead to extreme swings in market sentiment occurred.
This suggests that examinations of stock volatility may be affected by a
particular form of the "Peso problem". Some economists observing extreme
swings in stock prices ex post have conclude that there is no rational
explanation for them (Schiller 1981). ^ If possible regime switches that
ultimately failed to materialize were partly responsible, this would be
erroneous gchwert 1989b).2 As Schwert (1989b, 1146) argued: "With the
benefit of hindsight, we know that the U.S. and world economies came out of
the Depression quite well. At the time, however, investors could not have had
such confident expectations." The argument that political risk during the
Great Depression is partly to blame is supported by the recent finding that
unusually high levels of synchronicity of individual stock returns contributed
substantially to aggregate volatility (Morck, Yeung and Yu 2000). ^
1 Cf. the critique in Kleidon 1986.
2 Note that this is similar to the standard problem in bubble tests. Cf. Hamilton and
Whiteman 1985, Hamilton 1986.
3 They also demonstrate that lower synchronicity is systematically associated with "better
government" (defined as a composite measure of the risk of expropriation, government
corruption, and the risk of the government repudiating contracts).
This paper adopts a simple strategy to test the Schwert/Merton
hypothesis empirically. We use a data set on political risk and stock price
variability in a group of 10 countries during the interwar period, 1919-1939.
If fear of a collapse of capitalism was to blame for the extreme stock volatility
in the US, countries facing a higher probability of communist takeover or
other severe disruptions of the civic and legal order should have experienced
particularly large equity return volatility. Our data set, which contains a
number of relatively advanced. countries from Europe (Germany, France,
Sweden, Italy, UK, Netherlands, Belgium, Norway, and Switzerland) plus the
US is useful in testing this proposition. While some of these nations — such as
Germany, France, and the UK - went through extreme social upheavals and
political turmoil, others such as Switzerland were largely unaffected. If the
volatility of stock markets increased in response to mounting challenges to
the capitalist order, we should find systematic associations in our panel both
in the cross-sections and within each country over time. In view of the recent
literature on the political economy of democratization, the 1920s and 1930s
are also a particularly useful period to study. Acemoglu and Robinson argue
that, over the last 200 years, extending the franchise has effectively been a
way for capital owners to commit credibly to future redistribution (Acemoglu
and Robinson 1999, 2000). If this is true, then any challenges by disaffected
workers should be much more threatening once universal suffrage has been
granted, and the 'ruling classes' have run out of 'franchise cards' to play.
Since most countries had more or less completed the process of giving the
vote to the lower classes by the end of World War II, credible promises of
future redistribution became increasingly hard to make within the existing
political and social order.
The exercise is similar in spirit to recent work on interwar Germany
(Bittlingmayer 1998) and on emerging markets (Pekaert and Harvey 1997,
Mei 1999). Bittlingmayer argues that the extreme levels of volatility in
Germany during the early 1920s are driven by exogenous political events,
such as the revolution of 1918/19, the Hitler putsch in Munich, and the
French invasion of the Ruhr. Bekaert and Harvey show that country credit
ratings based on surveys of business men are weakly associated with stock
market volatility, and Mei argues that stock prices become less stable during
elections. While Bittlingmayer presents no systematic test of the connection
between political instability and stock return variability, Bekaert and Harvey
only find a small effect from political risk. Also, their variable is - as they
admit - a composite measure of political and macroeconomic uncertainty
(Bekaert and Harvey 1997).
There is a voluminous literature on the determinants of revolutions
and their relation to demographic, economic and social conditions, with
contributions from sociologists and economists PeFronzo 1991; Goldstone
1991; Goldstone and Merton 1986; Grossman 1999). While the interactions
are far more complicated than a simple immiserization model would predict —
with economic distress leading to revolutionary bids for power — inequality
and instability appear reliably associated (f\lesina and Perotti 1996, Muller
and Seligson 1987). There is also some indication that revolutions are
significantly more likely during recessions, when opportunity costs are
relatively low (Acemoglu and Robinson 1999, 2000; Gasiorowski 1995;
Prezworski et al. 1996). There are therefore strong reasons to believe that the
Great Depression should have been a good period for revolutionaries, and
that this realization concerned contemporaries. The slump was protracted
and led to unprecedented levels of unemployment. In countries where the
1920s had seen great increases in prosperity, inequality had reached extreme
levels (Galbraith 1962).
Our panel data set does not contain information on the threat of
communist takeover and revolution itself. Instead, we use a number of
variables that could reasonably be expected to help contemporaries gauge the
strength of workers militancy and the dangers to the established economic
and legal system. These include the number of general strikes, of riots and
anti-government demonstrations, of violent attempts to overthrow the
established order, as well as indicators of the stability of governments.
I find that these political indicators can help to explain the history of
stock market volatility in the interwar period. After controlling for
macroeconomic sources of variability, many - but by no means all —
indicators of worker militancy and left-wing radicalism led to significant and
large swings in the value of equities. Also, crack-downs on the opposition and
purges clearly helped to stabilize expectations, leading to lower volatility.
Periods of unstable government also appear to be weakly associated with
greater volatility.
I. Data
The stock indices in this study are similar to the set employed by Jorion and
Goetzmann (1999), and made available through Global Financial Data.^ They
are all broad market indices, relative to the size of the domestic equity
market that they represent. In most cases GFD has attempted to reconstruct
the equivalent of commonly used indices such as the S&P-500 for more
distant periods in the past (details in the data appendix). All series were
deflated by the consumer price index. Despite these broad similarities, some
differences should be noted. The number of shares varies considerably — the
Norwegian stock index is modeled on the OBX-25, containing the 25 largest
stocks by market capitalization, whereas the British and Dutch series
represent all-share indices. Differences in the composition of indices (and the
relative concentration of capitalizations) can have considerable influence on
aggregate measures of volatility (Bekaert and Harvey 1997). In the empirical
4 The Jorion and Goetzmann dataset is not publicly available.
part of the paper, we will try to adjust for this by using fixed-effect
regressions.
Average share prices could swing wildly - in June 1923, the German
index gained 61 percent, only to lose 52 percent in August. By far the highest
level of average volatility is recorded for Germany, which during the years
1919-39 shows a yearly standard deviation of monthly of 10.3 percent.
Belgium and the US are markedly more stable, with average volatility of 7
percent. At the opposite end of the spectrum, the UK and Norway recorded
standard deviations of only 3.1 and 2.6 percent.
All the countries in our sample show higher than average levels of
volatility during the Great Depression, with one notable exception. Germany
saw the highest standard deviation of monthly returns during 1923, when the
hyperinflation reached fever pitch, the French invaded the Ruhr, and the
country was fighting for its survival as a nation state (Feldman 1993).
1931/1932 are by far the most common years for maximum variability of
share prices - eight out of our ten countries see the peak in equity volatility
in one of these two years. Maximum volatility was again highest in Germany,
both in absolute terms and relative to the average for the country during the
period 1919-1938 as a whole. In 1923, the standard deviation was more than
four times higher than normal, reaching 43.5 percent. In absolute terms, the
US, Sweden and Belgium recorded relatively high levels of variability.
Relative to average share price volatility, a broadly similar ranking emerges.
In six out of ten countries, the standard deviation more than doubled, led by
Germany, Sweden, the US and the UK. In Belgium, on the other hand,
volatility in 1931 rose by only half Table I also presents the statistics on
skewness and kurtosis. Jarque-Bera tests (not reported) demonstrate that, in
each case, the null of normality can be rejected.
Table I
Real stock returns in 10 countries, 1919-1938
Continuously compounded monthly returns and measures of volatility, based on monthly
returns. The standard deviation is calculated on the basis of monthly returns for each year.
For details of the data, cf. the Data Appendix.
average
largest
largest
average
highest
year of
ratio
skew-
kur-
volatility
monthly
monthly
annual
volatility
highest
max/
ness
tosis
gain
loss
return
volatility
average
Germany
0.103
0.61
-0.52
0.055
0.435
1923
4.22
0.07
4.92
UK
0.031
0.11
-0.11
0.018
0.064
1931
2.08
-0.46
1.39
Belgium
0.070
0.27
-0.17
-0.064
0.105
1931
1.50
0.59
1,08
USA
0.071
0.35
-0.35
0.037
0.182
1932
2.58
-0.13
4.54
France
0.057
0.20
-0.18
-0.027
0.095
1936
1.67
0.21
0.62
Italy
0.050
0.24
-0.21
-0.047
0.099
1932
1.98
0.30
2.96
Nether-
0.044
0.23
-0.15
-0.025
0.085
1932
1.94
-0.02
2.32
lands
Sweden
0.046
0.18
-0.39
0.016
0.148
1932
3.22
-1.20
9.75
Norway
0.026
0.10
-0.09
0.010
0.053
1932
2.06
-0.22
1.13
Switzer-
0.041
0.27
-0.23
0.048
0.088
1931
2.16
-0.12
6.97
land
The data on civic unrest and political stability is from the cross-national data
set compiled by Arthur Banks under the auspices of the Center for
Comparative Political Research at the State University of New York. In
addition to a set of demographic and economic variables, it also contains
information on the nature of the political system and social instability for a
set of 166 over the period 1815-1973. Table II compares the main indicators
for our subsample of ten countries, and the data set as a whole. Overall, the
interwar data set for a number of countries that are developed today shows a
relatively high level of political instability and violence. For most indicators
of political uncertainty, the levels are twice the average observed in the
larger data set. This is true of the number of assassinations, of general
strikes, government crises, riots, and anti-government demonstrations. In
three categories, the subsample actually appears more stable - there were
fewer revolutions, purges and acts of guerrilla warfare than in the 166
country sample. The variability of our measures of political instability is
considerable, ranging from a coefficient of variation of 3.9 in the case of
revolutions to 1.98 for government crises. While Germany scores very high on
almost all measures of political fragility, recording a total of 188 events of
unrest, Switzerland marks the opposite extreme. Only three acts indicating
instability are recorded - two assassinations (in 1919 and 1923) and one riot
(in 1932).
