A Monthly International Dataset for the Interwar Period

A Monthly International Dataset for the Interwar Period:
Taking the Debate to the Next Level I
Thilo Albers
School of Business and Economics, Humboldt University Berlin
Martin Uebele
Faculty of Arts, University of Groningen
Abstract
Recent research on the Great Depression has increased the demand for detailed data on business activity. This paper
presents a novel dataset that expands the data in three dimensions: (i) a larger cross-section with 28 countries, (ii)
a higher (monthly) frequency of real economy data, and (iii) disaggregated data ensuring representativeness. We
aggregate in total 415 single series using Principal Component Analysis to construct a business activity indicator for
individual countries for the years 1925–1936. We date business cycle peaks and troughs during the Great Depression,
and find that while pre-Depression peaks occurred in almost all countries in 1929, some countries were severely hit by
the crisis as late as 1930 and 1931. Furthermore, we analyse business cycle comovement during the interwar period
and find that exchange controls made countries more independent from the global economy, while the gold standard
induced comovement.
Keywords: Great Depression, time series analysis, gold standard, business cycle comovement,
exchange controls
JEL codes: C38, E32, E58, N14
1. Introduction
This paper introduces a novel dataset on business activity during the Great Depression. The
dataset expands the so far available data in three dimensions: number of countries, time-frequency,
and representativeness. Based on aggregated and disaggregated data, it covers 28 countries for a
period of up to twelve years. The need and the rationale for composing such a dataset stems from
I
We thank Ulrich Pfister, Marvin Suesse and Nikolaus Wolf for their comments. Furthermore, we thank Chris
Meissner for answering our questions concerning the methodology in Mathy and Meissner (2011). Jan Tore Klovland
provided his industrial production estimates. Moreover, we are deeply indebted to Daniel Gallardo-Albarrán, who
transcribed about the half of the seemingly endless pages of data that are a building bloc of this research project. All
errors are ours.
Email addresses: [email protected] (Thilo Albers), [email protected] (Martin
Uebele)
the historiography of the Great Depression, its development throughout the last 80 years, and the
progress in econometrics and computing.
The Gold Standard Literature, which emerged in the 1980s, has shaped today’s understanding
of the Great Depression and international macroeconomics. It suggested that the gold standard
was not only important for the transmission of the crisis, but that it was in fact causal to the Great
Depression (Eichengreen, 1992). The innovation that this literature brought to the scholarship of
the Depression was simple and path-breaking at the same time: instead of analysing only one
country (mainly the United States) it compared many countries. Related to the cross-sectional
data dimension, Ben Bernanke (1995, p. 1) argued: “By effectively expanding the dataset from
one observation to twenty, thirty, or more, the shift to a comparative perspective substantially
improves our ability to identify – in the strict econometric sense – the forces responsible for the
world depression.”
The Gold Standard Literature challenged the previously dominant Friedman-Schwartz (1963)
view. They had argued that monetary policy mistakes were causal to the Great Depression, but
not the monetary system (of the gold standard) itself. However, during the course of the last
30 years, the comparative approach of the Gold Standard Literature became no less mainstream
than the Friedman-Schwartz view once had been. This development might have obscured another
dimension of “Depression data” - time or rather the frequency of observations. If one reads Ben
Bernanke’s statement in this context, a complementary question comes immediately into one’s
mind: Would expanding the dataset from 10 to 120 points in time – as the shift from annual to
monthly data for a 10 years period would imply – not improve our ability to identify (in the strict
econometric sense) the forces responsible for the world depression?
More recently, a new stream of literature, which we coin the Post-Gold Standard Literature, suggests that an expansion of datasets in the time-frequency dimension is important (Wolf, 2008;
Mathy and Meissner, 2011; Accominotti, 2012). They stress, among others, the relevance of
within-year timing during that relatively short period, which from the mid-1920s to the mid-1930s
does not span more than one typical business cycle. Using monthly instead of annual data comes
at a cost, though: either the reduction of the cross-sectional dimension of the dataset (Wolf, 2008;
Mathy and Meissner, 2011) or the focus on a particular sector (see Accominotti, 2012, for the
financial sector). Wolf (2008, p. 389) notes that the selection of countries is mainly determined
by the availability of key variables necessary for analysis. The most important key variable for
this type of research, though, is an economic activity or business cycle indicator, which is rarely
available. Our novel dataset provides a remedy for this problem.
The third dimension in which we expand the dataset is the representativeness of the business
cycle indicator. Clearly, pig iron, steel, and coal production contain information about the current
state of the economy. Bank clearings, employment numbers and many other indicators do so as
2
well. For agricultural countries, however, bacon exports might be more important than machinery
production, whereas the opposite is likely to be true in industrialised countries. We therefore use
a large number of disaggregated indicator series, and employ Principal Component Analysis as a
simple but powerful statistical aggregation procedure to estimate business activity in the respective
countries.
We collected the time series from two publications by the German official statistical office,
the Statistische Reichsamt (“Statistisches Handbuch der Weltwirtschaft,” 1936, 1937), which was
compiled to give an empirical overview on the world economy.1 It bears much resemblance to
related NBER publications (Burns and Mitchell, 1946), but differs in its global coverage. We
can estimate business activity indices for 17 countries from a total of 415 series of monthly data
covering ten to twelve years. More than 55,000 data points now provide us with a much better
estimate of the state of the economy in the countries of our sample, and thus with a better insight
into global crisis dynamics than ever before. After combining our business activity estimates
with eleven official counterparts contained in either the Handbuch or other sources (see Appendix
D.18), the dataset finally covers 28 countries for a period of up to twelve years. After describing
the dataset and methodological aspects of its compilation, the main aims of this paper are the
following:
Firstly, we track the course of the Depression on a monthly basis. While in most countries
the pre-Depression peak took place before autumn 1929, the crisis became more pronounced in
some agricultural economies only in 1930 and 1931. A possible explanation would be that they
were hit indirectly by the depression of world trade. Moreover, some countries showed signs of a
short-lived stabilisation in 1930. Such findings illustrate the value of this new dataset. By using
annual data, such developments typically stay behind the veil of temporal aggregation.
Secondly, we show the advantages of the dataset by analysing the determinants of business
cycle comovement. We find that the gold standard had indeed a positive impact on comovement,
and that exchange controls had a negative one. If it had been the policy-makers’ goal to make their
respective countries more independent from the world economy, they succeeded. As a comparison
with earlier evidence based on a smaller empirical base (Mathy and Meissner, 2011) shows, this
new result is most likely due to the improved quality and quantity of our new data.
In sum, this paper shows the merits of taking all three data dimensions into account: (i) crosssection, (ii) time-frequency, and (iii) representativeness. The next section discusses the data, while
the treatment of data is explained in Section 3. Thereafter, Section 4 discusses the course of the
monthly business activity indices, and then replicates earlier work on business cycle comovement
with the novel data. Section 5 concludes.
1
We are deeply indebted to Nikolaus Wolf, who pointed us towards this source.
3
2. A new monthly dataset for the interwar period
This section presents the key features of the disaggregated data that we employ for the business
activity estimates. The disaggregated dataset covers 17 countries and 415 indicator time series.
Each of them contains up to 144 monthly observations. This yields a total of more than 55,000
observations. All series have been manually transcribed from Statistische Reichsamt (1936, 1937).
These collections of interwar economic data provide an exceptional source for research on the
Great Depression, covering the period 1925–1936. The Statistische Reichsamt gathered the data
from the national statistical offices and from publications such as The Economist, Lloyds Register
of Shipping, and publications by private banks.
TABLE 1: S UMMARY TABLE - D ISAGGREGATED DATA
Time
Period
Number
of Series
Real
Indicators
Nominal
Indicators
Real
Economy
Trade
Trade
Prices
Prices
Money &
Banking
Gold Bloc
Belgium
Lithuania
Netherlands
Switzerland
1925–1936
1925–1936
1925–1936
1925–1936
24
22
32
30
12
8
11
6
12
14
21
24
10
2
12
5
7
10
2
7
2
3
2
3
5
7
16
15
Sterling Bloc
Australia
Finland
India
New Zealand
South Africa
1926–1936
1925–1936
1925–1936
1926–1936
1925–1936
18
30
21
11
24
5
11
12
6
9
13
19
9
5
15
1
5
4
3
8
12
9
11
5
2
2
2
3
2
2
3
14
3
1
12
Exchange Controls Bloc
Bulgaria
1926–1936
Hungary
1925–1936
Italy
1925–1935
1926–1936
Latvia
Romania
1925–1936
17
35
33
26
20
2
13
18
11
8
15
22
15
15
12
2
5
7
6
4
2
11
14
9
6
5
8
1
2
1
8
11
11
9
9
Other Countries with Depreciated Currencies
Chile
1927–1936
Estonia
1925–1936
Japan
1925–1936
14
23
35
8
9
11
6
14
24
6
4
9
4
7
9
0
3
2
4
9
15
Total
415
160
255
93
127
43
152
Country
Sources for bloc composition: See Appendix B. For Italy the data is only available until June 1935.
Table 1 provides an overview of the included countries. The series availability varies by country.
This may induce a selection bias. One could, however, also interpret it as a “qualitative filter.”
Those covered are likely to have been more relevant for the economy than those that have not
been documented. As the number of series per country usually lies between 20 and 30, the filter
interpretation seems to be more convincing.
Besides the time period and the total number of series per country, the table summarises the
variables by their type in two ways. The first distinction is between “real” and “nominal” variables.
This indicates whether or not the variable could be affected by price changes. Real indicators
include those that are based on simple counting (such as number of individuals unemployed) or
4
given in some kind of metric measure (such as tons of goods transported on railways). The column
for nominal indicators counts all variables that are either given in the respective currency, that are
price indices or that resemble any kind of interest rate. Whenever a variable was given as an index
number, it was checked whether this index was based either on a real variable such as employment
or on a nominal variable such as stock turnover in the respective currency. Out of the 415 variables,
about 40% are real indicator variables and 60% are nominal indicator variables.
It is quite important to have such a balanced setup. While economists generally believe in
the importance of the price mechanism, some might argue that prices in the interwar period are
particularly biased because of (bad) central banking. Others might argue that prices are indeed
a good indicator and especially in the interwar period, in which central banks did not impose
measures such as quantitative easing. Hence, this paper takes nominal indicators into account, but
considers it important that there is a substantial fraction of real indicators in the sample.
Another possible characterisation of the variables is a sectoral one.2 Each variable was assigned
to one of the four sectors: (i) real economy [22.4 %], (ii) trade [30.6 %], (iii) prices [10.4 %] or
(iv) money and banking [36.6 %]. The real economy group includes series such as transport
on rails and ships, unemployment numbers, agricultural production, and mining indices. The
trade group includes only variables that are either exports or imports. The price group subsumes
consumer and wholesale price indices. Finally, the money and banking group includes all variables
related to either (central, public, or private) banks or the stock market. It is notable that the
distribution between the groups is quite balanced.
3. Data treatment and aggregation
3.1. Seasonal adjustment
Clearly, seasonality becomes a concern with such heterogeneous monthly data. The first step in
preparing the data for further estimation is thus to remove seasonal variation from the series. This
section elaborates on the motivation for and the procedure of seasonal adjustment.
Not excluding seasonality could bias the aggregate, because some seasonality patterns might
work in different directions. For example, retail sales usually increase just before Christmas, while
employment is usually highest in summer. Since the included indicator series vary by country, this
bias would become even more relevant. A further potential bias emerges, because the degree of
seasonal variation may differ between series. Figure 1 illustrates this potential bias. The left panel
2
At this point, we follow more or less the grouping suggested by Statistische Reichsamt (1936, 1937), although
minor regroupings were considered necessary. Instead of putting the number of bankruptcies into “Zahlungsschwierigkeiten,” or payment problems, it was added to the real economy group. The same is true for unemployment
numbers.
5
(a) Total Unemployment (Belgium)
(b) Private Market Rate (Belgium)
Figure 1: Seasonal Adjustment of Indicators
shows Belgian unemployment numbers. For example, the seasonally adjusted series suggests
that unemployment was increasing towards the end of 1935, when it was in fact decreasing. In
contrast, the Belgian private market rate exhibits almost no seasonal pattern (Figure 1(b)). If
one standardises and aggregates the raw data, seasonal unemployment patterns might obscure
variations in the private market rate.
We apply a S3×5 seasonal filter to each of our series. This procedure is based on a moving
average and allows the seasonal component to vary over time (see Mathworks, 2013). This is
especially important for longer time spans than ours. For example, one can think of the seasonality
of wheat yields before and after the emergence of irrigation systems. However, it might even be
relevant for our case, considering e.g. the endogeneity of agricultural production and prices of
agricultural goods.3
In sum, seasonal adjustment eliminates a potential bias that could arise because of the heterogeneous nature of the dataset. Furthermore, if the data exhibit no seasonal patterns, it would remain
unchanged by applying the procedure described above. Thus there is no good reason for keeping
the seasonal components.
3.2. Detrending
After having carried out the seasonal adjustment procedure, one could estimate the principal
components analysis (PCA) with either trended or stationary data. Since PCA ultimately rests
on an estimated covariance matrix, one might argue that applying it to non-stationary data might
yield spurious results. On the other hand, we want to filter out common trends and do not make
3
A detailed documentation of this procedure can be found in (Mathworks, 2013).
6
any causality statements. Nevertheless, de-trending the data before estimating the principal components is certainly the more robust way in a strict econometric sense. Like many authors, we use
the Hodrick-Prescott Filter (HP-filter henceforth) to separate the cycle from the trend (Hodrick
and Prescott, 1997).
However, there are two particular issues connected with this: (i) Our monthly dataset spans
about one business cycle which is by its very nature extremely volatile. It is actually the main
aim of this paper and the literature it belongs to to document the crisis from peak to trough and
back and not filter out whatever a statistical procedure may typically label as “trend.” Therefore, a
mechanical application of typical filters needs to be reconsidered.
(ii) The differing degrees to which the individual series were hit by the crisis render making the
individual series stationary particularly cumbersome. Instead of “adjusting the Hodrick-Prescott
Filter for the frequency of observations” (Ravn and Uhlig, 2002), adjusting the HP-filter for the
nature of the indicator was therefore necessary in some cases. This is because the Depression data
provokes a critical problem that might not be relevant in other applications of the HP-filter: Its
inability to induce mean stationarity to all series with the most common filter parameters.4
To distinguish between the cyclical and the trend component, first differencing would be the
standard way to ensure stationarity, unless the underlying trend has a polynomial far higher than
one. However, one reason for the emergence of other filter techniques is the loss of information
associated with such a way of filtering. Canova (1998) provides an extensive overview about
the discussion, while A’Hearn and Woitek (2001) provide a focused discussion in the case of
historical business cycles, albeit for annual data. It is beyond the scope of this article to discuss
the filter’s ability to transform a signal to a stationary series, while leaving its information content
unchanged at the desired frequencies. Instead, we only focus on the ability to transform the signal
to a stationary one at all, since this is the primary aim of filtering, and has to be ensured before
discussing how well the filter performs beyond that task. Thus, when choosing the smoothing
parameter λ for the HP-filter, we adjust it to the outcome of a unit root test ex post.
