Download Reinventing Export-led Growth: Sweden in the 1930s Preliminary

Reinventing Export-led Growth: Sweden in
the 1930s
Preliminary Version - Please do not cite
without permission
Alexander Rathke∗, Tobias Straumann, and Ulrich Woitek
IERE, University of Zurich
European Historical Economics Society (EHES)
The 8th biennial Conference, Geneva, September 3-6, 2009
Abstract
Our main focus in this paper is the role of Sweden’s exchange rate
policy in Sweden’s early and strong recovery after the Great Depression. We estimate a small-scale, structural general equilibrium model
of a small open economy using Bayesian methods. We find that indeed terms of trade movements are the mayor driving force for the
strong recovery, suggesting that Sweden’s exchange-rate policy aimed
at maintaining the krona at an undervalued level significantly contributed to the growth miracle in the 1930s. This finding is supported
by archival sources of the Swedish central bank (Riksbank). They
show that the Riksbank governor was predominantly concerned about
two issues, namely the level of foreign exchange reserves and the performance of Swedish exports.
Corresponding author; Institute for Empirical Research in Economics, Winterthurerstrasse 30, CH-8006 Zurich, Switzerland, [email protected]
∗
1
This finding is relevant for the interpretation of the 1930s from a
long-term perspective. Export-led growth is not a typical phenomenon
of the post-1945 monetary system as Dooley et al. (2003) claim, but
started shortly after the end of the gold-exchange standard in 1931.
2
1
Introduction
Although a small country, Sweden has always been a preferred object of
research for scholars interested in innovative macroeconomic policies. In the
early 1990s, Sweden was among the first countries to introduce inflation
targeting after abandoning the ecu peg of the krona (Bernanke et al., 1999).
Currently, the restructuring of the banking sector in the 1990s has found
new attention (Reinhart and Rogoff, 2008). And in the era of the Great
Depression, Sweden was widely admired for its early and strong recovery from
the deepest crisis of the 20th century (Fisher, 1935). In fact, the increase
of industrial production from 1933 to 1939 was higher than anywhere else
in Europe. Accordingly, the 1930s are not only remembered as a decade of
crisis, but also of renewal (Schön, 2000).
Despite of this admiration, however, the reasons for this impressive recovery have not entirely been understood yet. The crucial question is still
the same Charles Kindleberger asked 25 years ago in his classic account of
the Great Depression. He observed that “[t]he more significant issue is the
extent to which Sweden recovered through its own efforts, and especially
through developing a public works policy to accompany cheap money, and
to what extent the recovery represented simple exchange depreciation in excess of that of the pound, plus spillover from the British building and later
armament boom.” (Kindleberger, 1973, p. 182)
In this paper we try to give a new answer to this old question by going
beyond Lundberg (1983) who was the first to study the implications of an
undervalued krona. To assess the effects of an undervalued currency, we estimate a dynamic stochastic general equilibrium (DSGE) model for a small
open economy. Our interest is not only historical, however. Inspired by the
ongoing discussion on Bretton Woods II, we ask whether Sweden pioneered
the strategy that some East-Asian countries have pursued in recent times
(Dooley et al., 2003). Our empirical results suggest that this was in fact the
case. By consciously keeping the currency undervalued the Swedish authorities fostered growth in a unprecedented way. Of course, export-led growth
had been a familiar pattern under the classical gold standard before 1914.
3
Sweden, to take the same example, also knew periods during which growth
was mainly driven by export. But by consciously keeping the exchange rate
undervalued in order to boost exports Sweden invented a new kind of growth
strategy that had been impossible under the gold standard.
The paper is structured as follows. Section 2 provides a brief summary of
the historical setting and the literature. Section 3 sketches Sweden’s exchange
rate policy on the basis of narrative and statistical evidence. The model is
explained in Section 4, and Section 5 outlines the estimation approach and
the empirical findings. The paper ends with a conclusion.