Table II
Measures of Political Instability
The data is from Banks 1976, and shows the number of events per country and year. All data
is for the years 1919-1939, where available. The countries are Belgium, Switzerland, France,
Germany, Italy, Netherlands, Norway, Sweden, UK and US. The last column gives the ratio
of the average number of events in the 10 country sample divided by the average number of
events in the 166 nation sample.
10 Country Interwar 166 Nation
Sample Sample
average st.dev. max N average st.dev. max N ratio
averages
number of
0.28
0.77
5
233
0.14
0.51
9
4066
2.01
assassinations
general strikes
0.26
0.62
3
233
0.11
0.51
13
4066
2.37
guerrilla warfare
0.22
0.81
7
233
0.28
1.09
34
4066
0.79
government
0.60
1.19
6
233
0.30
0.73
7
4066
2.00
crises
purges
0.27
0.75
4
233
0.34
1.01
34
4066
0.78
riots
1.47
2.99
22
233
0.64
2.18
55
4066
2.29
revolutions
0.07
0.27
2
233
0.20
0.56
6
4066
0.34
anti-government
0.75
1.62
11
233
0.35
1.69
60
4066
2.14
demonstrations
There is also plenty of change over time. While 1919 saw, for example, four
times the average number of assassinations in the subsample of 10 countries,
there were none in 1936-38. The number of anti-government demonstrations
reached more than twice is average level in 1932, and the number of riots
peaked in 1934 at almost twice its normal frequency. Unsurprisingly, the
tendency of governments to resort to violent acts of repression also peaked
during the tumultuous years of the Great Depression, with the frequency of
purges reaching a high of 2.6 times its average level in 1934.
II. Political Instability and Civic Unrest during the Interwar Period
Europe and the US experienced two waves of turmoil and increasing
uncertainty. In each case, the continued existence of the established political
and economic order was in question. Following the end of World War I and
the Russian Revolution in 1917, chaos and civic unrest broke out in
numerous countries. After the end of the Habsburg dynasty and the
disintegration of the Austro-Hungarian Empire, a large number of new
nation states was formed. In Germany, the Emperor abdicated; revolution
came when Navy sailors mutinied and widespread strikes broke out.
Returning troops supporting the Social Democratic government were fighting
former comrades who sought to establish a German equivalent to the Soviet
Union, led by two leading communist intellectuals of the day, Rosa
Luxembourg and Karl Liebknecht (Winkler 1985). Right-wing putsches such
as the Kapp Putsch in 1920 and the Hitler Putsch in 1923 destabilized the
new democratic order, already undermined by the harsh terms of the
Versailles treaty. Leading political figures such as Matthias Erzberger and
Walter Rathenau fell victim to political murder. A Belgian-French invasion of
the industrial heartland, the Ruhr, as well as Communist uprisings in
Saxony and Thuringia compounded problems Pittlingmayer 1998). In the
years 1919-23, there were 13 government crises, the same number of riots,
and three general strikes. In France, there were waves of strikes in 1919 and
1920, considered by some observers as "a concerted attack upon the structure
of bourgeois society" (Lorwin 1968, 334). Nonetheless, these attacks
ultimately failed -the trade union activist Merrheim said he "found in France
a revolutionary situation without ... any revolutionary spirit in the working
classes" (Lorwin 1968: 335).
In the US and Britain, demobilizations and the end of war did not lead
to the same degree of extreme instability as in continental Europe. However,
the very sharp contractions in output and employment in 1920/21, engineered
in part as an attempt to reduce prices and return to the gold standard at pre-
war parities, led to a considerable rise in worker militancy. This occurred
against the background of a considerable strengthening of organized labor.
As in the other belligerent countries, the position of labor had strengthened
as a result of the war effort - governments recognized unions and encouraged
cooperation between them and employers. ^ Trade union membership in the
TUC (Trades Union Congress) soared from 2.2 million in 1913 to 6.5 million
in 1920. In our data set, Britain records 39 riots between 1919 and 1922, 12
assassinations, 6 general or politically motivated strikes, and 5 major
government crises over the period. The average number of days lost in
industrial disputes soared from 4.2 million in 1915-18 to 35.6 million in 1919-
23, the highest recorded value.^ Dissatisfaction with the established order
could take a number of forms. In the US, there were 5 assassinations and
four general or politically motivated strikes in 1919-23. Only one riot broke
out, but 17 anti-government demonstrations were recorded. The total number
of strikes increased sharply, to 3,630 in 1919, involving 4.2 million workers
(Foner 1988). Fear of a Communist takeover took the form of the so-called
"Red Scare". Following the founding of the Third International in March, two
Communist parties were formed in 1919, and quickly became active in
propaganda (gchmidt 2000). In response to bombs mailed to politicians by
terrorists, a widespread crack-down, led by the Justice Department's Radical
Division under J. Edgar Hoover, began.
5 Cf. Flanders 1968, 8-9; Lorwin 1968, 330-333; Taft 1958, 272-4.
6 Flanders 1968, p. 65.
anti-government
demonslfalions
revolutions
purges
governmeni
crises
guerrilla
warfare
general
strikes
assassinations
la B^-:
I
\ «
\ I
ipMj
\ 19 :"j
^
!i
1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938
Figure 1: Political and civic unrest in 10-Country-Sample, 1919-1938
The figure shows the total number of events per year, broken down by category. The data is
from Banks 1976.
The second half of the 1920s saw a considerable decline in worker militancy
and political violence. The 'roaring twenties' brought prosperity to many
countries, with some exceptions. The US economy expanded rapidly, France
reaped the benefits of currency stabilization under Poincare, and Germany,
with the help of foreign loans, experienced an upsurge in activity after the
end of the hyperinflation ^ichengreen 1992, Balderston 1993, Borchardt
1991). At the same time, Britain's economy - tied to gold at an overvalued
exchange rate - continued to languish (Moggridge 1972). But even in those
countries that didn't experience booms, labor militancy was on the wane.
With the exception of the general strike in Britain in 1926 (Flanders 1968),
labor movements created few troubles. The democracies of central Europe
appeared to be stabilizing (Maier 1975). Riots declined to less than one-third
their average frequency in the preceding half-decade; government crises,
which had been running at an average of more than 10 per year in the early
1920s, fell to 3 in 1927, 2 in 1928, and 5 in 1929.
The second wave of unrest and politically motivated violence began in
1930, with the start of the Great Depression. Over the course of the crisis,
industrial output in the US and Germany fell by 40-50 percent from peak to
trough, and between a quarter and a fifth of all industrial workers were
unemployed over the period 1930-38 (Feinstein, Temin and Toniolo 1997). In
the face of massive capital outflows and pressure on reserves as a result of
banking panics in Germany, Austria and the US, central banks first tried to
defend the gold standard by a policy of deflation (Eichengreen 1992).
10
Eventually, more and more countries abandoned the peg, either by devaluing
or via a system of capital controls. Countries that remained on gold for a long
time experienced the most severe contractions. France, which had initially
avoided problems, eventually experienced major difficulties. Faced with a
slump that extended into the second half of the 1930s, it was eventually
forced to devalue in June 1937. Britain, which was amongst the first to
abandon the gold standard, escaped relatively lightly.'' Recovery came faster
and in a more robust way to the countries that abandoned gold first
(Eichengreen and Sachs 1985).
Economic difficulties were quickly reflected in the politics of the street
and the factory floor. The total number of anti-government demonstrations
soared from 22 in 1925-29 to 72 in 1930-34; riots rose from 62 to 108. The
number of politically motivated general strikes increased from 7 to 10. In
Germany, there is clear evidence that high rates of unemployment did much
to boost the fortunes of the Communist party, already one of the strongest in
the world {''alter 1991). Recent research also demonstrates that areas in
which incomes contracted particularly sharply saw the largest increase in
votes for the Nazis (Stogbauer 2001). In Britain, the Bank of England decided
to leave the gold standard instead of raising the (relatively low) discount rate
- a decision that can only be understood as an attempt to avoid any further
rise in unemployment, and the threat of instability that would follow from it
(Eichengreen and Jeanne 1998). Apprehensiveness was accentuated by the
mutiny of the Royal Navy in the port of Inverness in 1931.