Figure 2 illustrates the rationale for our concerns. It shows a stock market index and total
unemployment in the Netherlands. Each of the panels contains the seasonally adjusted original
data, the trend and the cyclical component separated by the HP-filter with the smoothing parameter
λ = 129, 600. In Figure 2(a) the filter with λ = 129, 600 is unable to induce stationarity to the
cyclical component as the mean of the cyclical component changes over time, and local trends in
the series become apparent (1925–1930, 1930–1932, 1932–1936). The Dickey-Fuller (Dickey and
4
Hodrick and Prescott (1997) suggest λ = 14, 400 for monthly data, while Ravn and Uhlig (2002) argue that
λ = 129, 600 for monthly data would be more appropriate, because it performs better in filtering out the same trend
from the same data at different frequencies.
7
(a) Stock Market (Netherlands)
(b) Total Unemployment (Netherlands)
Figure 2: Application of the HP-filter using the Ravn-Uhlig Criterion (λ = 129, 600)
Fuller, 1979) test confirms the visual analysis, and is unable to reject the null hypothesis of a unit
root. In contrast, eyeballing and the Dickey-Fuller test find the unemployment series in Figure 2(b)
to be stationary.5
In a heterogeneous dataset, a mix of stationary and non-stationary series creates a serious problem for PCA. Because of their higher degree of correlations, the non-stationary series tend to
dominate the first component. This imposes obviously a problem for our empirical strategy. Thus,
we propose an algorithm that initially runs the HP-filter with λ = 129, 600. Thereafter, the algorithm runs a Dickey-Fuller test to check for the existence of a unit root. If it is not able to reject
non-stationarity, the algorithm reduces λ by 10x , where x is an arbitrary value chosen as a compromise between precision and computing time, and tests again for a unit root, until stationarity
is ensured. For this paper, we run the algorithm manually for the (very few) cases, in which the
HP-filter with λ = 14, 400 is unable to induce stationarity. The advantage of this approach is that
it is rule-based, and that one could even argue that a “one size fits all” parameter is overly ad-hoc.
In sum, we choose individual smoothing parameters for the trend-cycle decomposition, since
the “standard” values do not always induce stationarity. Despite the econometric importance of
de-trending, there is also a more substantial question attached to this: In times of a major crisis,
what is trend and what is cycle? The trend-cycle decomposition literature usually operates with
decades of annual time series within a growth regime. In our case, between the mid-1920s to the
mid-1930s, this long run trend was presumably constant for most economies. Thus, the Great
5
This problem would be smaller if we took logarithms before de-trending. Unfortunately, there are zero observations in the dataset, e.g. in bankruptcies. An alternative is to replace the zeros by infinitesimal small values and then
take natural logarithms.
8
Depression and the recovery afterwards are actually the cycle, not the trend. Yet, all standard
smoothing values advise filtering out large parts of this “trend” – which creates a major paradox.
As the next subsection will show, we therefore chose an agnostic aggregation procedure, applicable
to stationary and non-stationary data, which leaves the decision of this question ultimately to the
reader.
3.3. Aggregation
There are several techniques to aggregate single time series to a business cycle index. Historically, they include averaging, principal component analysis, and various versions of factor analysis
including classical and Bayesian dynamic factor analysis (Stock and Watson, 2002; Kose et al.,
2003).
Any form of averaging is certainly the most basic method of aggregation. To choose appropriate weights, however, is quite a challenging task, especially in hindsight. A discussion in The
Economist (1933) about its own business activity index illustrates such concerns. A previous version of the index, an arithmetic average, was not considered satisfactory: “On reflection, however,
it was thought that although the general trend of the curve might be accurate enough, some of its
detailed fluctuations were undoubtedly distorted by the sudden movement of one or more series
whose importance in the economic life of the country was considerably exaggerated by the lack of
weighting.” After taking into account several measures of the importance and usefulness of every
particular series, experts assigned weights to the 18 series and calculated a weighted arithmetic
average (The Economist, 1933, p. 8).
While this kind of aggregation is certainly a good approximation for business activity, it requires much expert knowledge of the particular time period and country. Such an approach to generate business indices is still not unusual, as Klovland’s (1998) estimates for the Nordic industrial
production show.6 It does, however, not seem applicable for the scope of this paper: Estimating
business indices for 17 countries. In general, it is questionable how well a historian can perform
such a weighting compared to contemporary experts.
As opposed to this ad-hoc procedure, there are a variety of agnostic procedures to estimate
business cycles such as PCA (Rhodes, 1937).7 Today, PCA is a usual tool in econometrics and part
of many textbooks (see e.g. Jolliffe, 2002). In a nutshell, it aims to explain the variance in a dataset
6
For example, Mathy and Meissner (2011) employ Klovland’s estimates.
As Mitchell et al. (2012, p. 545) note, Stock and Watson (2002), while not aware of Rhodes’ suggestion, brought
this methodology back on the agenda. For the British interwar business cycle, Rhodes (1937, p. 37) shows the
goodness of the fit of his approach compared to the index constructed by The Economist. Appendix A provides a case
study on Belgium, illustrating that PCA is likely to reflect the state of the economy similarly but better than industrial
production indices.
7
9
of n indicator variables (our data for one country) by using n eigenvectors - the so-called principal
components. These are derived from an eigenvector-eigenvalue analysis of the variance-covariance
matrix8 of the observables, which is standard linear algebra. The first component explains the most
of the variance, the second the second most and so on. Following this idea, if one employs the
transformed variance-covariance matrix to predict the first component over time, it would resemble
an index of business activity.9 This methodology is arguably applicable to non-stationary data as
well under certain circumstances (see e.g. Hall et al., 1999). We do not aim at settling this debate
here but agnostically offer aggregates obtained from stationary and non-stationary data instead.
Our regression analysis, however, will be strictly applied to stationary series.
4. Comovement during the depression
In this section we present results from the analysis of the new dataset. The first part is a short
qualitative discussion of evidence on the timing and nature of the Great Depression. The second
part is a structural analysis of the relationship between business cycle comovement and monetary
policy during the Great Depression. It replicates the empirical setup of Mathy and Meissner (2011)
and makes use of the new dataset, which yields new results.
4.1. Timing
This section’s purpose is to date the Great Depression in a more accurate way than it has been
done before. We accomplish that by interacting two data dimensions. Firstly, we employ monthly
instead of quarterly or annual data. Secondly, the cross-sectional dimension covers 28 countries,
including 17 estimated business indices and 11 official indices of either industrial production
or business activity (see Appendix D.18). The focus of this exercise lies on the estimates with
non-stationary data for reasons stated above. Three preliminary findings emerge. First, the preDepression peak occurred in almost all countries in 1929. Second, the Great Depression hit some
agricultural economies more severely as late as 1930 and 1931. Finally, some countries showed a
short period of stabilisation in 1930 before continuing their economic free fall.
Table 2 provides a simple overview of the peaks and troughs for the period 1929–1934. We
restricted the search routine to find peaks between 1929 and 1931 and troughs between 1929 and
1934.10 In France, Lithuania, India, Denmark, Sweden, Bulgaria, and Estonia the pre-Depression
peak took place in 1930 (Table 2, column 1). For almost all other countries, it occurred in the
8
Naturally, one standardises the data before estimating the covariances.
In some cases, the second rather than the first component reflects business activity (see Appendix C).
10
We excluded 1935 and 1936 because idiosyncratic events such as the strikes in the heavy industries in Belgium
would bias the overall picture otherwise (see The Economist (1936) and Appendix A).
9
10
first two quarters of 1929. Notably, no country experienced its pre-Depression peak in the direct
aftermath of “Black Thursday,” (October 24).11 In most countries, the economy was already in
a downturn when the New York stock market crash hit the world economy. In sum, the preDepression peaks indicate that the Great Depression started in 1929 in most countries. However, it
became more pronounced, especially for agricultural economies, in 1930 and 1931 (see Appendix
D for the particular indices). Most likely, this is due to the collapse of world market prices for
agricultural goods and the onset of protectionist measures. Leaving an econometric assessment
aside at this stage, it provides some evidence for Madsen’s (2001) hypothesis that agriculture
markets constituted an important transmission channel for the Great Depression.
TABLE 2: P EAKS AND T ROUGHS IN THE G REAT D EPRESSION , 1929–1934
Country
Pre-Depression Peak
(1929–1931)
non-stationary
stationary
Depression Trough
(1929–1934)
non-stationary
stationary
Start of
Recovery
1930
Stabilisation
Gold Bloc
Belgium
France
Lithuania
Netherlands
Poland
Switzerland
1929/6
1930/2
1930/12
1929/5
1929/1
1929/4
1929/6
1931/10
1930/12
1929/8
1929/4
1931/4
1932/7
1932/7
1934/12
1932/7
1933/3
1934/11
1933/5
1933/7
1934/5
1934/3
1934/3
1934/4
1935
1935
1935
1933
-
yes
yes
yes
-
Sterling Bloc
Australia
India
Denmark
Finland
Great Britain
New Zealand
Norway
South Africa
Sweden
1929/1
1930/3
1930/7
1929/1
1929/7
1929/5
1929/8
1929/7
1930/1
1929/11
1929/11
1930/6
1930/7
1929/7
1930/6
1929/8
1931/9
1930/1
1933/3
1931/9
1932/7
1932/3
1932/9
1932/8
1931/9
1932/7
1932/7
1932/8
1932/9
1934/3
1934/6
1934/1
1934/2
1932/6
1933/10
1934/6
1933
1931
1932
1932
1932
1933
1932
1932
1932
yes
yes
yes
-
Foreign Exchange Controls Bloc
Austria
1929/8
Bulgaria
1930/1
1929/5
Czechoslovakia
Germany
1929/4
Hungary
1929/4
Italy
1929/9
Latvia
1929/5
Romania
1929/1
1929/8
1929/4
1929/5
1929/11
1929/4
1929/7
1931/1
1930/8
1933/4
1934/7
1933/3
1932/8
1933/8
1932/5
1932/7
1933/1
1934/3
1932/6
1934/4
1934/3
1932/2
1933/7
1933/9
1932/12
1933
1932
1934
1932
1932
-
yes
yes
yes
-
Other Countries with Depreciated Currencies
Canada
1929/1
Chile
1929/3
Estonia
1930/3
Japan
1929/7
UnitedStates
1929/7
1929/1
1929/4
1931/6
1929/7
1929/7
1933/2
1932/5
1933/4
1931/11
1932/7
1934/5
1933/6
1934/6
1932/12
1933/7
1933
1932
1933
1932
1933
yes
The timing of the Depression troughs varies considerably throughout the sample. For example,
Great Britain’s business activity index touched bottom only in September 1932, which is somewhat at odds with the belief that the economy recovered immediately after the suspension of the
11
See Temin (1991, p. 46) for a discussion on the irrelevance of the stock market crash for the start of the Great
Depression.
11
gold standard in September 1931. It does not necessarily contradict the idea that countries leaving
the gold standard early recovered faster (Eichengreen and Sachs, 1985), since reaching the trough
does not mean entering the recovery. Some countries, although they hit bottom at some point
between 1931 and 1934, remained depressed throughout the entire period under consideration.
Furthermore, on average members of the Sterling bloc and other countries with depreciated currencies hit bottom earlier than their counterparts from the two other blocs,12 which is consistent
with the Eichengreen-Sachs hypothesis. The longest-lasting recessions occurred in Switzerland,
Lithuania, and Bulgaria, which reached the trough only in 1934.
Eyeballing of the individual indices (see Appendix D) suggests that a small period of stabilisation in 1930 occurred in ten countries (Table 2, column 6). This does not necessarily imply that the
economic downturn did not continue, but at least it lost pace for up to six months. In most cases
the recessionary slump picked up its old pace in 1931 or even before. There is no clear pattern in
terms of bloc membership, though. A possible explanation for this short-lived stabilisation is that
the rise of protectionist policies hindered this stabilisation to continue (Eichengreen and Irwin,
1995, 2010).
It turns out that, if recovery occurred at all up to 1936, it started the latest in the gold bloc
(with Poland being the outlier). In contrast, the Sterling bloc, consisting of the British Empire
and the Nordic countries, entered the recovery quite early. The results do not suggest that foreign
exchange controls were an impediment to the start of the recovery: Four of eight foreign exchange
controls bloc members started to recover in 1932 or 1933. The fact that most members of the
group “Other Countries with Depreciated Currencies” imposed exchange controls as well and
started their recoveries in 1932 or 1933 supports this finding. This questions Eichengreen and
Irwin’s (2010, p. 894) assessment that exchange controls were an “ultimately futile effort to stem
the decline in output and rise in unemployment.”
In sum, while we did not look at the intensity of the recovery, its dating yields three interesting
results. It supports the idea that adhering to gold standard postponed the recovery (see e.g. Eichengreen and Sachs, 1985). Furthermore, exchange controls do not seem to have been an impediment
to recovery per se. Finally, the Great Depression hit some agricultural economies more severely
only in 1930 and 1931.
4.2. The role of exchange controls revisited
This section illustrates another application of our dataset. It replicates recent work by Mathy
and Meissner (2011), while making use of the greater number of observations. Mathy and Meiss12
The column showing the depression troughs based on the analysis of stationary data deviates slightly, especially
in the British case. Because it omits the downward trend for two years, a small interruption of the recovery in January
1934 shows up as the low point of the business cycle.
12
ner (2011) analyse business cycle comovement during the Great Depression and find a positive
correlation between business cycle comovement and fixed exchange rate regimes. They also find a
positive effect of exchange controls on business cycle comovement (Mathy and Meissner, 2011, p.
371), which is rather counter-intuitive. A potential explanation offered by them is sample selection
bias. With the larger sample of 26 countries for a ten-year period,13 we are able to show that their
finding on exchange controls is indeed a result of their small cross-section. When expanding the
number of countries, which is now possible, exchange controls lead to international disintegration
of business cycles.
The remainder of this section is organised as follows. Firstly, we present Mathy’s and Meissner’s empirical approach. Secondly, we discuss differences in our empirical setup. Thereafter, we
elaborate on the results. Finally, we conclude and discuss what else may be done to assess business
cycle comovement during the interwar period based on the new dataset.
4.2.1. The Method - Mathy and Meissner’s (2011) Empirical Setup
Since this section starts with a replication exercise, the following is in part a summary of the
methods section in Mathy and Meissner (2011). Mathy and Meissner (2011, p. 366) define the
dependent variable as correlation between two countries’ industrial production indices. Their
baseline sample includes ten countries and covers the period 1920–1938.14 They split up their
sample period in nine two-year periods and estimate the respective correlations for those periods.