In the US, the Communist party expanded rapidly during the Great
Depression, and union membership soared. As "Hoovervilles" spread around
American cities, bitterness against the rich and civic unrest became more
widespread. Arthur Schlesinger noted about the year 1931 that "a malaise
was seizing many Americans, a sense at once depressing and exhilarating,
that capitalism itself was finished" (Schlesinger 1957, 205). The Hoover
administration - despite its general willingness to balance the budget by
whatever means necessary - opposed a cut in Army infantry units in 1931
because it would "lessen our means of maintaining domestic peace and
order." (Schlesinger 1957, 256). In a secret message to Congress, the
President urged that troops be exempted from a 10 percent pay cut so that
the nation would not have to rely on disaffected troops in case of internal
troubles. William Z. Foster, one of the most outspoken Communists in the
US, published his book Toward Soviet America in 1932. The party found rich
grounds for its agitation amongst the millions of unemployed and
impoverished <3chlesinger 1957, 256, 219). In the same year, the so-called
Bonus Army marched on Washington - veterans demanding that their
bonuses be paid ahead of time. It took cavalry, infantry and tanks,
■7 The relatively limited scale of the slump in Britain must be put in the context of its
sluggish performance over the period 1920-30. Cf. Feinstein, Temin and Toniolo 1997.
11
commanded by General Douglas MacArthur, to regain control (Zinn 1999,
381-2).
Perhaps even more importantly, the crisis rapidly increased the
chances of Franklin D. Roosevelt gaining office. While even the most
conservative businessmen did not equate this with a communist take-over,
worries about the continued existence of "capitalism as we know it" were
rampant. As Schlesinger noted, the "New York governor was the only
presidential candidate in either major party who consistently criticized
business leadership, who demanded drastic (if unspecified) changes in the
economic system, who called for bold experimentation and comprehensive
planning." (^chlesinger 1957, 290-1) Worries about future economic policy
was compounded by the increasing realization that a return to the so-called
"New Era" of prosperity and growth was impossible. Faced with growing
labor militancy and an increasing willingness to contemplate central
planning among the mainstream parties, right-wing radicalism also began to
gain a following. Some observers and politicians, including prominent US
senators, began to call for a Mussolini-style government, and magazines such
as Vanity Fair and Liberty argued the case for a dictatorship (^chlesinger
1957, 268).
III. Unrest and Volatility
What, then, were the effects of civic unrest and political uncertainty? Average
volatility in our sample shows two peaks, one during the early 1920, and a
second one during the Great Depression (Figure 2). The high point in 1923 is
driven by the extremes of stock price volatility seen during the hyperinflation
in Germany, as the difference between the mean and the mode in our sample
makes clear. These run broadly in parallel with the upsurges in political
violence and worker militancy. In this section, I discuss the extent to which
we can find a systematic association between the two.
12
0 09 .
j
1
% 1 i
1
1
A J ±
0.07 ■
0.06 ■
0.05-
A
j
1
£1 1 i
7 \
1 / F 1
_ S~A. .A\
1 1 1 1 /
jT^-^-^^^^^^ ii\ \y^y
Vx2!I^^ 1 jS^:.
^ „ ^ „ ^_ ^ 1 l
0.01 ■
0-
1927 1929
Year
Figure 2: Stock Price Volatility in 10-Country Sample, 1919-1938
The figure plots the mean and median of the monthly standard deviation of continuously
compounded real stock returns. For sources, see Data Appendix.
Some of our measures of political instability appear highly correlated with
the volatility of stock returns, as well as with each other. Table 111 gives the
results. Assassinations, strikes, acts of guerrilla warfare, riots, purges and
revolutions are frequently correlated with each other. The correlation of stock
price volatility with government crises is also evident and significant at the 5
percent level, as is the impact of riots and demonstrations (significant at the
1 percent level). Share price volatility is also strongly and significantly
correlated with the volatility of inflation. ^
8 This is in contrast to the results by Schwert (1989b), who finds that the predicted volatility
of the producer price index is only weakly correlated with stock price variability. Our results
are largely unchanged when we use the conditional variance of inflation from a GARCH (1,1)
model instead of actual variability of price changes.
13
Table III
Correlations of Indicators of Political Instability, Share Price
Volatility, and the Volatility of Inflation
The number of events in each country per year is correlated with the volatility of
continuously compounded monthly real return in the same year, and the volatility of
monthly rates of inflation. ASS is the number of assassinations per year, STRIKE the
number of politically motivated or general strikes, GUE are acts of guerrilla warfare, CRISIS
refers to the number of government crises, PURGES are the violent crackdowns on the
opposition, by the government or forces sympathetic to the government, RIOT is the number
of violent demonstrations and riots, REV is the number of attempted revolutions (successful
or not), and DEMO is the number of anti-government demonstrations not directed against
foreign powers. For sources, cf. the Data Appendix.
STRIKE
GUE
CRISIS
PURGES
RIOT
REV
DEMO
SVOL
PVOL
ASS
0.26**
0.35**
0,15*
0,24**
0.27**
0.18*
0.10
0.01
-0.01
STRIKE
1.00
0.35**
0.16*
0.04
0.38**
0.23**
0.30**
0.13
0.10
GUE
1.00
0.10
0.08
0.30**
0.41**
0.00
-0.01
-0.01
CRISIS
1.00
0.04
0.32**
0.19*
0.15
0.15
0.05
PURGES
1.00
0.09
0.10
0.11
-0.08
-0.03
RIOT
1.00
0.35**
0.46**
0.21
-0.00
REV
1.00
0.05
-0.02
-0.01
DEMO
1.00
0.24**
0.01
SVOL
1.00
0.68**
To test for connections between the degree of political uncertainty and stock
market volatility more formally, 1 estimate panel regressions of the type:
CT,=c^+p,X, + l3,P,+e (1)
where dt is the standard deviation of continuously compounded monthly real
stock returns in country i at time t, Xit is a set of macroeconomic controls, and
Pit are the indicators of political and social instability discussed above. Table
IV reports the results of estimating (1) with generalized least squares for the
full sample over the period 1919-1939. Some of the indicators of political
unrest emerge as highly significant. Anti-government demonstrations are
important in driving up volatility, as are government crises. Collinearity
between the demonstrations variable and those for riots and strikes leads to
some imprecisely estimated coefficients (Table IV, eq. 1). 1 therefore
calculated a summary variable, CHAOS, equal to the (unweighted) sum of
strikes, riots and demonstrations. It emerges as consistently and highly
significant. To illustrate the nature of the variable, consider Figure 3, which
plots the component series of CHAOS alongside stock price volatility.
14
g
CO
">
0)
■D
T3
0.00
15
10
-5
— 1 — I — I — I — 1 — I — I — I — I — I — I — I — I — I — I — I — I — I — I —
20 22 . 24 26 28 30 32 34 36 38
'-:-0
--5
cr
(D
CD
<
CD
SVOL_US —
DEMO US —
- RIOT_US
— STRIKE US
Figure 3: Share Price Volatility and Unrest in the United States
The same is true of PURGE, which indicates that crackdowns on mihtants
significantly reduced the volatihty of equity values. Higher inflation
variability leads to greater volatility of stock prices. The use of fixed effects
has little effect on our results. These effects are large in an economic sense. A
one standard deviation increase in the number of demonstrations would have
raised stock price volatility by 14 percent; a one standard deviation rise in
our CHAOS variable has an impact of 22 percent. For the PURGE variable,
on the other hand, the effect is a reduction by 9.5 percent. While we are able
to explain between 7 and 8 percent of the total variation in stock price
volatility with political variables, inflation volatility alone can explain up to
45 percent. The fixed effect dummies add another 9 percent. The results
demonstrate that, while civic unrest and politically motivated violence clearly
had an effect on stock prices during the interwar years, it's explanatory
power is not overwhelming. Controlling for the level of inflation does not alter
this result (equation 11).^ A number of variables are not significant - as is
the case for changes in the executive (EXECCH), the number of elections
(NELECT), the number of assassinations (ASS) and revolutions (REV).
9 Note, however, that the negative and significant coefficient is not robust to changes in the
specification - estimating in logs (to cope with the extreme values observed during the
German hyperinflation) yields an insignificant coefficient.
01
o
Oh
T3
c«
> S
H >
O
o
^
Cfi
o
o
be
4J
O)
!»