This yields a panel structure to employ the following regression equation:
ρijt = Xijt β + γjt + µjt + δt + ijt ,
(4.1)
where ρijt is the bilateral correlation of countries i and j during the two-year-period t, and X is a
set of exogenous variables. γjt and µjt represent time-varying fixed effects to take out the effect of
idiosyncratic time-variant policy changes. δt controls for period-specific shocks and represents
the pair-specific error term. In some instances, they also include pairwise country fixed effects.
As independent variables, they employ a measure of bilateral trade activity, a “peg-measure,”
and binary indicators for being on gold and exercising exchange controls. The measure of bilateral
trade activity equals bilateral trade flows divided by the sum of two countries’ GDP in the first
year of the two-year period (Mathy and Meissner, 2011, p. 367). To measure the peg, they
modify Shambaugh’s (2004) approach of determining de facto pegs. They first count the number
13
We restrict our sample to this period and do not take into account the Chilean and Italian business cycle in order
to have a balanced sample.
14
The countries included are: Austria, Belgium Canada, Denmark, France, Japan, Norway, Sweden, the United
States and the United Kingdom. Their broader sample includes six more countries.
13
of months in which the exchange rate stayed within a 2%-band and none of the two countries
exercised exchange controls, and then divide the result by 24. They argue that this procedure
proxies the time on a de facto peg. In some regressions, they also include a binary variable that
indicates whether both countries were on the gold standard and another that indicates whether at
least one country imposed exchange controls.
4.2.2. Variations and additions
Although we aim to replicate the paper by Mathy and Meissner as closely as possible, we
change the data treatment and empirical setup to enhance the robustness of the results. The following section discusses six changes, which include (i) seasonal adjustment, (ii) an adaptation
of the HP-filter to crisis data, (iii) an alternative peg measure, (iv) a modification of the foreign
exchange controls measure, (v) the choice of different fixed effects, and (vi) the treatment of insignificant correlations. Furthermore, we reduced the length of the respective periods from 24 to
18 months, which increases the number periods.
Before going through the list, a general disclaimer may be necessary for the representativeness
of our activity indicators: While we estimate indices that describe general business activity, we
also use official industrial production indices for some countries such as the France.15 In consequence, we have a somewhat heterogeneous dataset. Since Mathy and Meissner (2011) employ
only industrial production indices, their analysis suffers from a similar problem: For three countries out of ten they build indices only based on pig iron and steel production (Mathy and Meissner,
2011, p. 366). At the same time, they employ some indices that provide a wider picture of the
industrial sector such as those estimated by Klovland (1998). However, those could also be misleading because they measure industrial production in largely agricultural economies, which might
not accurately reflect general business activity.
(i) A major difference in our data treatment is seasonal adjustment (Mathy and Meissner (2011)
do not carry out seasonal adjustment). If not correcting for seasonality, countries with similar
seasonal patterns such as agrarian countries might exhibit higher correlation than others. Since
we want to exclude this potential bias, we run a seasonal adjustment procedure on all indices as
discussed in Section 3.
(ii) Mathy and Meissner’s (2011, p. 366f) application of the HP-filter is not reproduced oneby-one. They follow the smoothing parameter setting (λ = 129, 600) suggested by Ravn and
Uhlig (2002) but as discussed in Section 3.2, stationarity is a major concern for business activity
estimates and official indices because if they are not stationary, the pairwise correlations will be
spurious. Since the period of interest in Mathy and Meissner (2011) is longer than in our case, this
15
See Appendix D.18 for the classification of the official indices.
14
must not necessarily be a problem for their results. Furthermore, they log-linearise the data before
de-trending it. To ensure stationarity of all business activity and industrial production indices, we
employ the value for the smoothing parameter for monthly data λ = 14, 400 suggested by Hodrick
and Prescott (1997).16 This procedure circumvents the non-stationarity problem that would have
occurred if we had applied the smoothing parameter suggested by Ravn and Uhlig (2002).
(iii) Mathy and Meissner’s (2011, p. 367) measure of de facto pegs might in our view be harmed
by endogeneity. The exchange rate reflects the relative value of two currencies, which depends on
relative price levels that in turn depend on economic activity, amongst others. Hence, if economic
activity moves in the same direction in both countries, relative price levels are likely to move
accordingly, ceteris paribus, and thus a change in the exchange rate is rather unlikely.
Consequently, we employ a different criterion. We calculate the percentage of time in our 18months periods in which each country was officially on the gold standard and did not devalue or
impose foreign exchange controls (Bernanke, 1995, p. 8). For each pair, we multiply the two
resulting percentages to calculate our “On Gold” variable. Since these criteria were due to policy
decisions, the endogeneity potential described above should not be present.
(iv) Mathy and Meissner (2011) employ a binary variable of foreign exchange controls assigning
“1” if any country of a pair imposed foreign exchange controls and “0” otherwise. In contrast,
we construct the measure as follows: For each country, we calculate the percentage of months in
each 18-months-period in which it imposed foreign exchange controls. We then sum up the two
fractions and call the variable “FX Controls.” This procedure accounts for differing degrees of
exchange controls exercised by each country pair.
(v) We are still in the process of collecting bilateral trade series, and therefore do not include a
variable for bilateral trade integration such as Mathy and Meissner (2011, p. 367). However,
since bilateral trade flows turn out to be very persistent over time, the application of country-pair
fixed effects compensates to a large extent for this (Mathy and Meissner, 2011, p. 368).17 Furthermore, country-pair fixed effects control for slow-changing bilateral policies or other persistent
phenomena (Mathy and Meissner, 2011, p. 368).
(vi) Finally, our large sample enables us to acknowledge the fact that the dependent variable is
itself an estimate and set correlations to zero, when their significance level is lower than 0.85. This
16
In three cases, it was necessary to decrease the smoothing parameter λ even more to induce stationarity. Those
cases are Czechoslovakia (λ = 7000), France (λ = 400), and South Africa (λ = 1500). See Section 3.2 for a
discussion of this strategy.
17
As a robustness check, Mathy and Meissner (2011, p. 368) perform a regression with country-pair fixed effects.
They comment on the insignificant coefficient of the trade variable as follows: “given the strong persistence of bilateral
trade relationships, the coefficient on trade is no longer significant.” On the persistence of trade flows, see also
Eichengreen and Irwin (1998) and Wolf and Ritschl (2011).
15
provides an important robustness check, because including insignificant correlations may lead to
understated standard errors. As it turns out, however, the results do not change significantly.18
In sum, we change the setup by Mathy and Meissner (2011) in six ways. We thereby account for
the lack of bilateral trade data and emphasise some important aspects of the empirical setup. The
most important change, however, is the number of countries included in our estimations, resulting
in 325 country pairs for each of the 7 periods. Such a large number of observations facilitate
the application of country-pair and individual time-varying country fixed effects at the same time,
which further improves the robustness of the results.19
4.2.3. Results
This section presents the regression results. The focus lies on two sub-periods, July 1927 –
December 1931 and January 1932 – June 1936, since foreign exchange controls were only introduced in the 1930s. Furthermore, pairs that effectively stayed on gold after 1932 constitute less
than 5% of the whole sample. Table 3 reports the results of the baseline specification, including
country-pair fixed effects and individual time-varying country fixed effects. Apart from “On Gold”
and “FX Controls” as explained above we also report “Av. CP FE,” the average value of country
pair fixed effects that gives an impression of the level of bilateral comovement during the period
in question, which rests on pair-specific factors such as trade. The three columns marked with a
star show results for the same specification, but treating insignificant correlations as zeros.
There are two important findings from this exercise: First, we find a positive and significant
effect for being on gold during the heyday of the interwar gold standard (1927–1931), and second,
exchange controls effectively worked against business cycle comovement.
On the first result, the magnitude of the coefficient “On Gold” is with 0.21 about the same as in
Mathy and Meissner (2011, p. 370) with 0.18. However, the estimate is insignificant in their setup.
The result is robust against interpreting insignificant pairwise correlation as zeros (column 1*). It
does neither harm the magnitude of the coefficient nor its preciseness. Mathy and Meissner (2011,
p. 370) argue that the common shock of the Depression might explain most of the variance, and
thus being on gold might therefore be found to be insignificant. However, if the gold standard was
an important transmission channel for the global crisis, the coefficient on gold should be significant
for this sub-period, too. Thus we think the small sample is more likely to be the reason for the
imprecise estimation.20 For the full period 1926–1936, we find gold to be significant as well. It
18
Of course, setting the correlation to zero is neither the most appropriate way. A proper approach would build
on a two-stage framework, correcting the standard errors in the comovement regression for the impreciseness of the
pairwise correlation estimate.
19
See Baldwin and Taglioni (2006) on this particular choice of fixed effects.
20
Moreover, Mathy and Meissner (2011, p. 370) find their peg measure constructed from exchange rates to be
16
Table 3: Regression Results by Period
Insignificant correlations
not treated as zeros
(1)
(2)
(3)
1927-1931 1932–1936 1925–1936
0.211∗
0.178
0.142∗∗
(0.117)
(0.117)
(0.063)
Insignificant correlations
treated as zeros
∗
(1 )
(2∗ )
(3∗ )
1927-1931 1932–1936 1925–1936
0.212∗
0.0408
0.068
(0.110)
(0.112)
(0.063)
-0.187∗∗
(0.095)
-0.197∗∗
(0.095)
-0.191∗∗
(0.085)
-0.145∗
(0.086)
0.0304
(0.075)
0.210∗∗∗
(0.076)
0.148∗∗
(0.060)
0.218∗∗∗
(0.069)
0.141∗∗
(0.058)
Country-Pair
Yes
Yes
Yes
Yes
Yes
Yes
TV Country
Yes
Yes
Yes
Yes
Yes
Yes
975
0.177
975
0.172
2275
0.125
975
0.138
2275
0.109
On Gold
FX Controls
Av. CP FE
0.001
(0.071)
Fixed Effects
N
adj. R2
975
0.161
Dependent variable is the pairwise correlation of the cyclical components of the business activity indices.
Standard errors in parentheses; ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
does not survive our robustness check in column 3*, though. The reason is fairly simple. Since
there are very few country pairs after 1935 (namely the gold bloc, consisting of five countries), the
sample becomes too small for a reasonable estimate – outliers would dominate the picture. Thus,
being on the gold standard increased comovement, especially during the crisis, which is in line
with the idea that the gold standard transmitted shocks throughout the world.
The second important finding is that the coefficient of “FX Controls” indicates that exchange
controls effectively reduced business cycle comovement. If policymakers aimed to make their
country more independent from global influences after the start of the crisis, this policy instrument
suited its purpose. The finding is robust for treating insignificant correlations as zeros. In contrast
to the gold standard variable, it survives even in the full period sample, although there were no
exchange controls before 1930 in our sample.
There might be a multicollinearity problem between the foreign exchange controls variable and
the time-varying country fixed effects, though. However, in a robustness check, which includes
significant for the entire period but not for the heyday of the gold standard. This may suggest that unofficial and
semi-official pegs against pound and dollar after 1931 (see League of Nations, 1940, p. 219) drive the coefficient.
17
time and country fixed effects, multicollinearity seems to be no problem (not shown). The foreign
exchange control coefficient for the sub-period 1932–1936 remains stable, whereas the gold coefficient becomes significant and larger. This result is at odds with Mathy and Meissner (2011, p.
370), who find a positive effect of exchange controls on business cycle comovement. They argue
that their result might be due to the inclusion of many Reichsmark bloc members. We can only
speculate that it might also be due to the binary nature of their exchange controls variable, or the
smaller sample compared to ours.
In sum, our analysis employing the new dataset yields two results that more easily reconcile
theory and empirics of the international effect of the Great Depression than some previous work.
For the heyday of the gold standard, we find a positive and significant influence of being on gold
on economic comovement. Furthermore, the introduction of foreign exchange controls indeed led
to a decrease of comovement in the 1930s. If it had been the policymakers’ goal to make their
economies more independent from recessionary dynamics in the aftermath of the Great Depression, they in fact succeeded.
5. Conclusion
This paper contributes to the literature on the Great Depression in three ways. Firstly, it shows
the course of the Depression in different countries. Secondly, it analyses business cycle comovement during the Great Depression. Thirdly, and most importantly, this paper provides a novel
dataset for research on the Great Depression. It allows researchers for using more frequent and
more representative business cycle indicators for more countries than ever and thus enables them
to empirically analyse hypotheses that were nearly impossible to test up to now.
The need for this new dataset can be deduced from the historiography of the Great Depression.
The Friedman-Schwartz view and the Gold Standard Literature differ in their regional focus and
methodological approach. The Gold Standard Literature took a larger number of countries into
account than the previous research and thus extended the cross-sectional dimension of the analysis. The merits of this literature are well known, but at the same time, the cross-sectional focus
obscured another dimension of datasets – the time-frequency dimension. The few articles of what
we call the Post-Gold Standard Literature show the merits of a larger time-frequency dimension,
but either suffer from a smaller cross-sectional one or focus on a particular sector of the economy.
The data underlying this paper expand all three dimensions, cross-sectional, time-frequency as
well as representativeness.
This paper demonstrates two fruitful applications of the new dataset. It describes the course of
the Depression more accurately than before. While pre-Depression peaks took place in the first
half of 1929 in most of the 28 country cases, the Depression hit some countries as late as 1930
18
or the beginning of 1931, especially agricultural economies. This is consistent with the literature
on intensified protectionist measures from 1930 and 1931 onwards (e.g. Eichengreen and Irwin,
2010). Moreover it supports the idea that the agricultural sector was an important transmission
channel of crisis dynamics (Madsen, 2001).
The second application is a structural analysis concerning crisis spillovers, an important economic policy question. For the heyday of the gold standard, we find a positive effect for gold
adherence on business cycle comovement. Moreover, we find that by restricting foreign capital
exchange in the 1930s policy-makers succeeded in making their countries more independent from
the world economy – assuming this was indeed their goal. These results profit substantially from
the broader and more frequent dataset we employ, as a comparison with Mathy and Meissner
(2011) indicates.
The future agenda for this line of research is the following. Firstly, one should analyse the
Depression in agricultural economies more deeply. Secondly, one should employ a more formal
approach for “dating the Depression.” Finally, the dataset facilitates a reasonable application of
time series econometrics to a large number of countries in order to intensify research on the propagation of the Great Depression.
19
6. References
Accominotti, O., 2012. Asymmetric propagation of financial crises during the Great Depression. Modern and comparative economic history seminar, January 26, 2012, London School of Economics and Political Science. Online
at http://eprints.lse.ac.uk/41704/ (accessed on 16.7.2013).
A’Hearn, B., Woitek, U., 2001. More international evidence on the historical properties of business cycles. Journal of
Monetary Economics 47, 321–346.
Baldwin, R., Taglioni, D., September 2006. Gravity for dummies and dummies for gravity equations. Working Paper
12516, National Bureau of Economic Research.
Bernanke, B., 1995. The macroeconomics of the Great Depression: A comparative approach. Journal of Money, Credit
and Banking 27 (1), 1–28.
Bernanke, B., James, H., 1991. The gold standard, deflation, and financial crisis in the Great Depression: An international comparison. In: Hubbard, G. N. (Ed.), Financial Markets and Financial Crises. University of Chicago Press.