(1)
-us
w
^
+
>
O
0)
<D 0)
g «
E-" he
CO JH
t3
o o
o ^-'
O CO
§5
<y> CO
o $S
o
o
o
o
o
O C~
O ^
o
f~. o
CO
o 00
£ +
ft en
^ +
5 II
s
a
o
o
CO
4.3
CB
CO
c«
t3
-^
s~t
01
o
H
fo
o eg
eg 02
d d -^
1^ ,-.
o 5J
*
eq -*
CO CD
o
o
CO
,,_^
00
0
LO
0
0
1 — 1
0
0
d_
d
d
CD
^
i>
0
00
0
■ CO <^
o r-
, o ^
d ^-'
eg
00
1^ ,— ^
Tf CO CO
" O O IT)
O • c-3 •
O !M ■ "-I
eg
o
o
o
o
CD 10
eg 03
o
o
eg
^ —I
eg
o
o
-1 P d
0
0
0
^
s
0
ID
w
w
CU
p
Q
Di
o
o
M
m
M
0
>
H- 1
1— I
0
w
«
y^
P3
0
w
O
2
^
^ «• P
at ^ — t^
CO
M
22
CO
eg
►J
^
M
o
o
12;
o
LO CO
-^ o
d <^
CO CO
^ CO
CO o
d eg
00 CD
LO o
d
M
0 <£)
w
10 0
>"
0 ^
CO CO
CO o
eg CD
CO o
o <^
eg CO
CD o
d =^
CO CO
LO o
c= eg
t- CO
o o
d^
O o
d <^
CO M
T3 -M Q:;
■rH 55 T3 ^_
18
An obvious concern with our regressions in Table IV are possible correlations
in the error terms. As our historical narrative stressed, the risk of revolutions
and other challenges to the established economic order was often highly
correlated across countries - as could be seen in the wave of strikes and
attempts at revolution after the end of World War I, or during the Great
Depression. To ignore the correlation in the error terms would be to overlook
a significant element in the history of the period. To deal with the issue, I
estimate seemingly unrelated regressions (SUR) of our baseline specification.
Table V gives the results. The coefficient for the indicators of civic unrest are
often somewhat smaller, but more tightly estimated than under GLS. The
negative and significant coefficient on PURGE is broadly confirmed, as is the
volatility-increasing impact of CHAOS. DEMO has a significant coefficient in
2 out of 3 cases, and CRISIS emerges again as significant. PVOL is also
highly correlated with stock price volatility. ^^ The main difference with the
results reported in Table IV is that there is now a clearer indication of the
number of elections in any one year increasing volatility (eq. 11). Also,
increased numbers of changes in the executive appear to undermine the
stability of share prices (eq. 10). These results are similar to the recent
finding that share price volatility in emerging markets is systematically
higher during elections (Mei 1999).
10 This is in line with recent findings by Hu and Willett 2000, who document evidence in
favour of the variability hypothesis.
as
u
o
p.*
a
a
en
> S
3 o
o
CIS
o
o
Ph
y
o
CO
Ml
0
43
CO
-t^
T3
o
a)
<p
o
p.
CO
CO
o
<D
ii
^
0)
+J
CD
CD
43
>
>
-d
o
CO
T3
o
0)
en
(p
cs
43
e
+J
CO ^
^^ CO
0) 0)
_co d
0) <
CO
^ +
£ +
CO
^ +
1^
^ b~
3 CO
a
5 co"
o ,--1,
>-
IC CD
o S
t-
^^
o
00
o
CO
o
IM
o
^
o
CM
q
d
<d
*
00
o
o
o
■OT
■ 0)
CO
o "
CO
o S
° d ■=! S
o -^■
3
p^ CO
en
03 CO
■ S'-S
43 o
CD ^^
CO
o w
■ rt IP
lo '5
CD CO
H 12.
o "^
,-1 cc
o '^
d o
d
O "^
O ^
^ CD ^ P
o iri o ■■ •
o
o ^
o d
o ^^
o in
o ^-'
O Tf
o ■— '
d d-
CO .
O CO
o ■— '
O
Q
Q
b — '^ — .
5 lo o in
o -i P -J
d ~-'° '-'
O
(ti
K
^
M
CQ
w
o
o
W
O
1 — 1
o
w
J
M
^
>
o
>
M
O
pc;
o
w
2;
Cu
2:
M
C/2
O
2
O
2
o
2
CO CD
^ o
d <^
^ CD
o <^
CO CD
CO o
d <^
CO CO
in o
d
CD CD
in o
d <M
IM CD
CD o
d <M
M
»-H
CO
W
CO
o
f^
d
IM
CD o
in o
di IM
in
in Sg
o o
d ^
in CD
q o
-O -w Pi
IP c^
S ,<P -A
t-l 0) CO 12
21
Another possible objection is that results might be driven by the
inclusion of Germany in our sample, where the connection between
stock price volatility and political chaos was particularly close
(Bittlingmayer 1998). I therefore re-estimate the principal results of
Table IV and Table V excluding the case of Germany (Table VI). The
coefficients on CHAOS appear largely unchanged if marginally
smaller, and PVOL again emerges as a large and significant factor
contributing to higher volatility. RIOT and DEMO also contribute to
higher variability of stock returns in all specifications except eq. (6),
where we estimate in logs. There, the lagged value of the number of
anti-government demonstrations is not significant.
Table VI
Stock Price Volatility and Civic Unrest - 9 Country Sample
The table reports results for the regression
cj„=cc,+l3,X„ + l5,P„+e
Estimation technique is seemingly unrelated regressions (SUR). T-statistics (based
on White heteroscedasticity-consistent covariances) in parentheses. For data sources,
cf. Data Appendix. *, ** indicate significance at the 10 and 5% level, respectively. The
sample contains all countries except Germany. The dependent variable is est except
in eq. (6), where it is In(oit).
1
2
3
4
5
6
PURGE
0.00036
(0.13)
0.0003
(0.11)
-4e-5
(0.02)
DEMO (-1)
0.0048**
(4.7)
0.0048**
(4.7)
0.0099
(0.56)
RIOT
0.0007*
(2.45)
0.0008*
(2.4)
0.037**
(2.6)
STRIKE
1.5e-5
(0.008)
-0.007
(0.3)
CHAOS
0.0012**
(3.9)
0.0007**
(2.6)
0.0007**
(2.6)
PVOL
0.53**
0.37**
0.31*
0.32*
0.16**
(4.02)
(2.3)
(1.78)
(1.8)
(2.97)
Fixed
NO
YES
YES
YES
YES
YES
effects
adj. R2
0.08
0.27
0.27 -
0.34
0.34
0.32
N
182
182
182
182
182
182
Using the standard deviation of monthly returns in country i at time t
as a dependent variable generates easily interpretable results.
However, since the days of pioneering studies (such as Officer 1973)
that used a similar approach, more advanced techniques have become
22
available.il One of the well-observed regularities of equity returns is
time-varying volatility - large (positive or negative) returns tend to be
followed by large (positive or negative) returns. Adding lagged values
of SVOL in the regressions in Table IV and Table V does not change
our results. An alternative approach is to derive conditional variances
from GARCH models, and to use these as dependent variables.
Table VII
Conditional Stock Price Volatility and Civic Unrest
The table reports results for the regression
Estimation technique is seemingly unrelated regressions (SUR). T-statistics (based
on White heteroscedasticity-consistent covariances) in parentheses. For data sources,
cf. Data Appendix. The sample contains all countries except Germany. The
dependent variable is the conditional variance from GARCH(1,1) models for each of
the 10 countries. *, ** indicate significance at the 10 and 5% level, respectively.
1
2
3
4
5
6
7
PURGE
■0.0004**
(3.4)
0.0002
(1.2)
DEMO
0.0006**
(5.0)
0.0005**
(4.8)
RIOT
0.0004**
(5.9)
CHAOS
0.0003**
(6.2)
0.0002**
(3.6)
PVOL
0.07**
(10.3)
0.07**
(11.1)
0.07**
Fixed
NO
NO
NO
NO
NO
NO
YES
effects
adj. R2
0.003
0.004
0.01
0.01
0.61
0.64
0.66
N
188
188
188
188
188
188
185
Table VII reports the results of re-estimating our models using the
conditional variances from GARCH(1,1) models as the dependent
variables. The coefficients on our indicators of unrest and militancy are
estimated tightly. If anything, chaos and turmoil are more helpful in
explaining conditional variances than the unadjusted ones — a one
standard deviation rise in CHAOS increases the conditional variance
by 27 percent relative to its mean, while a one standard deviation
change in DEMO has an impact of 20 percent (the respective values for
unadjusted variances were 22 and 14 percent).
Political chaos and unrest, especially acts of labor militancy
aimed against the government of the day and the political system more
11 For an overview, cf Campbell, Lo and MacKinlay 1997, ch. 12.2.
23
broadly, did contribute to higher volatihty of stock returns during the
interwar period. While the effect is not uniformly strong for all
indicators of instability, a number of variables emerge as consistently
significant. These are the number of strikes, riots and anti-government
demonstrations. Independently of the estimation strategy used, the
inclusion of fixed effects, and the selection of sub-samples, these
appear to be a considerable part of the story about high and increasing
variability of stock returns during the Great Depression in 10
relatively advanced countries.
IV. The Risk of Revolution
So far, we have implicitly used an indirect mapping from political
violence and worker unrest to stock price volatility. The logic of our
argument, however, suggests that the main cause of the impact of any
political unrest variable on stock price volatility should be changes in
the expected chances of survival of the established economic and
political order. I therefore examine the extent to which these variables
would actually have been useful in predicting revolutions — either
attempted ones or those that succeed. Logistic regressions show that
indicators of civic unrest and anti-government militancy are highly
useful predictors of revolutions. From these regressions, we can derive
the threat of revolution - similar to the threat of takeover examined in
the corporate finance literature (Agrawal and Knoeber 1998). The
probability of an attempted overthrow of the government can then be
used to explain stock price volatility. I find that changes in the
likelihood of revolutions alone is sufficient to explain about 7-20
percent of the variation in stock price volatility.
There are 14 revolutions in our data set - as well as 194 annual
observations at the country level showing no revolution. As a first step,
we model the likelihood of an (attempted) violent overthrow of the
government, depending on the indicators of political instability and
violence used above. The predicted values are then correlated with
stock price volatility. This is essentially a data reduction strategy,
similar to factor analysis - except that our new exogenous variable has
a clear interpretation. In Table VIII, I report the results for logistic
regressions with revolutions as the dependent variable.
Multicollinearity between the exogenous variables, as noted above,
sometimes leads to insignificant coefficients. Independent of the
specification used, we find that the number of government crises in
any one year is an important predictor of the risk of a violent bid for
power. Riots are also highly significant in all regressions with the
exception of (3). Purges and other acts of violent suppression are
clearly more frequent in the run-up to revolutionary events, as are acts
24
of guerrilla warfare. While the Pseudo-R^s are never high, the
percentage of events correctly predicted is always above 90 percent.
Table VIII
The Risk of Revolution - Logistic Regressions
The dependent variable is a dummy variable Qt=l if a (attempted) violent overthrow
of the established government occurred, 0 otherwise. The Pseudo-R^ is the
Nagelkerke-R2. Wald statistics in parentheses. For data sources, cf. Data Appendix.
** indicate significance at the 10 and 5% level, respectively.
1
2
3
4
STRII^ ~
0.47
0.49
0.038
(1.67)
(1.7)
(0.006)
CRISIS
0.397**
0.43**
0.45**
0.385**
(4.99)
(5.2)
(5.45)
(4.7)
RIOT
0.12*
0.14*
0.096
0.16**
(3.02)
(3.3)
(1.24)
(5.8)
PURGES
0.42*
0.42
0.4*
(2.8)
(2.42)
(2.7)
DEMO
-0.12
(0.4)
-0.008
(0.001)
ASS
-0.14
(0.14)
QUE
0.76**
(7.5)
Constant
-3.5**
-3.7**
-3.9**
-3.6**
(61.4)
(53.9)
(51.0)
(59.1)
Pseudo-R2
0.162
0.192
0.289
0.17
% correctly
93.75
93.75
93.27
92.79
predicted
X^
13.5
16.1
24.85
14.3
The risk of revolution varies widely in our sample. Based on the
predicted values from regression (1), Germany starts the period with a
22 percent risk of another revolution, and witnesses a peak of over 45
percent in the period immediately following the stabilization of the
currency in 1924/25. France, on the other hand, reaches the highest
risk level in 1932, when the risk of revolution surges to 40 percent. In
line with expectations, Switzerland is not a hothouse of social unrest,
consistently showing a risk of revolution below 3.5 percent during the
period. The mean risk in our sample as a whole climbs to an all-time
high in 1920, when it reaches 14.7 percent. After falling in the 1920s to
around 4.5 percent -- similar to Switzerland - it almost doubles to 8.4
percent in 1932. Using the forecasts from regression (4) again suggests
that the all-time peak is in 1920, at 11.9 percent, but that by 1932, the
second-highest value for the whole period is reached - 9.1 percent. The
medians tell a similar story. In 1932, they reach local maxima that are
25
between one fifth and one half higher than the average values for the
period as a whole. In line with the writings of many contemporary
observers and later historians, we also find evidence that 'strong'
authoritarian governments - where parliaments had only a small role
to play — were seen to provide a degree of safety against the risk of
revolution (Turner 1985, Nolte 1963). When we correlate the degree of
parliamentary responsibility (again taken from the Banks data set)
with the risk of revolution, we find a clear and positive association
with both the risk of revolutions and their actual number, i^
From the logistic regressions in Table VIII, we derive the
predicted probability of a revolution occurring in country j at time t. Is
this new variable significantly correlated with stock price volatility? To
examine this question, we use the predicted likelihood of a violent
attempt to overthrow the government as a regressor in equations
similar to (1). Table IX gives the results. There is a significant and
strong effect independent of the estimation strategy and the
specification of the variables. A rise by one standard deviation in the
risk of revolutions increases average stock price volatility by 0.4 to 0.7
percent -- equivalent to between 8 and 14 percent relative to the mean.
This is independent of controlling for the effects of price volatility, or
other socio-political indicators such as the frequency of purges (which
again reduce volatility). We therefore find strong and consistent
support for the Schwert/Merton h5^othesis. It seems natural to ask if
the "volatility puzzle" can thus be resolved. Figure 4 in the appendix
plots the residuals from our regression (5) in Table IX. They do not
remain within the 95 percent confidence interval for the entire time in
all countries. Germany experienced a significant unexplained spike in
1923/24, for example, and again in 1931, whereas the UK shows
deviations in 1931 and 1938. The currency crises in 1931 are probably
significant contributors to these levels of volatility. i3 In the US,
12 I use variable 121 to measure the extent of parliamentary responsibility. Cf. the
Data Appendix for definitions. Note, however, that only three countries in our
dataset receive less than the maximum score (of 3) in our sample - Germany, Italy,
and Switzerland.
13 I tested for the possibility that countries on the gold standard had systematically
lower share price volatility, or that transitions of the monetary regime raised
volatility. There was no consistent and large effect.
26
considerable residuals remain for 1929, 1931, 1932 and 1938. While
this is clearly unsatisfactory, it also suggests that our model explains
stock price volatility sufficiently well to reduce the extraordinary scale
of variability in 1932 to a relatively unspectacular deviation from
predicted levels.
0)
^
t3
CD
.B
10
T^
^
CO
r3
CJ
CM
a
0
■f-H
0
.— 1
g
■zl
(D
+j
CO
CO
CD
.—I
1— (
00
0
CO
<
lb
CO
a
CJ
CO
CJ
0
CO
iS
CO
'a
CO
0
CO
CO
T— I
H
<D
CO
a
CO
0
-3
c
"^ CD
_„- CD
Pi
C35
o
*
w
•fH
*
(nT h
to
* "
<« .2
0
0 3
■};
>
CO
•3
CO 0
CD >
00
*
So
v
CD
C!
2 ?^
T— (
10
P5
CD
ft
CD T^
cd
C|H
ft
j^ 0
o
<
sl
•K
T3
CO
■K
^^
Jii
a>
+J
,.— V 5-«
CO
00
Pi
CO
"a!
u
CO
Q
6^
t-
CO
0
d
10
CJ
CO
CD
CJ
. . 0
02 to
C CO
0} 'a
"hi
5-4
3
CO g
3 S
C3
0
CO
<x>
rt =«
a
CO
- CD
CO i>
H >.
CD
CD
CO
73
- CD
CO
»^
1
■a
a
CO
S-i
0
S a
0 CO
in
*
CD bJD
c^j
>
1
CO
<D
in
CO
a*
CO
0)
CO
c
ID
bD -^
CO
a ft
-*
d
pLt
+j
CO
ft
•= a
M
CO
CO
a
-2 0
U
c
-2
CD ^4-1
o
CO
o
to
t3
CD
CO
a>
CJ
a
CO
3 "^
>l
M
J
0
CO
bo
0^
'a
u
£<
f~t
a
G CO
01
a
bo
ftCcJ
Ol
*
CO
si
tn
CO
CO
■So
d
"3
CO
CD
a"
'co
a
0
u
^2
*
05
CO
CO
-1-3
o
+
S
CJ
a)
4J
-£? CO
■S CD
0
<6
sj
04
0
(1)
+
0
CO
CO
a> 0
ft m
CD CD
3
II
to
a
CJ
CD
2
W
S'
CO
CD
-1^
0
-1^
^ £
CO
£
^
H
b'
CO
CD
l.s
1— 1
^
■«? t. ^ H " §
CS : 2^ >-• d <^'
00 ■ -H
10 CO
q -^
1—1 CO
d ^^
o ■*
* S^ CO CO
eg °; M ^
=^. t.™ d
o
o -* ^
►5 o o
^ d '^
O -H ^
2 j^ CN
OS§
*
o
Sid
d <^
Z o 1^
•-I CO 00 CO
-*■ W "^ o
^ >^ O <>'
CD ^
O (Ni
i-H CO
d ^~'
-K 5^ CO <M CO
o
o ^ X
>5 o o
^ d "^
C-)
*
in '
o o)
d ^
o -^ ;2;
^ o o
■^ d "
c-q
o 00 se
•5 o P
■^ d
<M
CO
5 2
CO
CO
2
o
o
>
Oh
OS §
Z d 1^
CD
■- "TS ^
fe CO Z
29
By deriving estimates of the probability of revolution from
specifications such as those used in Table VIII, we implicitly assume
that agents at the time had information for the entire period 1919-
1939. An alternative approach re-estimates the logistic regressions for
every year, expanding the sample as time goes by. The probability of
revolution in year t will only be assessed based on information for the
period 1919 up to and including year t. I initially begin with the period
1919 to 1922 (to preserve a minimum number of degrees of freedom),
using specification 1 from Table VIII. The forecasts from these
regressions for each country in each one of these years form the first
entries for a new variable, CRISK. For 1923, I then estimate based on
1919-1923, deriving the probability of revolution in each country for
that year. Table IX reports the results of using these expanding-
sample forecasts. The earlier findings linking political uncertainty and
the risk of revolution to stock market volatility are considerably
strengthened, with larger coefficients that are also more statistically
significant.