Chicago, pp. 33–68.
Burns, A. F., Mitchell, W. C., 1946. Measuring Business Cycles. NBER Book Series Studies in Business Cycles.
National Bureau of Economic Research. New York.
Canova, F., 1998. Detrending and business cycle facts. Journal of Monetary Economics 41 (3), 475 – 512.
Dickey, D. A., Fuller, W. A., 1979. Distribution of the estimators for autoregressive time series with a unit root. Journal
of the American Statistical Association 74 (366a), 427–431.
Eichengreen, B., 1992. Golden fetters. The gold standard and the Great Depression, 1919–1939. Oxford University
Press. New York.
Eichengreen, B., Irwin, D., 1995. Trade blocs, currency blocs and the reorientation of world trade in the 1930s. Journal
of International Economics 38, 1–24.
Eichengreen, B., Irwin, D., 1998. The role of history in bilateral trade flows. In: The regionalization of the world
economy. National Bureau of Economic Research, pp. 33–62.
Eichengreen, B., Irwin, D., 2010. The slide to protectionism in the Great Depression: Who succumbed and why?
Journal of Economic History 70 (4), 871–897.
Eichengreen, B., Sachs, J., 1985. Exchange rates and economic recovery in the 1930s. Journal of Economic History
45 (4), 925–946.
Federal Reserve Board, 1928. Annual report of the Bank of Italy. Federal Reserve Bulletin July, 488–497.
Friedman, M., Schwartz, A. J., 1963. A monetary history of the United States, 1867–1960. Princeton University Press.
Princeton.
Hall, S., Lazarova, S., Urga, G., 1999. A principal components analysis of common stochastic trends in heterogeneous
panel data: Some monte carlo evidence. Oxford Bulletin of Economics and Statistics 61 (S1), 749–767.
Hodrick, R. J., Prescott, E. C., 1997. Postwar us business cycles: an empirical investigation. Journal of Money, Credit,
and Banking, 1–16.
Jolliffe, I., 2002. Principal component analysis. Vol. 2. John Wiley & Sons.
Karnups, V. P., 2012. The 1936 devaluation of the Lat and its effect on Latvian foreign trade. Humanaties and Social
Science Latvia. (1), 49–62.
Klovland, J., 1998. Monetary policy and business cycles in the interwar years: the Scandinavian experience. European
Review of Economic History 2 (3), 309–344.
Kose, A., Otrok, C., Whiteman, C. H., 2003. International business cycles: world, region and country-specific factors.
American Economic Review 93 (4), 1216–1239.
20
League of Nations, 1940. Statistical Yearbook 1938–39. Economic Intelligence Service. Geneva.
Madsen, J. B., 2001. Agricultural crises and the international transmission of the Great Depression. The Journal of
Economic History 61 (2), pp. 327–365.
Mathworks,
2013.
Seasonal
adjustment
using
sn×m
seasonal
filters.
Online
at
http://www.mathworks.de/de/help/econ/seasonal-adjustment-using-snxd7m-seasonal-filters.html.
Mathy, G. P., Meissner, C. M., 2011. Business cycle co-movement: Evidence from the Great Depression. Journal of
Monetary Economics 58 (4), 362 – 372.
Mattesini, F., Quintieri, B., 1997. Italy and the Great Depression: An analysis of the Italian economy,1929–1936.
Explorations in Economic History 34 (3), 265–294.
Mitchell, J., Solomou, S., Weale, M., 2012. Monthly GDP estimates for inter-war Britain. Explorations in Economic
History 49 (4), 543–556.
Ravn, M. O., Uhlig, H., 2002. On adjusting the Hodrick-Prescott filter for the frequency of observations. Review of
Economics and Statistics 84 (2), 371–376.
Rhodes, E. C., 1937. The construction of an index of business activity. Journal of the Royal Statistical Society 100 (1),
18–66.
Shambaugh, J. C., 2004. The effect of fixed exchange rates on monetary policy. The Quarterly Journal of Economics
119 (1), 301–352.
Statistische Reichsamt, 1936. Statistisches Handbuch der Weltwirtschaft. Verlag für Sozialpolitik, Wirtschaft und
Statistik. Berlin.
Statistische Reichsamt, 1937. Statistisches Handbuch der Weltwirtschaft. Verlag für Sozialpolitik, Wirtschaft und
Statistik. Berlin.
Stock, J. H., Watson, M. W., 2002. Macroeconomic forecasting using diffusion indexes. Journal of Business & Economic Statistics 20 (2), 147–162.
Temin, P., 1991. Lessons from the Great Depression. Vol. 1. The MIT Press. Cambridge.
The Economist, 1926. Belgium - government finance - debt valorisation - taxation receipts - trade balance - coal bourse. The Economist Historical Archive 1843–2008 27 November 1926, 918.
The Economist, 1931. Holland - Dutch East Indies finance - capital issues - money - saving banks - foreign trade. The
Economist Historical Archive 1843–2008 2 May 1931, 945–946.
The Economist, 1932. Holland - foreign trade - industry - commercial policy. The Economist Historical Archive
1843–2008 2 Jan 1932, 20–21.
The Economist, 1933. An index of business activity. The Economist Historical Archive 1843–2008 Supplement to the
Economist, 21 Oct. 1933, 1–8.
The Economist, 1934. A review of Lithuania’s economic situation for the year 1934. The Economist Historical Archive
1843–2008 16 Feb. 1935, 370–371.
The Economist, 1936. Strikes in Belgium. The Economist Historical Archive 1843–2008 20 Jun. 1936, 667.
Wolf, N., 2008. Scylla and Charybdis. explaining Europe’s exit from gold, January 1928–December 1936. Explorations in Economic History 45 (4), 383–401.
Wolf, N., Ritschl, A. O., 2011. Endogeneity of currency areas and trade blocs: Evidence from a natural experiment.
Kyklos 64 (2), 291–312.
21
Appendices
A. A Robustness Check - The Belgian Business Cycle, 1925–1936
The following section presents a case study on Belgium. The high correlation with the official
industrial production index from the Institut des sciences économiques (Statistische Reichsamt,
1936, 1937) verifies our use of Principal Component Analysis (PCA), especially for the the replication of Mathy and Meissner (2011). Additionally, we make use of the historical archive of The
Economist to underpin the validity of our estimates. Hence, this section provides a qualitative and
quantitative robustness check for the method. Besides this robustness check, two features of the
Belgian interwar experience attracted our attention. Belgium experienced sovereign debt problems
in the mid-1920s, which also led to a destabilisation of the business cycle (The Economist, 1926,
p. 918). Furthermore, Belgium stayed on the gold standard until March 1935 (see Appendix B).
If the gold standard hampered the economy, releasing the “golden fetters” should have fostered
the recovery (Eichengreen and Sachs, 1985). We should be able to identify these features of the
Belgian interwar experience in the data.
(b) Stationary Data (λ = 14, 400)
(a) Non-Stationary Data
Figure 3: Fit of the Indicators - Line
Figure 3 shows the plots of the official index and the business activity estimates based on nonstationary data (Figure 3(a)) and non-stationary data (Figure 3(b)).21 At first glance the correlation
seems quite high, while some differences become apparent. Apparently, the official industrial production index emphasises the downturn in mid-1936 more than the business activity estimate. Why
would this be the case, if both proxy business activity? Contemporary issues of The Economist,
covering the period 1925 to 1936, and the gold standard literature shed light on this question.
21
For the stationary case, the official index is hp-filtered with λ = 14, 400. We standardised the series such they all
have a zero mean and a standard deviation of 1 to improve the graphic comparability .
22
Hence, before going into detail about the correlation of the official and the estimated indices, it is
worthwhile to review some historical background. The period 1926–1927 exhibits strong growth.
The political re-orientation towards a balanced budget might have been the reason for this development (The Economist, 1926). Some institutional “bottlenecks” might have been removed
and facilitated rapid growth. This, however, remains speculation. After the economic free fall
from mid-1929 until mid-1932, the Belgian economy remains stagnant until early 1935. In March
1935, Belgium left the gold standard (see Appendix B). A quite significant recovery period until
mid-1936 follows, when strikes in the heavy industries occurred (The Economist, 1936). Since
the business activity index provides a broader picture of the economy, it emphasises the downturn
related to the strikes in the heavy industries less than the industrial production index does. In conclusion, narrative evidence and eyeballing underpin the quality of the business activity estimate.
(b) Stationary Data (λ = 14400)
(a) Non-Stationary Data
Figure 4: Fit of the Indicators - Scatter Plot
Another way of assessing the fit is to produce a scatter plot and draw a regression line. A
perfect correlation (all dots on one line) would be a surprising result, since we present a wider
defined index than “just” industrial production. If there was no correlation at all, our approach
would have been proven wrong. Figure 4 makes this exercise for PCA applied to stationary data
and to non-stationary data. The scatter plots illustrate that the estimates are good approximations
of the official index, which confirms the impression from the eyeballing exercise earlier on. The
coefficient for the regression line in Figure 4(a) is 0.95 with a t-statistic of 38.7 and an R2 of the
regression of about 0.91. For the stationary case (Figure 4(b)), the corresponding regression yields
a very significant correlation of 0.83 (t = 18.13) and an R2 of about 0.7.
The outliers in Figure 4(b) are compatible with the previous discussion. The industrial production index emphasises the strikes in the heavy industries more than the business activity estimate.
Another outlier seems quite interesting. In June 1929, the value for the business activity index is
significantly larger than the value of the industrial production index. This is in line with the idea
23
that industrial production indices are leading indicators. While there might have been optimism
throughout the economy in June 1929 (the pre-Depression peak), industrial production did not increase in such dimension.22 In sum, the overall fit is quite good. We now turn to the interpretation
of the coefficients that we estimated via PCA.
Table 4 shows the coefficients (weightings) for the business index that we estimated via PCA.23
Those resemble the first two columns of the matrix of eigenvectors. The appendix reports such a
table for each country. In the following we focus on the coefficients derived by PCA applied to
stationary data (last two columns).
TABLE 4: DATA B ELGIUM
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Glas
Coal
Raw Iron
Raw Steel
Steel Manufacturing
Textiles
Coal Storage
Wool Conditioning
Total Unemployment
Insured Unemployed in Percent
Index
Tons
Tons
Tons
Tons
Index
Tons
Tons
Individuals
%
14400
14400
14400
14400
14400
14400
1000
14400
14400
14400
0.24
0.10
0.22
0.21
0.24
0.24
-0.22
0.19
-0.23
-0.23
-0.15
0.25
0.18
0.19
0.12
0.01
0.13
0.06
0.16
0.17
0.12
0.20
0.30
0.29
0.29
0.14
-0.22
0.08
-0.17
-0.17
0.22
-0.20
0.21
0.25
0.24
-0.20
-0.23
-0.15
0.31
0.27
Trade
Total Imports
Total Exports
Coal Exports
Iron Exports
Machines Exports
Textile Exports
Glas Exports
Franc
Franc
Tons
Tons
Franc
Franc
Franc
14400
14400
14400
14400
14400
14400
14400
0.25
0.25
-0.01
0.18
0.23
0.26
0.25
0.11
0.16
0.32
0.15
0.14
0.04
0.02
0.28
0.32
0.08
0.21
0.19
0.27
0.23
-0.10
-0.09
-0.17
0.15
-0.17
-0.14
0.12
Prices
Wholesale Prices
Consumer Prices
Index
Index
10000
10000
0.26
0.15
0.03
0.33
0.29
0.22
0.09
-0.14
Money and Banking
Currency in Circulation
Public Deposits
Private Deposits
Bank Rate
Market Interest Rate
Franc
Franc
Franc
%
%
5000
14400
14400
14400
5000
-0.17
-0.12
-0.12
0.13
0.14
0.33
0.11
0.24
-0.38
-0.37
0.05
0.09
0.11
0.04
-0.02
0.05
-0.11
-0.18
0.34
0.37
57.48
19.09
26.72
14.32
Explained Variance in %
The coefficients in Table 4 indicate that heavy industries and prices were quite important for
the business cycle, whereas money and banking played a minor role. This looks quite different for
other countries such as Belgium’s neighbour the Netherlands (see Appendix D.13).24 Moreover,
22
This remains speculation at this stage of this research project.
It also reports the λ that we used for the HP filter and the units of the original series.
24
Typically series such as clearings or stock issues have heavy weights, because they are immediately related to the
real economy.
23
24
coal exports and the wool conditioning industry were not that decisive for the course of the business cycle. Naturally, both unemployment series have relatively high weights and a negative sign.
Hence, if the economy was growing, unemployment decreased - a quite straightforward interpretation. The high positive loadings for prices are strongly related to the monetary regime, i.e. the
gold standard. It forced Belgium into a severe deflation. In consequence, increasing prices (or at
least the stop of deflation) facilitated the Belgian economy to prosper.
Figure 5: Fit excluding 6 Production Series
Data Source: see Section D.2
Finally, Figure 5 resembles another robustness check for using PCA in this context. Let us
assume, there is only one production series available, which is typically coal. We drop the production of glass, wool conditioning, textiles, raw iron, raw steel and steel manufacturing from
the sample. How much would the “fit” of the method change? The goodness of the fit decreases
(Figure 5). However, considering that we excluded almost all series that represent some form of
industrial production themselves, the fit is surprisingly good. The underlying reason might be that
trade variables proxy production quite well.
In sum, the Belgian business activity estimate does not only match its official counterpart,
but is also consistent with contemporary reports of The Economist. Anecdotal and quantitative
evidence supports the validity of PCA for the aggregation of individual indicator series. It is
a good approximation of the industrial production index, even if industrial production indicator
variables themselves are excluded. Naturally, one could also argue that the estimate captures the
state of the economy better than an industrial production index, which is usually thought of as a
proxy itself.
25
B. Gold Adherence and Introduction of Foreign Exchange Controls - Overview
TABLE 5: A DHERENCE TO G OLD AND I NTRODUCTION OF F OREIGN E XCHANGE C ONTROLS
Country
Return to Gold
Suspension of Gold Standard/Devaluation
Introduction of Foreign Exchange Controls
Gold Bloc
Belgium
France
Lithuania
Netherlands
Poland
Switzerland
Oct-26
Jun-28
Jan-25
Apr-25
Oct-27
Jan-25
Mar 35- Apr 35
Oct-36
Oct-36
Oct-36
Sep-36
Oct-35
Apr-36
-
Sterling Bloc
Australia
India
Denmark
Finland
Great Britain
New Zealand
Norway
South Africa
Sweden
Apr-25
Jan-25
Jan-27
Jan-26
May-25
Apr-25
May-28
Jan-25
Apr-24
Dec-29
Sep-31
Sep-31
Oct-31
Sep-31
Apr-30
Sep-31
Dec-32
Sep-31
Nov-31
Oct-1931–Dec-1931
-
Foreign Exchange Controls Bloc
Austria
Bulgaria
Czechoslovakia
Germany
Hungary
Italy
Latvia
Romania
Apr-25
Jan-29
Apr-26
Sep-24
Apr-25
Dec-27
Aug-22
Feb-29
Apr-33
Feb-34
Oct-36
-
Oct-31
Oct-31
Sep-31
Jul-31
Jul-31
May-34
Oct-31
May-32
Other Countries with Depreciated Currencies
Canada
Chile
Estonia
Japan
UnitedStates
Jul-26
Jan-26
Jan-28
Dec-30
Jun-19
Sep-31
Apr-32
Jun-33
Dec-31
Mar-33
Jul-31
Nov-31
Jul-32
Apr-33
Sources include mainly Bernanke and James (1991) and their cited references. In some cases, Statistisches Reichsamt (1936,1937) and League of Nations (1940) could
add information. Lithuania, Switzerland, India and South Africa entered the gold standard before January 1925. It was not possible to find out the exact date, though.
Sources for bloc composition: Eichengreen and Irwin (2010); Bernanke and James (1991); Wolf and Ritschl (2011).
26
C. Problems with the Estimation of the Remaining Business Activity Indices
In general both estimates, employing non-stationary data and stationary data, yield plausible
and compatible results. In some cases of the estimation with non-stationary data (Japan, Lithuania,
Bulgaria, Italy, Estonia, India), the coefficients and the associated scores for the first component
are implausible. In these instances, the second component turns out to reflect the business cycle
more accurately. This section discusses Japan as the typical case for this problem. It shows that
trends in the banking sector are too dominant in order to interpret the first component as a general
business activity indicator. This finding does only relate to the estimates with non-stationary data.
(a) Non-stationary Data
(b) Stationary Data
Figure 6: Principal Components for the Japan
Figure 6 shows the first and second component for Japan. From the comparison with an official
index that was compiled from Statistische Reichsamt (1936, p. 379) from 1929 on, it is obvious
that the first principal component could hardly reflect the business cycle. The index by the Reichsamt has the following yearly averages: 1929: 110; 1930: 102.5: 1931: 102.1; 1932: 117.1.
It seems that the Japanese economy was by no means in the downward spiral from 1925 on suggested by the first principal component in Figure 6(a). The Depression hit the country mainly in
1930 and 1931, with growth recurring from 1932 onwards. The second component suggests this
course of the Depression. A glance at the first component in the right panel (stationary estimates)
also supports this development of Japanese business activity.
If so, why would the first component in the left panel not reflect business activity as it usually
does? The analysis of the coefficients of the loadings in Table 6 sheds light on this question.
High loadings for almost all Money and Banking series indicate that the first component rather
reflects some credit tightening than a general business indicator. Moreover, more than half of the
27
Production, Transport and Employment indicators do not have the expected sign and most of the
Trade indicator series have the wrong sign, a coefficient near zero or both.
TABLE 6: DATA JAPAN
Variable Name
λ
Unit
PC1
Principal Component Coefficients
Non-Stationary
Stationary
PC2
PC1
PC2
Production, Transport and Employment
Employment (Industry)
Coal Production
Textiles Production
Silk Production
Turnover of Warehouses
Stocks of Warehouses
Transported Goods (Railway)
Wages
Real Wages
Index (1926)
1000 t
Index (1930)
t
Mill. Yen
Mill. Yen
1000 t
Index (1926)
Index (1926)
500
14400
14400
14400
14400
7000
14400
10000
14400
0.06
-0.11
-0.23
-0.12
0.08
-0.13
-0.12
0.24
0.18
0.31
0.27
0.11
-0.09
0.28
0.19
0.26
-0.04
0.10
0.15
0.19
0.16
0.03
0.14
0.11
0.29
0.12
0.27
-0.11
0.06
-0.04
0.01
0.23
0.11
-0.01
-0.09
0.01
Mill. Yen
Mill. Yen
1000 t
t
1000 t
1000 t
1000 Yen
t
t
14400
14400
14400
14400
14400
14400
14400
14400
14400
0.01
-0.03
-0.08
-0.17
-0.23
-0.18
-0.01
-0.08
0.18
0.31
0.31
0.10
-0.03
0.12
0.18
0.22
-0.04
0.12
0.26
0.28
0.06
-0.06
0.13
0.07
0.23
0.00
-0.03
0.04
0.07
0.03
0.00
0.09
0.03
-0.07
0.06
-0.06
Index (Juli 1914)
Index (Juli 1914)
3000
6000
0.21
0.23
0.19
0.17
0.27
0.29
0.04
0.06
Gold Stock (Central Bank)
Bills of Exchange and Advances (Central Bank)
Currency in Circulation (Central Bank)
Deposits by the Government (Central Bank)
Deposits by Banks (Central Bank)
Clearings
Bank Rate (Central Bank)
Market Rate
Market Rate - Demand Deposits
Clearing Banks - Cash Position
Clearing Banks - Stocks
Clearing Banks - Bills of Exchange and Advances
Clearing Banks - Deposits
Giro Banks - Savings
Stock Emissions
Mill.
Mill.
Mill.
Mill.
Mill.
Mill.
%
%
%
Mill.
Mill.
Mill.
Mill.
Mill.
Mill.
1500
14400
14400
14400
14400
14400
14400
4000
14400
14400
14400
14400
9000
14400
14400
0.23
-0.18
0.06
0.23
0.04
0.13
0.26
0.23
0.21
-0.12
-0.25
0.12
-0.22
-0.26
0.02
0.04
-0.03
0.25
0.06
-0.04
0.21
-0.06
0.01
-0.02
0.08
0.08
0.15
0.16
-0.03
0.23
0.14
-0.05
0.13
-0.06
0.18
0.14
-0.14
-0.17
-0.16
0.19
0.24
-0.03
0.20
-0.15
0.04
-0.01
0.41
0.16
0.30
0.23
-0.34
-0.14
0.01
-0.15
0.14
-0.10
-0.40
-0.32
0.29
-0.10
Explained Variance in %
%
39.00
26.47
20.87
12.95
Trade
Imports
Exports
Imports - Raw Cotton
Imports - Wool
Imports - Coal
Imports - Pig Iron
Imports - Machines
Exports - Raw Silk
Exports - Cotton Threads
Prices
Wholesale Prices
Consumer Prices (Tokyo)
Money and Banking
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Hence, it is important to check the coefficients and then decide whether or not, the first component could accurately reflect business activity. We could verify our suggestion that credit tightening drives the first component by simply excluding Money and Banking series from the estimation. Figure 7 shows that our analysis was right. The first principal component scores now exhibit
turning points similar to those suggested by the estimate of the first principal component with stationary data (Figure 6(b)) and those suggested by the estimate of the second principal component
with non-stationary data (Figure 6(a)).
In general, production and employment indicators should have relatively high coefficients and
yield the expected signs in order to interpret the particular component as economic activity. The
development of a proper econometric rule-based framework lies beyond the scope of this paper.
28
Figure 7: Japanese Business Activity Index, 1925–1936
Comment: first principal component excluding Money and Banking indicators.
Eyeballing suggests that the problem occurs only with the business cycle estimates (with nonstationary data) in the cases of Japan, Lithuania, Bulgaria, Italy, Estonia, and India.25 To enable
the reader to judge the validity of this ad-hoc procedure, coefficient tables for each country can
be found in Appendix D. In section 4.1, the second principal component provides a proxy for
business activity for the countries mentioned above. For all other countries, we employ the first
one. The model in Section 4.2 relies on the estimates with stationary data. Hence, the problems
described in this section do not apply.
25
The Italian case is ambiguous to some extent. A comparison with the index of the Ministry of Corporations (see
Mattesini and Quintieri, 1997), however, suggests that the second rather than the first component reflects business
activity.
29
D. Data - Countries
D.1. Australia
TABLE A 1: DATA AUSTRALIA
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Agriculture
Wheat Stocks
1000 t
14400
-0.19
-0.12
-0.02
-0.06
Trade
Imports
Exports
Imports - Food and Beverages
Imports - Raw and Semi-processed Goods
Imports - Processed Goods
Exports - Food and Beverages
Exports- Raw and Semi-processed Goods
Exports - Wheat
Exports - Budder
Exports - Wool (non-washed)
Exports - Wool (washed)
Exports - Gold and Silver
Prices
1000 Pounds
1000 Pounds
1000 Pounds
1000 Pounds
1000 Pounds
1000 Pounds
1000 Pounds
1000 t
1000 t
1000 t
1000 t
1000 Pounds
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
0.30
0.20
0.31
0.29
0.30
0.00
0.22
-0.16
-0.27
-0.05
-0.15
-0.06
0.09
0.42
0.02
0.07
0.10
0.09
0.43
-0.16
0.01
0.31
0.46
-0.17
0.42
-0.10
0.38
0.39
0.40
-0.13
-0.04
-0.24
-0.12
-0.07
-0.02
0.17
0.03
0.48
0.08
0.00
0.04
0.07
0.53
0.02
-0.23
0.45
0.38
-0.10
Wholesale Prices
Food Prices (Consumer)
Index (1911)
Index (1923/27)
14400
14400
0.31
0.31
0.00
-0.13
0.30
0.36
0.13
0.04
Money and Banking
Gold Stock (Central Bank)
Currency in Circulation (Central Bank)
Bank Rate (Central Bank)
Mill. Pounds
Mill. Pounds
%
14400
14400
14400
0.26
0.28
0.24
-0.28
-0.15
-0.33
0.11
0.04
-0.06
-0.12
0.11
-0.10
54.18
13.28
25.47
11.64
Explained Variance in %
Comments
→ Number of total series: 18
→ Broken Stick Criterion (non-stationary | stationary data) : 4 | 6
30
(a) Non-stationary Data
(b) Stationary Data
Figure A 1: Principal Components for Australia
31
D.2. Belgium
TABLE A 2: DATA B ELGIUM
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Glas
Coal
Raw Iron
Raw Steel
Steel Manufacturing
Textiles
Coal Storage
Wool Conditioning
Total Unemployment
Insured Unemployed in Percent
Index
Tons
Tons
Tons
Tons
Index
Tons
Tons
Individuals
%
14400
14400
14400
14400
14400
14400
1000
14400
14400
14400
0.24
0.10
0.22
0.21
0.24
0.24
-0.22
0.19
-0.23
-0.23
-0.15
0.25
0.18
0.19
0.12
0.01
0.13
0.06
0.16
0.17
0.12
0.20
0.30
0.29
0.29
0.14
-0.22
0.08
-0.17
-0.17
0.22
-0.20
0.21
0.25
0.24
-0.20
-0.23
-0.15
0.31
0.27
Trade
Total Imports
Total Exports
Coal Exports
Iron Exports
Machines Exports
Textile Exports
Glas Exports
Franc
Franc
Tons
Tons
Franc
Franc
Franc
14400
14400
14400
14400
14400
14400
14400
0.25
0.25
-0.01
0.18
0.23
0.26
0.25
0.11
0.16
0.32
0.15
0.14
0.04
0.02
0.28
0.32
0.08
0.21
0.19
0.27
0.23
-0.10
-0.09
-0.17
0.15
-0.17
-0.14
0.12
Prices
Wholesale Prices
Consumer Prices
Index
Index
10000
10000
0.26
0.15
0.03
0.33
0.29
0.22
0.09
-0.14
Money and Banking
Currency in Circulation
Public Deposits
Private Deposits
Bank Rate
Market Interest Rate
Franc
Franc
Franc
%
%
5000
14400
14400
14400
5000
-0.17
-0.12
-0.12
0.13
0.14
0.33
0.11
0.24
-0.38
-0.37
0.05
0.09
0.11
0.04
-0.02
0.05
-0.11
-0.18
0.34
0.37
57.48
19.09
26.72
14.32
Explained Variance in %
Comments
→ Number of total series: 24
→ Broken Stick Criterion (non-stationary | stationary data) : 3 | 7
32
(a) Non-stationary Data
(b) Stationary Data
Figure A 2: Principal Components for Belgium
33
D.3. Bulgaria
TABLE A 3: DATA B ULGARIA
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Coal Production
Transported Goods (Shipping - Incoming)
1000 t
1000 NRT
14400
14400
-0.13
-0.16
0.42
0.37
0.21
-0.06
0.07
0.12
Trade
Imports
Exports
Mill. Leva
Mill. Leva
14400
14400
0.27
0.24
0.07
0.15
0.30
-0.05
-0.16
-0.06
Prices
Wholesale Prices
Wholesale Prices - Foodstuff (no meat)
Wholesale Prices - Foodstuff (meat only)
Consumer Prices
Consumer Prices - Foodstuff
Index(1932/34)
Index(1932/34)
Index(1932/34)
Index(1933/34)
Index(1933/34)
5000
5000
14400
500
500
0.29
0.28
0.29
0.28
0.29
0.06
-0.07
0.02
0.03
0.02
0.42
0.40
0.38
0.33
0.34
0.08
-0.10
0.08
0.32
0.26
Money and Banking
Gold Stock (Central Bank)
Foreign Exchange (Central Bank)
Bills of Exchange and Advances (Central Bank)
Advances to the Government (Central Bank)
Currency in Circulation (Central Bank)
Deposits (Central Bank)
Bank Rate (Central Bank)
Protested Bills of Exchange (Central Bank)
Mill.
Mill.
Mill.
Mill.
Mill.
Mill.
%
Mill.
14400
4.25
4.25
10000
5000
10000
10000
10000
-0.26
0.17
0.26
0.20
0.28
0.18
0.24
0.19
0.25
0.37
0.22
-0.47
0.11
-0.23
0.12
0.32
0.12
0.02
-0.02
-0.10
0.13
0.30
-0.13
-0.05
0.06
-0.08
-0.03
-0.25
-0.42
-0.27
0.41
0.52
67.83
12.00
26.72
14.32
Leva
Leva
Leva
Leva
Leva
Leva
Leva
Explained Variance in %
Comments
→ Total series: 17
→ Broken Stick Criterion (non-stationary | stationary data) : 3 | 6
→ Price indices were matched by re-basing them.
→ We converted the pre-1928 gold stock values by employing the factor 3.83/100 using the information of the series
“Goldwert der Valuta.”
34
(a) Non-stationary Data
(b) Stationary Data
Figure A 3: Principal Components for Bulgaria
35
D.4. Chile
TABLE A 4: DATA C HILE
Variable Name
Production, Transport and Employment
Mining - Production Index
Mining - Employed in Salpetre Industry
Mining - Employed in Cooper Industry
Cooper Production
Coal Production
Retail Sailes (Santiago)
Trade
Imports
Exports
Exports - Wool (washed)
Exports - Cooper
Money and Banking
Clearings
Bank Rate (Central Bank)
Savings
Stock Market Index
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Index (1927/29)
1000
1000
1000 t
1000 t
Index
(1932/1934)
14400
1000
1000
14400
14400
14400
0.38
0.36
0.36
0.33
0.11
0.31
-0.03
-0.15
-0.10
0.10
0.42
0.13
0.29
0.07
0.16
0.20
0.33
0.30
0.36
-0.14
0.25
0.47
-0.11
-0.29
Mill. Pesos
Mill. Pesos
t
1000 t
14400
14400
14400
14400
0.32
0.34
0.03
0.30
-0.24
-0.22
-0.01
0.08
0.39
0.29
0.03
0.16
-0.01
0.30
0.07
0.43
Mill. Pesos
%
Mill. Pesos
Index (1927)
14400
14400
1000
5000
0.20
0.13
0.08
0.11
0.39
-0.40
0.42
0.41
0.42
-0.24
0.20
0.32
-0.19
0.24
-0.19
-0.25
44.48
30.92
27.51
21.8
Explained Variance in %
Comments
→ Total series: 14
→ Broken Stick Criterion (non-stationary | stationary data) : 3 | 5
→ Workers in Saltpetre Industry: Value for February 1934 was missing and linearly interpolated.