Similar results can be obtained if we use the conditional
variances from GARCH(1,1) models, as in our previous exercise with
the variables on demonstrations, riots and strikes. I employ the three
alternative definitions of the risk of revolutions, as before. As a further
robustness test, I add an AR(1) term to our specification. Table X gives
the results. Results are largely unchanged. The danger of a violent
overthrow of the established order always leads to higher stock market
volatility - a one standard deviation increase in the risk of revolution
(RISK) is associated with a 18 percent higher conditional variances.
The overall share of variation explained with the revolutionary threat
model is not very large, but the size and significance of the effect is
unchanged if we include fixed effects or the variance of inflation rates.
The only variable whose statistical significance appears somewhat
fragile is CRISK (based on expanding-sample forecasts of the
probability of revolution), which is not significant in eq. (9).
30
Table X
Conditional Stock Price Volatility and the Risk of Revolution
The table reports results for the regression
Estimation technique is seemingly unrelated regressions (SUR). T-statistics (based
on White heteroscedasticity-consistent covariances) in parentheses. For data sources,
cf. Data Appendix. The dependent variable is the conditional variance from
GARCH(1,1) models for all 10. *, ** indicate significance at the 10 and 5% level,
respectively. ._
RISK
0.011**
(7.1)
0.004**
(2.6)
0.0006**
(2.5)
RISK2
0.011**
(7.1)
0.001*
(1.7)
0.0009**
(3.44)
CRISK
0.008**
(3.8)
0.007**
(3.5)
0.0002
(0.87)
PVOL
0.06**
0.06**
0.07**
0.026**
0.024**
0.026**
(11.2)
(9.6)
(10.9)
(5.9)
(5.5)
(5.8)
AR(1)
0.87**
(22.8)
0.89**
(25.6)
0.87**
(23.0)
Fixed
NO
NO
NO
YES
YES
YES
NO
NO
NO
effects
adj. R2
0.002
0.0006
0.008
0.65
0.63
0.66
0.63
0.61
0.63
N
188
188
188
185
183
185
176
174
176
So far, we have mainly focussed on the strength of the threat that
could be mounted by disaffected segments of society - as might have
been perceived by stockholders. However, in order to analyse the
chances of capitalism's survival, the strength of the current system
should arguably matter in addition to the degree of turmoil and unrest
created by the opposing forces. Parliamentary systems vary widely in
the extent to which they are able to produce strong governments.
While systems of proportional representation often allow even very
small splinter groups to gain seats in parliament, other systems (such
as those with a first-past-the-post rule for MPs) create strong
majorities out of relatively small absolute differences in voter behavior.
During our period, Weimar Germany marks one extreme - parties that
managed to poll 60,000 votes in the entire country were represented in
the Reichstag. At the opposite end of the spectrum, Britain's electoral
rules continued to return governments with sizeable parliamentary
majorities, even if the voting was close. Did it matter? For our
hypothesis to be confirmed, we would expect that greater
fractionalization should lead to more instability - for any given
revolutionary threat, the established order should be in greater risk of
decline and fall.
a
B
ei
Ah
o
a
o
N
s:!
o
^
o
a
o
>
o
Oh
o
o
o
-o
O)
CO
ni
.2 a
Pi
en
C
o
to
CO
ho ^
Td fl CO
CD CP
-s =^
Q) <;
=> CO
>.o
<u CO
CO
^ of
« CO
22©^
o
o
^ 00
o oj ^
o
05
2! in
CO
LO
o N '^ 2
q -d ^
o
(M
"? "-"l oq
® CO o N
CD ^ * CO
CO "^
O (M
O
2 -^
o ^
n
*
*
CD S"
CO -^^
CD S-
cSS
o
d
o o
d ""^
i> -^ S "^.
i Co S
CO o
^ +
CD
"^ +
cS
CD ecJ
3. -S
-d o
S-
s s
^ b~
H &
CD Cvi
CM
I — I
Pi
2;
o
>
Oh
M IM CO
U CD o
P d '>^
OT
00 ^
W
lO o
>-'
d <>'
M
o
2
O
Ol
w
* as
zn
^ ^
w
"•. r-,
>^
w
o
2;
<» CO
d <M
CD 00
d
o
o
o
qj CO
00 r-4
lO o
d s^
OO CO
lO o
<T> cx>
i-H o
o <^
O CO
o P
CO
I— I Mh x) t--
(x, W ca Z
32
Table XI reports the results of regressing stock price volatility on the
fractionalization index. The weaker the parliamentary system, the greater
the instability in national stock markets during the interwar period. The
effect is unambiguous in terms of statistical significance, even if the size of
the coefficient varies. Also, once we control for the type of parliamentary
system, the number of elections (which, a priori, should also be associated
with greater volatility) has the expected sign. A one standard deviation rise
in fractionalization increases share price volatility by 14.5 percent relative to
its mean.
Schwert (1989b) uses nominal returns in his calculations. When price
changes are not too dramatic, this is unlikely to introduce biases. In the
baseline results, I used real returns to avoid the extreme impact that
inflation had in some of the countries in our sample - it would be incorrect to
infer that share price volatility was particularly high if most of the variance
resulted from greater inflation variability. To examine the robustness of our
results, I re-estimated the main results using the standard deviation of
nominal monthly returns in any given year as the dependent variable. Table
XV and Table XVI in Appendix II give the results (with and without the years
of the German hyperinflation included). The results for our political variables
are largely unchanged, and never become insignificant, with two exceptions.
The variable for the number of anti-government demonstrations is not
significant in the full sample, but strongly so if we exclude the hyperinflation.
In the sample that excludes the German observations for 1919-24, the effect
of purges is not tightly estimated, and one of the tree measures of the risk of
revolution is marginally below statistical significance at customary levels.
V. Omitted Variable Bias
Reverse causation - with stock market volatility leading to higher risks of a
fundamental change in the political and economic system - is unlikely to be a
large problem. Of greater concern is potential omitted variable bias.
Recessions are known to be systematically associated with higher
stockmarket volatility (Schwert 1990, Schwert 1989b). If generally poor
economic performance led simulateously to both greater worker militancy,
associated with a higher perceived risk of political turmoil, and to higher
stock volatility, the documented impact of our indicators of civic unrest might
be spurious. There is also the possibility that higher volatility of output
results in unemployment, leading to a rise in economic misery and
(perceived) threats to the survival of the capitalist system.
To control for the differential effects of recessions, I include the
percentage change in real per capita income in the regressions. Since the
effect might well be non-linear, I also experiment with the square of the
33
annual growth rate. Both variables clearly play a role, but the significance of
risk-of-revolution variable and the chaos indicator is not undermined.
Table XII
Stock Price Volatility and Macroeconomic Performance
The table reports results for the regression
Estimation technique is seemingly unrelated regressions (SUR). T-statistics (based on White
heteroscedasticity-consistent covariances) in parentheses. For data sources, of. Data
Appendix. The sample contains all countries. *, ** indicate significance at the 10 and 5%
level, respectively.
1
2
3
4
5
6
7
8
CHAOS
0.002**
0,004**
0.0013**
RISK
(6.2)
0.065**
(2.7)
0.18**
(4.99)
0.036*
RISK2
(5.11)
0.05**
(2.5)
(2.2)
CRISK
(4.4)
0.027**
GROWTH
-0.028
-0.038*
-0.04*
(2.4)
-0.043*
0.08
0.11
-0.089**
-0.098**
GROWTH 2
(1.1)
(1.68)
(1.9)
(1.7)
(0.07)
(1.0)
(2.6)
0.0013**
(3.0)
1.99**
Fixed effects
NO
NO
NO
NO
NO
NO
(3.6)
NO
(5.4)
NO
adj. R2
0.054
0.026
0.011
0.016
0.007
0.008
0.19
0.17
N
196
196
196
196
195
196
196
196
VI. Summary and Conclusions
Did fear about social unrest and the danger of a violent challenge to the
economic status quo contribute to share price volatility in the interwar
period? I find strong evidence in favor of such a link. Anti-government strikes
and demonstrations as well as riots appear to have made equity investors in
a set of 10 developed countries significantly more jittery. These results help
to explain why some countries saw extraordinarily wide swings in share
prices in the course of a single year — more than 40 percent in Germany in
the early 1920s, and approximately half this level in the US in 1932. This
provides direct evidence in favor of the Schwert/Merton hypothesis - the
"volatility puzzle" during the Great Depression can partly be resolved if we
account for the danger of a political discontinuity, brought on by the social
dislocation of the slump. Not only are dangers to the capitalist system in the
US an explanatory factor during the Slump, they are also important in
understanding the extreme volatility seen in some European stock markets
during the 1920s and 1930s. It is therefore no accident that the "heyday of
34
American communism" (Klehr 1984) also saw violent swings in share prices;
similar forces were operative in countries where the establishment had
reasons to worry about the ability to beat back revolutionary movements that
tried to profit from the depression. As Schlesinger observed in the case of the
United States: "Now depression was offering radicalism its long awated
chance." (Schlesinger 1957, 206).