→ Retail Sales (Santiago): Two indices were linked via indexing. One is based on 20 companies and their branches,
the other one on 18.
→ Exports - Wool (washed): For January–March 1929, only the quarterly value was available. Hence monthly values
have been assumed to be 1/3 of the quarterly value for each month.
36
(a) Non-stationary Data
(b) Stationary Data
Figure A 4: Principal Components for Chile
37
D.5. Estonia
TABLE A 5: DATA E STONIA
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Employment (Mining and Industry)
Transported Goods (Railway)
Transported Goods (Shipping - Incoming)
number
1000 t
1000 NRT
14400
14400
14400
0.00
-0.04
-0.04
0.25
0.37
0.23
0.25
0.23
0.12
0.16
0.08
0.30
Trade
Imports
Exports
Exports - Flax
Exports - Butter
Exports - Eggs
Exports - Sawn Wood
Exports - Paper
1000 Ekr
1000 Ekr
t
t
1000s
1000 cbm
t
14400
14400
14400
14400
14400
14400
14400
0.30
0.30
0.16
-0.02
-0.08
0.14
0.30
0.13
0.13
-0.10
0.26
0.20
-0.03
0.01
0.31
0.30
0.01
0.09
0.06
0.05
0.03
0.01
0.18
0.08
0.26
0.16
0.03
0.35
Prices
Wholesale Prices
Consumer Prices (in Reval)
Consumer Prices - Foodstuff (in Reval)
Unemployed
Index (1913)
Index (1913)
Index (1913)
Anzahl
10000
14400
10000
14400
0.33
0.33
0.34
-0.16
-0.06
0.05
-0.03
-0.16
0.26
0.20
0.18
-0.24
-0.29
-0.35
-0.28
-0.28
Money and Banking
Gold Stock (Central Bank)
Foreign Exchange (Central Bank)
Bills of Exchange and Advances (Central Bank)
Currency in Circulation (Central Bank)
Clearings
Bank Rate (Central Bank)
Private Banking - Bills of Exchange and Advances
Private Banking - Deposits
Protested Bills of Exchange
Mill. Ekr.
Mill. Ekr.
Mill. Ekr.
Mill. Ekr.
Mill. Ekr.
%
Mill. Ekr.
Mill. Ekr.
1000 Ekr.
10000
10000
144000
144000
144000
144000
144000
144000
144000
-0.24
0.20
0.25
0.04
0.17
0.29
0.04
-0.02
0.19
0.16
0.21
-0.23
0.20
0.32
-0.18
0.33
0.39
0.06
0.09
0.29
-0.26
0.13
0.27
0.07
0.32
0.34
0.03
-0.09
-0.19
0.25
-0.16
0.10
0.29
0.08
-0.02
0.17
Explained Variance in %
%
35.93
25.98
24.23
11.70
Comments
→ Total series: 23
→ Broken Stick Criterion (non-stationary | stationary data) : 3 | 7
→ “Flax Exports” have been linearly interpolated for two month (July 1928 and October 1930).
→ The “Employment (Mining and Industry)” series has been linearly interpolated from July–December 1926.
38
(a) Non-stationary Data
(b) Stationary Data
Figure A 5: Principal Components for Estonia
39
D.6. Finland
TABLE A 6: DATA F INLAND
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Wholesale Turnover
Transported Goods (Railway)
Transported Goods (Shipping)
Bankruptcies - Total
Bankruptcies - Agriculture
Mill Fmk
1000t
1000 NRT
real number
real number
14400
14400
14400
14400
14400
0.26
0.27
0.27
-0.15
-0.14
-0.06
-0.11
0.05
0.28
0.30
0.23
0.25
0.18
-0.20
-0.22
0.18
0.00
0.04
0.01
-0.03
Trade
Imports
Exports
Imports Stone Coal and Coke
Imports Cotton
Exports Butter
Exports Wood
Exports Cellulose
Exports Wood Pulp
Exports Paper
Mill Fmk
Mill Fmk
1000t
t
t
1000 cbm
1000t
1000t
1000t
14400
14400
14400
14400
14400
14400
14400
14400
14400
0.10
0.20
0.19
0.19
-0.09
0.06
0.23
0.23
0.26
-0.26
-0.17
0.09
-0.06
0.08
-0.18
0.18
0.16
0.12
0.27
0.14
0.06
0.12
-0.08
0.10
0.03
0.04
0.11
0.15
0.22
0.00
-0.05
0.01
0.13
0.26
0.12
0.09
Prices
Wholesale Prices
Consumer Prices
Index (1926)
Index (=1914)
14400
14400
-0.08
-0.14
-0.34
-0.24
0.16
0.15
0.27
0.28
Money and Banking
Gold Stock (Central Bank)
Foreign Exchange (Central Bank)
Bills of Exchange and Advances (Central Bank)
Currency in Circulation (Central Bank)
Deposits (Central Bank)
Clearings - Foreign and Domestic Stocks
Bank Rate (Central Bank)
Private Banking - Bills of Exchange and Advances
Private Banking - Deposits
Saving Institutions - Savings
New Life Insurances
Stocks
Stock Turnover
Protested Bills of Exchange
Mill Fmk
Mill Fmk
Mill Fmk
Mill Fmk
Mill Fmk
Mill Fmk
%
Mill Fmk
Mill Fmk
Mill Fmk
Mill Fmk
Index
Mill Fmk
Mill Fmk
3000
14400
3000
14400
14400
14400
14400
14400
14400
7000
14400
8000
14400
14400
0.22
0.16
0.04
0.11
0.13
0.25
-0.27
-0.03
0.13
0.23
0.08
0.25
0.07
-0.15
-0.05
-0.10
0.14
-0.21
0.27
-0.06
-0.11
0.19
0.27
0.23
-0.12
-0.09
-0.16
0.23
0.06
0.10
-0.05
0.27
0.31
0.33
-0.12
0.14
0.31
0.14
0.05
0.28
0.06
-0.19
0.00
-0.38
0.41
0.02
-0.16
-0.05
0.32
0.31
-0.16
0.04
-0.05
-0.19
-0.14
0.06
35.13
24.11
17.92
15.47
Explained Variance in %
Comments
→ Total series: 30
→ Broken Stick Criterion (non-stationary | stationary data) : 5 | 9
→ Values of the Gold Stock for 1925 have been converted into the parity, which was in law from 1926 on.
40
(a) Non-stationary Data
(b) Stationary Data
Figure A 6: Principal Components for Finland
41
D.7. Hungary
TABLE A 7: DATA H UNGARY
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Brown Coal Mining
Iron Mining
Transported Goods (Railway)
Number of Applications per 100 Job Advertisements
Unemployed Union Members
1000 t
1000 t
1000 t
Individuals
1000s
14400
14400
14400
14400
14400
0.00
0.07
0.19
-0.17
-0.06
0.33
0.30
0.20
-0.08
-0.35
0.17
0.06
0.07
-0.08
-0.11
0.13
0.16
0.32
-0.23
-0.27
Trade
Imports
Exports
Imports - Raw Cotton
Imports - Wood
Imports Coal and Coke
Imports Machines
Imports - Cotton Materials
Exports - Wheat
Exports - Flower
Exports - Cattle
Exports - Pigs
Mill. Pengö
Mill. Pengö
t
1000 t
1000 t
1000 Pengö
t
1000 t
1000 t
1000s
1000s
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
0.21
0.19
-0.18
0.21
0.20
0.21
0.21
-0.03
0.17
0.05
-0.01
0.09
0.08
0.13
0.07
0.11
0.11
-0.04
0.16
0.06
-0.03
0.17
0.16
-0.10
0.02
0.12
0.07
0.11
0.08
-0.04
0.06
-0.20
-0.07
0.31
0.25
0.16
0.21
0.27
0.19
0.16
0.15
0.24
0.16
0.06
Prices
Wholesale Prices - Price-elastic Goods
Wholesale Prices - All Goods
Wholesale Prices - Agricultural Goods
Wholesale Prices - Industry
Wholesale Prices - Wheat
Consumer Price Index
Consumer Price Index - Foodstuff
Consumer Price Index - Textiles
Index (1925/27)
Index (1913)
Index (1913)
Index (1913)
Pengö per 100 kg
Index (1913)
Index (1913)
Index (1913)
14400
14400
14400
14400
14400
14400
14400
14400
0.22
0.21
0.21
0.22
0.20
0.20
0.22
0.20
0.02
-0.06
-0.04
-0.03
-0.09
0.10
0.00
-0.08
0.04
0.31
0.30
0.28
0.13
0.31
0.31
0.21
0.07
-0.15
-0.15
0.06
-0.04
-0.04
-0.05
0.02
Money and Banking
Gold Stock (Central Bank)
Foreign Exchange (Central Bank)
Bills of Exchange and Advances (Central Bank)
Advances to the Government (Central Bank)
Currency in Circulation (Central Bank)
Deposits by the Government (Central Bank)
Deposits Others (Central Bank)
Bank Rate (Central Bank)
Market Rate
Current Account(Private Banks)
Savings (Private Banks)
Mill.
Mill.
Mill.
Mill.
Mill.
Mill.
Mill.
%
%
Mill.
Mill.
14400
6000
14400
10000
14400
10000
10000
14400
14400
14400
5000
0.13
0.16
-0.18
0.19
0.17
0.18
-0.10
0.15
0.16
-0.05
-0.14
0.26
-0.22
0.14
-0.14
0.21
0.03
-0.06
-0.17
-0.23
0.35
0.28
0.02
-0.06
0.26
-0.13
0.10
0.22
0.12
0.23
0.23
0.11
-0.12
0.06
-0.03
-0.05
-0.08
-0.10
-0.04
-0.21
-0.04
-0.06
0.22
0.27
56.99
17.01
22.12
12.61
Pengö
Pengö
Pengö
Pengö
Pengö
Pengö
Pengö
Pengö
Pengö
Explained Variance in %
Comments
→ Total series: 35
→ Broken Stick Criterion (non-stationary | stationary data) : 5 | 10
→ Wholesale prices of “All Goods”, “Agriculture Goods”, and “Industry” are based on a new series from 1929 on.
However, the Reichsamt already matched the indices. They exhibit no structural breaks.
→ The “Gold Stock” series shows little variation from 1932 on, which might be due to new regulation.
42
(a) Non-stationary Data
(b) Stationary Data
Figure A 7: Principal Components for Hungary
43
D.8. India
TABLE A 8: DATA I NDIA
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Cotton Thread Production
Processed Cotton Production
Transported Goods (Shipping - Incoming)
Sea Freight Index
1000t
1000t
1000 NRT
Index
144000
144000
144000
144000
-0.24
-0.25
0.00
0.24
0.10
0.14
0.42
-0.02
0.05
0.03
0.13
-0.02
0.49
0.46
-0.14
0.31
Trade
Imports
Exports
Imports Gasoline
Imports Iron and Iron Goods
Imports Machines
Imports Processed Cotton
Exports Raw Cotton
Exports - Raw Jute
Exports - Processed Cotton in Pieces
Exports - Jute Goods
Exports - Saks
Mill Rupies
Mill Rupies
1000 hl
1000t
Mill Rupies
Mill Meter
1000t
1000t
Mill Meter
1000t
1000t
144000
144000
144000
144000
144000
144000
144000
144000
144000
144000
144000
0.29
0.29
0.05
0.28
0.22
0.26
0.10
0.09
0.27
0.21
0.10
0.02
-0.01
0.23
0.07
0.28
0.02
0.23
0.26
-0.11
0.28
0.37
0.35
0.34
0.09
0.24
0.28
0.31
0.06
0.08
0.00
0.25
0.25
0.09
-0.23
-0.03
-0.10
-0.19
0.21
-0.32
-0.06
0.30
0.04
0.05
Prices
Wholesale Prices - Calcutta
Wholesale Prices - Bombay
Consumer Prices (Bombay)
Index (1914)
Index (1914)
Index (1914)
144000
144000
4000
0.30
0.29
0.29
-0.07
-0.12
-0.08
0.37
0.38
0.26
-0.06
0.07
0.23
Money and Banking
Bank Rate (Imperial Bank of India)
Stock Issues
Stock Value of 5 Indian Railway Companies in London
%
Mill Rupies
Mill Pound
144000
144000
5000
0.18
-0.08
-0.05
-0.17
0.23
0.44
-0.06
0.01
0.12
-0.05
0.04
-0.11
Explained Variance in %
%
51.55
15.88
22.45
15.01
Comments
→ Total series: 21
→ Broken Stick Criterion (non-stationary | stationary data) : 5 | 7
44
(a) Non-stationary Data
(b) Stationary Data
Figure A 8: Principal Components for India
45
D.9. Italy
TABLE A 9: DATA I TALY
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Raw Steel Production
Transported Goods (Railways)
Transported Goods (Shipping - Incoming)
Transported Goods (Shipping - Outgoing)
Unemployed
Unemployed - Agriculture
Unemployed - Industry
1000t
1000t
1000t
1000t
1000
1000
1000
14400
14400
14400
14400
6000
14400
2000
0.04
0.23
-0.06
-0.11
-0.24
-0.22
-0.23
0.41
0.05
0.41
0.35
-0.07
-0.09
-0.09
0.26
0.28
0.05
0.18
-0.30
-0.21
-0.25
0.15
0.05
0.17
0.06
-0.19
-0.15
-0.17
Trade
Imports
Exports
Imports - Wheat
Imports - Raw Cotton
Imports - Wool
Imports - Wood
Imports - Coal
Imports - Scrap Metal
Imports Machines
Exports - Mandarin, Orange and Citrus
Exports - Olive Oil
Exports - Cheese
Exports - Raw Silk
Exports - Cars
Mill Lira
Mill Lira
1000t
1000t
1000t
1000t
1000t
1000t
Mill Lira
t
t
t
t
Number
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
0.24
0.24
0.17
0.12
-0.11
0.19
0.07
0.10
0.22
0.03
0.06
0.13
0.18
0.22
0.02
-0.04
-0.01
-0.01
-0.11
0.10
0.35
0.32
0.06
-0.15
0.06
-0.10
-0.09
0.06
0.28
0.28
-0.09
0.13
0.10
0.19
0.12
0.16
0.29
0.02
0.08
-0.02
0.01
0.09
0.15
-0.04
0.36
-0.03
-0.08
0.09
-0.04
0.01
-0.03
0.09
0.01
0.09
0.13
0.04
Prices
Wholesale Prices
Index (1934
1000
0.23
-0.04
0.25
-0.01
Mill. Lira
Mill. Lira
Mill. Lira
Mill. Lira
Mill. Lira
Index
(1932/1934)
Mill. Lira
%
%
Mill. Lira
1000
14400
14400
14400
14400
14400
-0.22
0.16
0.13
0.24
-0.05
0.20
0.04
0.03
-0.14
-0.03
-0.38
-0.07
0.06
-0.18
0.21
0.11
-0.08
0.17
0.15
0.28
-0.27
0.12
-0.19
0.01
14400
8000
2500
14400
-0.21
0.20
0.20
0.12
0.09
-0.12
-0.15
0.08
-0.08
0.09
0.18
0.11
0.37
-0.41
-0.34
-0.04
Mill. Lira
14400
-0.15
0.02
-0.02
0.05
52.46
12.60
22.27
9.65
Money and Banking
Gold Stock (Central Bank)
Foreign Exchange (Central Bank)
Bills of Exchange and Advances (Central Bank)
Currency in Circulation (Central Bank)
Deposits (Central Bank)
Clearings
Clearings - Giro Cheques
Bank Rate (Central Bank)
Market Rate
Capital Increases or Start-Ups for Listed Stock Corporations
Capital Decreases or Liquidations of Listed Stock Corporations
Explained Variance in %
%
Comments
→ Total series: 33
→ Broken Stick Criterion (non-stationary | stationary data) : 6 | 10
→ Values for the first three years of the gold and foreign exchange series have been converted to the 1928 parity. We
follow the explanations for Article 3 of the decree law No. 253 given in the Federal Reserve Bulletin (Federal Reserve
Board, 1928, p. 493).