This argument can be reinforced by deriving a measure of the risk of
revolution, based on the observed correlations between social unrest, political
violence, and the revolutions that do occur in our sample period. This "threat
variable" is a highly significant predictor of higher stock price volatility,
lending further credence to the hypothesis that investors feared a possible
repeat of the Russian Revolution in other countries. The impact of such an
event is, as Schwert (1989b), has argued, similar to a "Peso problem" — not
easily measured ex post, but clearly relevant to the decision-making of
economic agents at the time. I also find that weaker democracies, as
indicated by greater fractionalization and more frequent elections, were more
prone to experience wild swings in equity prices.
I do not argue that political violence, worker militancy and civic unrest
were exogenous to changes in economic conditions. As the political and social
history of the countries in our samples makes abundantly clear, the strength
of radical movements ebbed and flowed with the economic fortunes of their
countries (Stogbauer 2001, Falter 1991). Many of the countries that
experienced particularly severe economic shocks saw considerable upheaval.
The extent to which political collapse followed economic misery varied
considerably. While Germany and Italy became dictatorships during our
period, the US democratic system survived an economic crisis that was as
severe as Germany's. As recent work by Bittlingmayer (1998) has shown,
uncertainty in general may well have aggravated the decline in industrial
activity that was in part behind the upsurge in political violence and worker
militancy. What our results do show is that in those countries where
economic shocks of the early 1920s and, again, in the early 1930s, led to
greater political instability or the risk thereof, stock prices began to swing
wildly. While political chaos contributed to extreme stock price volatility, my
results also document that a part of the "volatility puzzle" still remains — the
models developed in the empirical section are not able to fully predict the
variability actually observed.
These findings suggest a clear agenda for future research. Recent work
on US share returns that decomposes the variability of aggregate indices into
the volatility of individual shares and the degree of "synchronicity" (Campbell
et al. 2001, Morck, Yeung and Yu 2000) should be replicated for other
countries during the interwar period. The Schwert/Merton hypothesis would
receive further confirmation if panel evidence confirmed a systematic
association between synchronicity on the one hand, and indicators of social
unrest and political instability on the other. An alternative route for future
35
research would be to construct indices of political risk at higher frequencies,
using some of the techniques based on news reports that have been used in
an attempt to explain variability in post-war data sets (Cutler, Poterba and
Summers 1989). With these, fully specified GARCH-models could be
estimated that would allow a more detailed modelling of the transmission
process from political uncertainty to stock price volatility.
36
Data Appendix
The stock price indices are from Global Financial Data (at
http://www.globalfindata.com). The individual series used are given in Table
XIII.
Table XIII:
Data sources -
Stock Indices
and CPI
Stock index
file
file for cpi price
index
UK
UK-FTSEAll
Share Index
_FTSAV.csv
CPGBRM.csv
Belgium
CBB Spot Price
Index
_BSPTD.csv
CPBELM.csv
USA
S&P-500
_SPXD.csv
CPUSAM.csv
France
SBF-250
_SBF250D.csv
CPFRAM.csv
Italy
BCI General
BCIID.csv
CPITAM.csv
Netherlands
CBS All-Share
CBSAD.csv
CPNLDM.csv
Sweden
Affarsvarlden
General
_SWAVD.csv
CPSWEM.csv
Norway
OBX-25
_OBXD.csv
CPNORM.csv
Switzerland
Switzerland Price _SPIXD.csv
CPCHEM.csv
Index
The exception is Germany, where the hyperinflation limits data availability. I
use the series compiled by Gielen (1992). For the period after 1919, it is based
on statistics compiled by the Imperial Statistical Office. His series is now
widely accepted as the best available long-run equity index for Germany
(Bittlingmayer 1998, Jorion and Goetzmann 1999). The growth rates of GDP
per capita are taken from Maddison 1995; GROWTH is calculated as the
difference between the natural logarithms of GDP in the preceding year and
the current year.
The political variables and indicators of unrest are from Banks (1976).
The code numbers and definitions are given in Table XIV.
37
Table XIV
Definition of Political and Social Variables
The table gives the definitions of the various indicators of political violence, legislative
efficiency and social instability used in our study. Source: Banks 1976.
Variable
name
Variable
number
Definition
ASS 91 The number of assassinations, defined as any
politically motivated murder or attempted murder
of a high government official or politician.
STRIKE 92 The number of general strikes, defined as any
strike of 1,000 or more industrial or service workers
that involves more than one employer and that is
aimed at national government policies or authority.
GUE 93 The number of acts of guerrilla warfare, defined as
any armed activity, sabotage, or bombings carried
on by independent bands of citizens or irregular
forces and aimed at the overthrow of the present
regime.
CRISIS 94 The number of major government crises, defined as
any rapidly developing situation that threatens to
bring the downfall of the present regime - excluding
situations of revolt aimed at such overthrow.
PURGE 95 The number of purges, defined as any systematic
elimination by jailing or execution of political
opposition within the ranks of the regime or the
opposition.
RIOT 96 The number of riots, defined as any violent
demonstration or clash of more than 100 citizens
involving the use of physical force.
REV 97 The number of revolutions, defined as any illegal or
forced change in the top governmental elite, any
attempt at such a change, or any successful or
unsuccessful armed rebellion whose aim is
independence from the central government.
DEMO 98 The number of anti-government demonstrations,
defined as any peaceful public gathering of at least
100 people for the primary purpose of displaying or
voicing their opposition to government policies or
authority, excluding demonstrations of a distinctly
anti-foreign nature.
FRACTURE 113 Party fractionalization index, based on a formula
proposed by Rae 1968. The index is constructed as
follows:
38
F = l-
S(0
where t is the proportion of members associated
with the ith party in the lower house of the
legislature.
PARLRES 121 Parliamentary responsibility, defined as the degree
to which a premier must depend on the support of a
majority in the lower house of a legislature in order
to remain in office.
Code Definition
0 Irrelevant. Office of premier does not exist.
1 Absent. Office of premier exists, but there is
no parliamentary
responsibility.
2 Incomplete. The premier is, at least to some
extent, constitutionally responsible to the
legislature. Effective responsibility is, however,
limited.
3 Complete. The premier is constitutionally and
effectively dependent upon a legislative majority for
continuance in office.
CABCH 123 Major cabinet changes, defined as the number of
times in a year that a new premier is named and/or
50% of the cabinet posts are occupied by new
ministers.
EXECCH 124 The number of times in a year that effective control
of the executive power changes hands. Such a
change requires that the new executive be
independent of his predecessor.
NELECT 127 The number of elections held for the lower house of
a national legislature in a given year.
CHAOS strike+demo+riot
CO
<
\j :
■ I— I
T3
a.
a.
3 3 ? 3 3
3 :
E$
H
O
Ul
0)
W)
c3
3
0)
W)
o
>
X
o
s
>
(V
U
u
Oh
o
Cfi "
o
>
V
CO O)
«
O QJ
T3
s
^ >
oj
-p
■"^ ?s
M .
-^ 5^
^s
E3
g «
P
C*H
a)
O
en oj
5 -^
^
^S *^
5
73
.2 e
*^ TO
S
cd o
cS
to ^
^^ 03
P^ 0)
W S
.2 *
CO -K
CO
>- -3
oj cu
_ra c.
0) <
C 2
2? f
0)
^ CO »).
CO ^ _^
OT
M
53 (J2
O (X,
S 9 d
>-
CD --1
CO CO
CO _;
o C-
d
C<1
d
*
*
oo lo
.— I 00
* _
lO —
00 (Xi
o
IX>
o
d
o c-q
o ^
d "-'
O
Q
O
O
O
Cc3
(M
W
w
M
1— 1
S
Di
o
o
>
m
CO (X,
w
q a>
>-
o '-I
M
^
o ^
OT
^ C-Q
W
o o
>^
d <^
o
o
d^
o °
d^
o
2;
o
2;
O o
d <^
O "^
o S
^ -7
CO A
^
tf
o
12;
O +J
0) CO
.2"g
<i
M
>, _>,
o
• 1-1
-(J
s
1-^
CT3 O)
o
tS d.
>
Q) ra
a>
en u
tf
o
0) 0)
-1^
1^
a:
o O
tM
QJ <P
0
re -»^
^
S o
re id
c3 >>
CO
H .-S
pj^
•i-H
4J
P 2
0
a, S
>
li:
<0
m
to J, -
• fH
fH
' u
ho X
^
13 0)
^
o
o
o
0)
m
CO
S 2
^.^
?