→ Two wholesale price series were linked via re-basing to create the series “Wholesale Prices.”
→ The “Clearings” index was created by linking two series on clearings.
46
(a) Non-stationary Data
(b) Stationary Data
Figure A 9: Principal Components for Italy
47
D.10. Japan
TABLE A 10: DATA JAPAN
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Employment (Industry)
Coal Production
Textiles Production
Silk Production
Turnover of Warehouses
Stocks of Warehouses
Transported Goods (Railway)
Wages
Real Wages
Index (1926)
1000 t
Index (1930)
t
Mill. Yen
Mill. Yen
1000 t
Index (1926)
Index (1926)
500
14400
14400
14400
14400
7000
14400
10000
14400
0.06
-0.11
-0.23
-0.12
0.08
-0.13
-0.12
0.24
0.18
0.31
0.27
0.11
-0.09
0.28
0.19
0.26
-0.04
0.10
0.15
0.19
0.16
0.03
0.14
0.11
0.29
0.12
0.27
-0.11
0.06
-0.04
0.01
0.23
0.11
-0.01
-0.09
0.01
Trade
Imports
Exports
Imports - Raw Cotton
Imports - Wool
Imports - Coal
Imports - Pig Iron
Imports - Machines
Exports - Raw Silk
Exports - Cotton Threads
Mill. Yen
Mill. Yen
1000 t
t
1000 t
1000 t
1000 Yen
t
t
14400
14400
14400
14400
14400
14400
14400
14400
14400
0.01
-0.03
-0.08
-0.17
-0.23
-0.18
-0.01
-0.08
0.18
0.31
0.31
0.10
-0.03
0.12
0.18
0.22
-0.04
0.12
0.26
0.28
0.06
-0.06
0.13
0.07
0.23
0.00
-0.03
0.04
0.07
0.03
0.00
0.09
0.03
-0.07
0.06
-0.06
Prices
Wholesale Prices
Consumer Prices (Tokyo)
Index (July 1914)
Index (July 1914)
3000
6000
0.21
0.23
0.19
0.17
0.27
0.29
0.04
0.06
Money and Banking
Gold Stock (Central Bank)
Bills of Exchange and Advances (Central Bank)
Currency in Circulation (Central Bank)
Deposits by the Government (Central Bank)
Deposits by Banks (Central Bank)
Clearings
Bank Rate (Central Bank)
Market Rate
Market Rate - Demand Deposits
Clearing Banks - Cash Position
Clearing Banks - Stocks
Clearing Banks - Bills of Exchange and Advances
Clearing Banks - Deposits
Giro Banks - Savings
Stock Emissions
Mill.
Mill.
Mill.
Mill.
Mill.
Mill.
%
%
%
Mill.
Mill.
Mill.
Mill.
Mill.
Mill.
1500
14400
14400
14400
14400
14400
14400
4000
14400
14400
14400
14400
9000
14400
14400
0.23
-0.18
0.06
0.23
0.04
0.13
0.26
0.23
0.21
-0.12
-0.25
0.12
-0.22
-0.26
0.02
0.04
-0.03
0.25
0.06
-0.04
0.21
-0.06
0.01
-0.02
0.08
0.08
0.15
0.16
-0.03
0.23
0.14
-0.05
0.13
-0.06
0.18
0.14
-0.14
-0.17
-0.16
0.19
0.24
-0.03
0.20
-0.15
0.04
-0.01
0.41
0.16
0.30
0.23
-0.34
-0.14
0.01
-0.15
0.14
-0.10
-0.40
-0.32
0.29
-0.10
Explained Variance in %
%
39.00
26.47
20.87
12.95
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Yen
Comments
→ Total series: 35
→ Broken Stick Criterion (non-stationary | stationary data) : 4 | 10
→ The series on Bills of Exchange and Advances (Central Bank) were linked via indexing.
→ Statistische Reichsamt (1937, p. 134) reports slightly different values for the years 1933 and 1934 compared to
Statistische Reichsamt (1936, p. 385).
→ The value for December 1936 of the “Transported Goods (Railway)” series was missing and assumed to take the
same value as in November.
→ There is little variation in the gold stock series until 1930.
48
(a) Non-stationary Data
(b) Stationary Data
Figure A 10: Principal Components for Japan
49
D.11. Latvia
TABLE A 11: DATA L ATVIA
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Transported Goods (Railway)
Transported Goods (Shipping - Incoming)
Healthinsurance Members
Unemployed
Bankcruptcies (Number)
Bankcruptices (Volume)
1000t
1000 NRT
1000
real number
real number
1000 Lat
14400
14400
1000
10000
14400
14400
0.21
0.23
0.08
-0.14
-0.08
-0.03
0.23
0.04
0.37
-0.01
-0.05
-0.01
0.26
0.18
0.26
-0.30
-0.28
-0.10
0.06
0.16
0.19
-0.18
0.01
0.10
Trade
Imports
Exports
Imports - Textiles
Exports - Flax
Exports - Lineseed
Exports - Butter
Exports - Bacon
Exports - Plancks
Exports - Paper
Mill lat
Mill Lat
1000 Lat
t
t
t
t
t
t
14400
14400
14400
14400
14400
14400
14400
14400
14400
0.30
0.30
0.29
0.06
0.14
-0.12
0.00
0.14
0.17
-0.05
-0.04
-0.10
-0.21
-0.22
0.20
-0.24
0.14
0.16
0.30
0.18
0.33
-0.04
0.04
0.03
0.08
0.16
0.19
-0.11
0.17
-0.09
0.28
0.24
0.02
0.30
0.17
-0.18
Index
(1930/1933)
Index
(1930/1933)
10000
0.29
-0.12
0.11
-0.31
14400
0.28
-0.12
0.06
-0.33
Money and Banking
Gold Stock (Central Bank)
Foreign Exchange (Central Bank)
Bills of Exchange and Advances (Central Bank)
Currency in Circulation (Central Bank)
Deposits (Central Bank)
Bank Rate (Central Bank)
Private Banks - Bills of Exchange and Advances
Private Banks - Deposits
Protested Bills of Exchange
Mill Lat.
Mill Lat.
Mill Lat.
Mill Lat.
Mill Lat.
%
Mill Lat.
Mill Lat.
1000 Lat
1000
300
3000
8000
7000
14400
1000
1000
14400
-0.26
0.28
0.18
0.17
0.11
0.12
0.27
0.14
0.15
0.19
-0.04
0.11
0.28
0.35
-0.38
0.11
0.31
-0.18
-0.06
0.05
0.09
0.19
0.20
-0.12
0.32
0.30
-0.17
-0.11
-0.19
0.27
-0.18
-0.26
0.19
0.19
0.00
0.23
Explained Variance in %
%
52.46
12.60
22.27
9.65
Prices
Consumer Prices
Consumer Prices - Foodstuff
Comments
→ Total series: 26
→ Broken Stick Criterion (non-stationary | stationary data) : 5 | 8
→ For the consumer price series, new and old indices were linked.
→ The gold stock series exhibits little variance for most of the time.
→ The values for the gold stock and foreign exchange for September–December 1936 have been converted to the old
parity. The Lat was devalued by 40 % (Karnups, 2012, p. 56). See also “Goldwert der Valuta” series (Statistische
Reichsamt, 1937, p. 70).
50
(a) Non-stationary Data
(b) Stationary Data
Figure A 11: Principal Components for Latvia
51
D.12. Lithuania
TABLE A 12: DATA L ITHUANIA
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Transported Goods (Railway)
Transported Goods (Ships - Incoming)
1000 t
1000 NRT
14400
14400
-0.12
-0.26
0.31
0.07
0.17
0.21
0.18
0.19
Trade
Imports
Exports
Imported Agricultural Machines
Imports Textiles
Exports Flax
Exports Linseed
Exports Butter
Export Eggs
Export Planks
Exports Cellulose
1000 Lit
1000 Lit
1000 Lit
1000 Lit
t
t
t
t
t
t
14400
14400
14400
14400
14400
14400
14400
14400
14400
14400
0.28
0.24
0.23
0.25
0.09
0.07
-0.29
0.18
-0.14
-0.08
0.21
0.24
0.17
0.23
-0.21
-0.08
0.13
0.05
0.04
0.09
0.40
0.24
0.13
0.43
0.00
0.06
0.19
0.14
0.06
0.15
0.22
0.20
0.14
-0.03
0.05
-0.15
-0.11
-0.09
0.34
0.26
Prices
Wholesale Prices
Wholesale Prices (Flax)
Consumer Prices
Index (1913)
Lit per kg
Index (1913)
14400
14400
14400
0.33
0.26
0.33
-0.06
-0.17
-0.05
-0.09
0.03
-0.10
0.50
0.19
0.47
Money and Banking
Gold Stock (Central Bank)
Foreign Exchange (Central Bank)
Bills of Exchange and Advances (Central Bank)
Currency in Circulation (Central Bank)
Deposits (Central Bank)
Bank Rate (Central Bank)
Protested Bills of Exchange
Mill. Lit
Mill. Lit
Mill. Lit
Mill. Lit
Mill. Lit
%
1000 Lit
500
5000
5000
5000
1000
14400
14400
-0.29
0.23
-0.18
-0.10
-0.03
0.21
0.04
0.03
0.30
0.36
0.35
0.44
0.12
0.23
-0.07
0.38
-0.09
0.25
0.15
-0.20
-0.34
0.06
-0.15
0.17
0.02
-0.05
0.05
0.17
Explained Variance in %
%
40.42
20.31
19.02
11.98
Comments
→ Total series: 22
→ Broken Stick Criterion (non-stationary | stationary data) : 4 | 7
→ The Economist (1934) pictured Lithuania as an agrarian country, which was hindered in its development mainly
because of the global depression. The exports of agrarian goods fell, but even more did those of manufactured
goods (The Economist, 1934, p. 370). Interestingly, the correspondent states that "while the condition of the beforementioned economic branches was due to the results of the world crisis, more or less unsteady, it further remains
to mention briefly the stable financial and monetary situation in the country" (The Economist, 1934, p. 371). This
statement might be reflected by the high scores of the second component for the Money and Banking variables
regarding the central bank.
52
(a) Non-stationary Data
(b) Stationary Data
Figure A 12: Principal Components for Lithuania
53
D.13. Netherlands
TABLE A 13: DATA N ETHERLANDS
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
1000t
Mill hlf
1000t
1000 NRT
Real Number
%
%
%
%
%
14400
14400
14400
14400
14400
14400
4.25
4.25
4.25
14400
-0.13
0.16
0.16
0.17
-0.17
-0.21
-0.21
-0.18
-0.21
-0.19
-0.08
-0.14
-0.23
0.08
0.05
0.10
0.11
0.25
0.06
0.22
0.03
0.17
0.07
0.15
-0.29
-0.12
-0.08
-0.01
-0.05
-0.24
-0.14
-0.15
-0.07
-0.03
0.13
-0.01
0.09
0.04
0.11
0.06
%
%
14400
14400
-0.21
-0.20
0.08
0.13
-0.28
-0.28
0.06
0.14
Trade
Total Imports
Total Exports
Mill hlf
Mill hlf
14400
14400
0.21
0.21
-0.07
-0.04
0.27
0.24
0.00
0.04
Prices
Wholesale Prices
Food Prices
Index
Index
7000
10000
0.21
0.21
0.02
0.03
0.22
0.14
0.22
0.25
Money and Banking
Gold Stock (Central Bank)
Foreign Currency (Central Bank)
Bills of Exchange (Central Bank)
Currency in Circulation (Central Bank)
Deposits Total (Central Bank)
Private Deposits (Central Bank)
Clearings
Giro Cheque Turnover
Bank Rate (Central Bank)
Market Rate
Carryover Rate
Deposits Saving Banks
Total Issues of Obligation and Stocks
Stock Market Index
Stock Market Index - Dutch Stocks only
Revenues from Stock Exchange Tax
Mill Hfl
Mill Hfl
Mill Hfl
Mill Hfl
Mill Hfl
Mill Hfl
Mill Hfl
Mill Hfl
%
%
%
Mill Hfl
Mill Hfl
Index (21/25)
Index (21/25)
100 Hlf
5000
14400
14400
4000
14400
14400
14400
14400
14400
14400
14400
14400
14400
9000
9000
14400
-0.19
0.18
0.08
-0.09
-0.17
-0.16
0.21
-0.14
0.13
0.15
0.16
-0.20
0.11
0.21
0.21
0.14
-0.17
-0.22
0.27
-0.38
-0.24
-0.28
0.02
0.14
0.28
0.30
0.28
0.08
-0.14
0.03
0.02
0.12
-0.24
0.06
0.07
-0.12
-0.25
-0.25
0.22
0.09
0.08
0.11
0.09
-0.17
-0.03
0.23
0.23
0.10
0.14
-0.34
0.26
0.24
0.05
0.07
0.01
-0.13
0.36
0.33
0.38
0.15
-0.13
0.15
0.18
0.14
Explained Variance in %
%
68.26
10.00
23.88
12.09
Variable Name
Production, Transport and Employment
Coal Production
Completed Constructions
Transported Goods (Railways)
Traported Goods (Ships - Incoming)
Bankcrupcies
Insured Unemployed
Lost Days due to Unemployment (total)
Lost Days due to Unemployment (Coal Industry)
Lost Days due to Unemployment (Metal Industry)
Lost Days due to Unemployment (Foodprocessing Industry)
Lost Days due to Unemployment (Textile Industry)
Lost Days due to Unemployment (Construction)
Comments
→ Total series: 32
→ Broken Stick Criterion (non-stationary | stationary data) : 4 | 8
→ The negative score for coal production seems surprising. However, The Economist (1932, p. 21) reports: “The
Netherlands coal industry, however, continues to distinguish itself by performing a satisfactory exception to the general
depression."