-J re
C8
cm
^°
• fH
^
S
^
o S
CO
M
^
CD CO
d —
M in
o ^
d ^^
g°^
M -* CO
o S
dSi
3
5. «
05:
CO
T^
fl
^
O
^'
Xi
u
CD
o
P.
Qd
-M
co'
!-<
a
0)
^
+
o
CO
0)
J2
re
II
re
CP
+J
tH
^
H
b'
CO
re
CD
i
M
^
W
^
15 "
d
C<1
O l^
O ^
d ^
O
2
d
o 2
d
d
O
2
CO lO
o
2
00 to
q OS
CO
en
O CO
d ^^
O
2
C^ CD
d
2 t
Pi
w
c«
p
m
w
H
O
o
o
Oi
o
PU
fn
Oi
t^
«
W
00
1— 1
Pi
M
s
o
>
Q O „
M W tf
fe W re Z
References
Acemoglu, D. and J. A. Robinson (1999). A Theory of Political Transitions. In,
Massachusetts Institute of Technology, Department of Economics Working
Paper: 99/26: 31.
(2000). "Why Did the West Extend the Franchise? Democracy, Inequality, and
Growth in Historical Perspective." Quarterly Journal of Economics 115(4):
1167-99.
Agrawal, A. and C. R. Knoeber (1998). "Managerial Compensation and the Threat of
Takeover." Journal of Financial Economics 47(2): 219-39.
Alesina, A. and R. Perotti (1996). "Income Distribution, Political Instability, and
Investment." European Economic Review 40(6): 1203-28.
Balderston, T. (1993). The Origins and Course of the German Crisis. Berlin, Haude
& Spener.
Banks, A. S. (1976). Cross-National Time Series, 1815-1973. In, Inter-University
Consortium for Political and Social Research.
Bekaert, G. and C. R. Harvey (1997). "Emerging Equity Market Volatility." Journal
of Financial Economics 43(1): 29-77.
Bitthngmayer, G. (1998). "Output, Stock Volatility, and PoUtical Uncertainty in a
Natural Experiment: Germany, 1880-1940." Journal of Finance 53(6): 2243-
57.
Borchardt, K. (1991). Perspectives on Modern German Economic History and Policy.
Cambridge ; New York, Cambridge University Press.
Campbell, J. Y., et al. (2001). "Have Individual Stocks Become More Volatile? An
Empirical Exploration of Idiosyncratic Risk." Journal of Finance 56(1): 1-43.
Campbell, J. Y., A. W. Lo and A. C. MacKinlay (1997). The Econometrics of Financial
Markets. Princeton, N.J., Princeton University Press.
Cutler, D. M., J. M. Poterba and L. H. Summers (1989). "What Moves Stock Prices?"
Journal of Portfolio Management 15(3): 4-12.
David, A. and P. Veronesi (2001). "Inflation and Earnings Uncertainty and the
Volatility of Asset Prices: An Empirical Investigation." unpublished
manuscript. Chicago.
DeFronzo, J. (1991). Revolutions and revolutionary movements. Boulder, Westview
Press.
Eichengreen, B. and O. Jeanne (1998). "Currency Crisis and Unemployment:
Sterling in 1931." Centre for Economic Policy Research, Discussion Paper:
1898.
Eichengreen, B. and J. Sachs (1985). "Exchange Rates and Economic Recovery in the
1930s." Journal of Economic History 45(4): 925-46.
Eichengreen, B. J. (1992). Golden Fetters : The Gold Standard and the Great
Depression, 1919-1939. New York, Oxford University Press.
Falter, J. W. (1991). Hitlers Wdhler. Miinchen, Beck.
Feinstein, C. H., P. Temin and G. Toniolo (1997). The European Economy Between
the Wars. Oxford ; New York, Oxford University Press.
43
Feldman, G. D. (1993). The Great Disorder: Politics, Economics and Society in the
German Hyperinflation, 1914-1924. Oxford, Oxford University Press.
Flanders, A. (1968). "Great Britain". In: Comparative Labor Movements, edited by
W. Galenson. New York, Russell & Russell.
Foner, P. (1988). History of the Labor Movement of the United States. New York,
International Publishers.
Galbraith, J. K. (1962). The Great Crash, 1929. New York, Time.
Gasiorowski, M. J. (1995). "Economic Crisis and Political Regime Change: An Event
History Analysis." American Political Science Review 89(4): 882-97. '
Goldstone, J. A. (1991). Revolution and rebellion in the early modern world.
Berkeley, University of California Press.
Goldstone, J. A. and R. K. Merton (1986). Revolutions : theoretical, comparative and
historical studies. San Diego.
Grossman, H. I. (1999). "Kleptocracy and Revolutions." Oxford Economic Papers
51(2): 267-83.
Hamilton, J. (1986). "On Testing for Self-fulfilling Speculative Price Bubbles."
International Economic Review 27(3): 545-52.
Hamilton, J. and C. Whiteman (1985). "The Observable Implications of Self-fulfilling
Expectations." Journal of Monetary Economics 16(3): 353-73.
Hu, X. and T. D. Willett (2000). "The Variability of Inflation and Real Stock
^etnms." Applied Financial Economics 10(6): 655-65.
Jorion, P. and W. N. Goetzmann (1999). "Global Stock Markets in the Twentieth
Century." Journal of Finance 54(3): 953-80.
Klehr, H. (1984). The Heyday of American Communism : The Depression Decade.
New York, Basic Books.
Kleidon, A. (1986). "Variance Bound Tests and Stock Price Valuation Models."
Journal of Political Economy 94: 953-1001.
Lorwin, V. (1968). "France". In: Comparative Labor Movements, edited by W.
Galenson. New York, Russell & Russell.
Maddison, A. (1995). Monitoring the World Economy. Paris, OECD.
Maier, C. S. (1975). Recasting Bourgeois Europe : Stabilization in France, Germany,
and Raly in the Decade after World War I. Princeton, N.J., Princeton
University Press.
Mei, J. (1999). "Political Risk, Financial Crisis, and Market VolatiUty." NYU
Business Department Working Paper.
Moggridge, D. E. (1972). British Monetary Policy, 1924-1931: The Norman Conquest
of $4.86. Cambridge, UK, Cambridge University Press.
Morck, R., B. Yeung and W. Yu (2000). "The Information Content of Stock Markets:
Why Do Emerging Markets Have Synchronous Stock Price Movements?"
Journal of Financial Economics 58(1-2): 215-60.
Muller, E. and M. Seligson (1987). "Inequality and Insurrections." American
Political Science Review 81: 425-451.
Nolte, E. (1963). Der Faschisjnus in seiner Epoche: die Action frangaise, der
italienische Faschismus, der Nationalsozialismus . Miinchen, R. Piper.
Officer, R. R. (1973). "The Variability of the Market Factor of the New York Stock
Exchange." Journal of Business 46(3): 434-53.
Prezworski, A., et al. (1996). "What Makes Democracy Endure." Journal of
Democracy 7(7): 39-55.
44
Rae, D. (1968). "A Note on the Fractionalization of Some European Party Systems."
Comparative Political Studies 1; 413-418.
Schiller, R. (1981). "Do Stock Prices Move too Much to Be Explained by Subsequent
Changes in Dividends?" America?! Economic Review 75: 421-436.
Schlesinger, A. (1957). The Crisis of the Old Order, 1919-1933. Boston, Houghton
Mifflin.
Schmidt, R. (2000). Red Scare : FBI and the Origins of Anticommunism in the United
States, 19191943. Copenhagen, Museum Tusculanum Press University of
Copenhagen.
Schwert, G. W. (1989a). "Business Cycles, Financial Crises, and Stock Volatility.
Reply." Carnegie-Rochester Conference Series on Public Policy 31(0): 133-37.
(1989b). "Why Does Stock Market Volatility Change over Time?" Journal of
Finance 44(5): 1115-53.
(1990). "Stock Returns and Real Activity: A Century of Evidence." Journal of
Finance 45(4): 1237-57.
Stogbauer, C. (2001). "Measuring the Impact of the Depression on the Radical Vote
in the Weimar Republic: A Spatio-temporal Approach." Economic History
Working Paper. Munich.
Taft, P. (1958). "Germany". In: Comparative Labor Movements, edited by W.
Galenson. New York, Russell & Russell.
Turner, H. A. (1985). German Big Business and the Rise of Hitler. New York, Oxford
University Press.
Winkler, H. A. (1985). Von der Revolution zur Stabilisierung : Arbeiter und
Arbeiterbewegung in der Weimarer Republik, 1918 bis 1924. Berlin, J.H.W.
Dietz.
Zinn, H. (1999). A People's History of the United States : 1492-Present. New York,
NY, HarperCollins Publishers.
Date Due 3/^/^
Lib-26-67
MIT LIBRARIES
3 9080 02246 1005
tM0i:
■mm
I ''it
■III
liiltiTI