→ The series "Gold Stock" exhibits a structural break in 1931. Before that date, the gold stock had shown little
variation, but grew rapidly afterwards. The foreign exchange series shows the reverse. Portfolio changes due to the
tensions on foreign exchange markets are the most likely explanation (see The Economist, 1931, p. 945).
54
(a) Non-stationary Data
(b) Stationary Data
Figure A 13: Principal Components for the Netherlands
55
D.14. New Zealand
TABLE A 14: DATA N EW Z EALAND
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Construction Permits
Incoming Ships
Unemployed
1000 Pounds
1000 NRT
Index
14400
14400
5000
0.40
-0.18
-0.41
0.16
0.37
0.01
0.28
0.12
-0.38
-0.10
0.00
0.16
Trade
Exports
Exports - Butter
Exports - Cheese
Exports - Meat
Exports - Wool
1000 Pounds
t
t
t
t
14400
14400
14400
14400
14400
0.15
-0.30
-0.14
-0.27
-0.13
0.58
0.33
0.23
0.17
0.37
0.29
0.14
0.19
0.04
0.13
0.53
0.47
0.35
0.29
0.38
Prices
Wholesale Prices
Food Prices (for consumers)
Index (1909/13)
Index (1926/30)
8000
8000
0.42
0.41
0.06
0.04
0.49
0.46
-0.20
-0.22
Money and Banking
Stocks
Index (1926)
5000
0.27
0.40
0.40
-0.14
Explained Variance in %
%
49.28
19.99
27.65
46.82
Comments
→ Total series: 11
→ Broken Stick Criterion (non-stationary | stationary data) : 2 | 4
→ The unemployment series was extrapolated with male unemployed from 1935 onwards.
→ Stock market values for December 1926, 1927, 1928, 1930 were missing and therefore linearly interpolated.
56
(a) Non-stationary Data
(b) Stationary Data
Figure A 14: Principal Components for New Zealand
57
D.15. South Africa
TABLE A 15: DATA S OUTH A FRICA
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Coal Production
Gold Production
Employed Europeans in Mining
Employed Non-Europeans in Mining
Employed Europeans in Gold Mining
Employed Non-Europeans in Gold Mining
Transported Goods (Shipping - Incoming)
Unemployed Europeans (Job Wanted Advertisements)
1000 t
t
1000s
1000s
1000s
1000s
1000 NRT
Individuals
14400
14400
14400
14400
2000
14400
14400
14400
0.18
0.10
0.22
0.25
0.27
0.27
0.23
0.08
0.29
-0.31
0.17
0.10
-0.08
-0.11
0.08
-0.38
0.16
-0.28
0.16
0.18
0.09
0.08
0.02
-0.18
0.16
-0.01
0.07
0.08
0.42
0.22
0.01
-0.25
Trade
Imports
Exports - Wool
1000 Pounds
t
14400
14400
0.12
-0.06
0.35
-0.19
0.27
-0.18
0.01
0.03
Prices
Consumer Prices
Foodstuff Prices (Consumer)
Index (1910)
Index (1910)
14400
4000
-0.19
-0.13
0.30
0.36
0.22
0.29
0.31
0.11
1000 Pounds
1000 Pounds
1000 Pounds
1000 Pounds
Mill. Pounds
%
Mill. Pounds
Mill. Pounds
Mill. Pounds
1000 Pounds
1000 Pounds
Index (1 Jan.
1923)
5000
4000
14400
14400
14400
14400
4000
4000
14400
14400
500
5000
0.27
0.11
-0.16
0.26
0.26
-0.25
0.08
0.28
-0.13
0.15
0.28
0.25
-0.01
-0.02
-0.17
0.05
0.01
0.09
0.18
0.03
-0.20
-0.33
-0.02
0.09
0.17
0.25
-0.20
0.25
0.16
-0.25
-0.01
0.30
0.13
-0.28
0.12
0.26
0.22
-0.26
0.23
0.03
-0.28
0.29
0.33
-0.18
-0.05
0.07
0.27
-0.08
52.02
23.98
31.14
13.38
Money and Banking
Gold Stock (Central Bank)
Foreign Exchange (Central Bank)
Bills of Exchange and Advances (Central Bank)
Currency in Circulation (Central Bank)
Clearings
Bank Rate (Central Bank)
Private Banking - Bills of Exchange and Advances
Private Banking - Demand Deposits
Private Banking - Long-term Deposits
Land and Agricultural Bank - Advances
Saving Banks - Deposits
Stock Market (6 Gold Mining Stocks in London)
Explained Variance in %
%
Comments
→ Total series: 24
→ Broken Stick Criterion (non-stationary | stationary data) : 4 | 6
→ Unemployed Europeans: the value for November 1931 was missing and linearly interpolated. From July 1934 on,
women are included. There is however no structural break in the series at that point.
→ Stock Market Index: the value for December 1936 was missing and assumed to take the same value as the one for
November 1936.
58
(a) Non-stationary Data
(b) Stationary Data
Figure A 15: Principal Components for South Africa
59
D.16. Switzerland
TABLE A 16: DATA S WITZERLAND
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Production of Watches
Transported Persons (Railway)
Transported Goods (Railway)
Unemployment (job requests)
Bankcrupcies
1000s
1000s
1000 t
Individuals
Number
144000
144000
144000
5000
14400
0.21
0.02
0.17
-0.20
-0.16
-0.07
0.44
0.26
-0.11
-0.08
0.12
0.12
0.20
-0.23
-0.08
-0.13
-0.06
0.10
-0.18
-0.10
Trade
Imports
Exports
Imports - Raw and Semi-processed Goods
Imports- Processed Goods
Exports - Raw Material
Exports - Processed Goods
Imports - Coal
Mill. Fr.
Mill. Fr.
Mill. Fr.
Mill. Fr.
Mill. Fr.
Mill. Fr.
1000 t
144000
144000
144000
144000
144000
144000
144000
0.20
0.21
0.21
0.18
0.20
0.21
0.00
0.13
0.03
0.06
0.24
-0.04
0.03
0.31
0.25
0.25
0.20
0.23
0.11
0.24
0.06
0.07
-0.09
-0.04
0.06
0.05
-0.11
-0.03
Prices
Consumer Prices
Food Prices
Textiles Consumer Prices
Index (1914)
Index (1914)
Index (1914)
10000
10000
10000
0.21
0.21
0.21
0.02
-0.02
-0.01
0.24
0.24
0.16
-0.08
-0.07
-0.10
Money and Banking
Gold Stock (Central Bank)
Currency in Circulation (Central Bank)
Deposits (Central Bank)
Giro Transfers (Central Bank)
Clearings
Clearings - Giro Checks
Bank Rate (Central Bank)
Market Rate
Kantonalbanken - Bills of Exchanges and Advances
Kantonalbanken - Stocks on Balance Sheet
Kantonalbanken - Mortgages
Kantonalbanken - Savings
Kantonalbanken - Advances on Current Account
Kantonalbanken - Obligations
Stock Yields
Mill.
Mill.
Mill.
Mill.
Mill.
Mill.
%
%
Mill.
Mill.
Mill.
Mill.
Mill.
Mill.
%
1000
2000
2000
14400
14400
14400
5000
1000
14400
6000
14400
8000
14400
14400
7000
-0.19
-0.20
-0.17
0.17
0.21
-0.15
0.19
0.15
-0.05
-0.21
-0.21
-0.21
-0.15
-0.20
0.18
0.01
0.03
0.04
0.22
0.06
0.31
-0.13
-0.03
0.43
0.08
0.06
0.07
0.33
0.17
0.19
-0.18
-0.18
-0.17
0.20
0.24
0.22
-0.04
-0.07
0.21
-0.14
-0.15
-0.20
0.09
-0.05
0.23
-0.18
-0.11
-0.07
0.24
0.16
0.05
-0.37
-0.17
-0.13
0.32
0.36
0.31
0.25
0.39
0.11
Explained Variance in %
%
73.05
14.50
27.53
12.75
Fr.
Fr.
Fr.
Fr.
Fr.
Fr.
Fr.
Fr.
Fr.
Fr.
Fr.
Fr.
Comments
→ Total series: 30
→ Broken Stick Criterion (non-stationary | stationary data) : 3 | 9
→ From 1936 on, there is a new parity. The old one is 1 Frank = 0.290323 gram fine gold, thereafter it is: 0.215 fine
gold. Hence the last three values of the series have been adjusted to the old parity.
→ The series stock yields has a new basis from 1934 on. However, the series shows no inconsistencies.
→ The public banks “Kantonalbanken” change their balance sheets in 1932, which leads to minor changes in the
composition of the series. However, the series exhibit no structural breaks.
60
(a) Non-stationary Data
(b) Stationary Data
Figure A 16: Principal Components for Switzerland
61
D.17. Romania
TABLE A 17: DATA ROMANIA
Variable Name
Unit
λ
Principal Component Coefficients
Non-Stationary
Stationary
PC1
PC2
PC1
PC2
Production, Transport and Employment
Oil Production
Transported Goods (Railway)
Bankcruptcies
Moratoria
1000 t
1000 t
Anzahl
Anzahl
14400
14400
14400
14400
-0.28
-0.04
0.20
-0.12
0.02
-0.32
0.25
0.32
0.24
0.33
0.01
0.03
-0.09
-0.08
-0.18
0.23
Trade
Imports
Exports
Exports - Cattle
Exports - Corn
Exports - Wood
Exports - Gasoline
Mill. Lei
Mill. Lei
1000 Stück
1000 t
1000 t
1000 t
14400
14400
14400
14400
14400
14400
0.29
0.26
0.20
-0.04
0.29
-0.29
-0.10
-0.04
0.02
0.32
-0.10
0.03
0.13
0.38
0.27
0.19
0.08
0.16
-0.06
0.17
-0.29
0.34
-0.28
-0.01
Index
(1932/1934)
14400
0.30
0.02
0.09
0.16
Money and Banking
Gold Stock (Central Bank)
Currency (Central Bank)
Clearings (in Bukarest)
Bank Rate (Central Bank)
Real Interest Rate Fixed-yield Investments
Stock Issues
Stock Market - Interest Rate of Fixed-yield Bonds
Stocks - Turnover
Protested Bills of Exchange
Mill. Lei
Mill. Lei
Mill. Lei
%
%
Mill. Lei
Index (1926)
Mill. Lei
Mill. Lei
5000
4000
14400
14400
14400
14400
14400
14400
14400
-0.27
-0.19
0.27
0.12
-0.18
0.22
0.27
0.17
0.17
-0.10
-0.24
0.08
0.48
0.32
-0.08
-0.09
-0.30
0.31
0.11
-0.25
0.42
0.33
0.02
0.11
-0.15
-0.26
-0.19
0.14
0.19
-0.15
0.01
0.50
-0.04
-0.45
-0.16
0.06
Explained Variance in %
%
49.79
12.89
15.53
14.79
Prices
Retail Prices
Comments
→ Total series: 20
→ Broken Stick Criterion (non-stationary | stationary data) : 4 | 7
→ For the “Retail Prices,” two series were linked.
→ From May 1931 on, clearings declined rapidly. However, there are no comments on the series that indicate that it
is not valid anymore.
→ From 1934 on, the series “Real Interest Rate Fixed-yield Investments” does not take foreign treasury bills into
account, which lowers its value by around 2 %.
→ Values for the Gold Stock for November and December 1936 have been converted to the old parity: 0.009 vs the
new one: 0.00625174 (Statistische Reichsamt, 1937, p. 98).
62
(a) Non-stationary Data
(b) Stationary Data
Figure A 17: Principal Components for Romania
63
D.18. Data - Countries with Official Indices on Business Activity
For some countries, there were official industrial production indices available. For the United
States and the United Kingdom, there were business activity indices available. Since those are
from various sources, we present them in this separate data section. To make the graphs comparable, all indices have been based on 1925. Moreover, all indices have been seasonally adjusted in
the same fashion as the disaggregated data.
D.18.1. France
Figure A 18: French Industrial Production
Source & Comments
Type: industrial production index
Source: Statistische Reichsamt (1936, p. 92), Statistische Reichsamt (1937, p. 38)
citep
64
D.18.2. Belgium
Figure A 19: Belgian Industrial Production
Source & Comments
Type: industrial production index
Source: Statistische Reichsamt (1936, p. 44), Statistische Reichsamt (1937, p. 20)
D.18.3. Poland
Figure A 20: Polish Industrial Production
Source & Comments
Type: industrial production index
Source: Statistische Reichsamt (1936, p. 224), Statistische Reichsamt (1937, p. 88)
65
D.18.4. Denmark
Figure A 21: Danish Industrial Production
Source & Comments
Type: industrial production index
Source: Klovland (1998)
D.18.5. United Kingdom
Figure A 22: British Business Activity
Source & Comments
Type: business activity index
Source: The State of Trade Supplement of The Economist, various issues
66
D.18.6. Norway
Figure A 23: Norwegian Industrial Production
Source & Comments
Type: industrial production index
Source: Klovland (1998)
D.18.7. Sweden
Figure A 24: Swedish Industrial Production
Source & Comments
Type: industrial production index
Source: Klovland (1998)
67
D.18.8. Austria
Figure A 25: Austrian Business Activity
Source & Comments
Type: business activity index
Source: Statistische Reichsamt (1936, p. 210; Table 2), Statistische Reichsamt (1937, p. 83, Table 2)
Comment: Indices have been linked, via re-basing them to 1934
D.18.9. Czechoslovakia
Figure A 26: Czech Industrial Production
Source & Comments
Type: industrial production index
Source: Statistische Reichsamt (1936, p. 295), Statistische Reichsamt (1937, p. 112)
68
D.18.10. Germany
Figure A 27: German Industrial Production
Source & Comments
Type: industrial production index
Source: Statistische Reichsamt (1936, p. 20, Table 10), Statistische Reichsamt (1937, p. 12, Table 10)
D.18.11. Canada
Figure A 28: Canadian Industrial Production
Source & Comments
Type: industrial production index
Source: Statistische Reichsamt (1936, p. 456), Statistische Reichsamt (1937, p. 147)
69
D.18.12. United States
Figure A 29: American Business Activity
Source & Comments
Type: business activity index
Source: Statistische Reichsamt (1936, p. 502), Statistische Reichsamt (1937, p. 165)
70