2
Recovery from the Great Depression
It is well established that those countries which abandoned the gold standard
in the early 1930s enjoyed the most rapid recovery from the crisis (Eichengreen and Sachs, 1985; Eichengreen, 1992). Sweden was among those lucky
countries. In late September 1931, a week after the fall of the British pound,
the Swedish authorities abandoned the gold standard and let the krona depreciate. Denmark and Norway took the same step and enjoyed a fast recovery
as well. By contrast, the countries of the gold bloc (Belgium, France, Italy,
Netherlands, Poland, Switzerland) which maintained the gold standard until
1935/36 suffered from a protracted economic depression.
Yet Sweden not only experienced a shorter crisis, but enjoyed a particularly strong industrial recovery after 1932. In 1938, Sweden’s index of industrial production amounted to 153 (1929 = 100) while the index in Denmark,
Norway and the UK was at 135, 130 and 127 respectively (Figure 1). Thus,
there seems to be a particularly favorable link making Swedish production
exceptionally competitive during the 1930s.
4
Figure 1: Industrial Production
Sweden
/.
.
*- +
,
+
*
) $"
(
#
'
&$% "#!
Denmark
Norway
UK
To what extent was this growth miracle due to economic policies? Schön
(2000) provides a useful survey of the Swedish literature. In the postwar
decades, the heyday of Keynesian economics, the most popular explanation
highlighted fiscal policy. Supposedly, the Swedish Social Democrats who
came to power in 1932 revived the depressed economy by pursuing a model
counter-cyclical policy, based on the teachings of the younger generation of
the Stockholm school (Gunnar Myrdal and Bertil Ohlin). The figures, however, show that the fiscal deficit was too small to account for the recovery.
The Social Democrats did not implement a new fiscal regime. This explanation is therefore hard to maintain.
A second important explanation, advocated by Jonung (1984), focuses
on monetary policy. By leaving the gold standard, the Swedish central bank
(Riksbank) was able to free itself from the deflationary train set in motion
in 1930 and to implement a more expansionary monetary policy. In particular, the newly gained freedom enabled the Riksbank to act as lender of
last resort in the aftermath of the Kreuger crash in March 1932 and to avert
a major banking crisis. Moreover, the Riksbank adopted a new monetary
policy framework, the so-called price level targeting, based on the seminal
work of the eminent Swedish economist Knut Wicksell.
5
A third possibility is offered by Lundberg (1983). He highlights the importance of a weak krona throughout the major part of the 1930s and investigates why the undervaluation did not cause inflation at an earlier stage
of the expansion. His answer is twofold. First, the demand elasticity was
rather high, i.e. foreign customers of Swedish products strongly reacted to
the lowering of prices after the end of the gold standard in 1931. Second, the
access to natural resources was relatively cheap since they were provided by
domestic producers (wood, iron ore).
Which policy was more important for Sweden’s growth miracle, monetary
policy or exchange-rate policy? Jonung’s claim that leaving the gold standard
allowed the Riksbank to save the banking system from collapsing and to
pursue a more expansionary policy can hardly be disputed (Bernanke and
James, 1991). However, whether the Riksbank contributed to the Swedish
growth miracle by adopting a modern monetary policy framework is an open
issue. Straumann and Woitek (2009) show that the Riksbank did not adopt
price level targeting after 1931, but continued to peg the krona to sterling as
in the times of the classical gold standard and the gold exchange standard
of the 1920s. Statistical evidence (Bayesian VAR) shows that there was no
regime break in the early 1930s.
In addition, statements by Riksbank governor Ivar Rooth suggest that
the Riksbank was primarily concerned about the competitiveness of the exporting sectors. It is therefore important to examine the contribution of
both monetary and exchange-rate policy by using econometric tools. Before
we present the model we provide some descriptive statistics and narrative
evidence.
3
The Exchange-Rate Policy of the Riksbank
At first sight, it appears that the exchange-rate policy of the Riksbank vis-àvis the British pound, then by far the most important trading and financial
partner, can be divided into two phases (Figure 2). In the period ranging
from the suspension of the gold standard in late September 1931 to the
official pegging to sterling in July 1933, the rate of the krona/£ appears to
6
fluctuate relatively freely. A closer look, however, reveals that there were
two major attempts to stabilize the krona at the old parity of 18.16 kronor
per £ (Jonung, 1979). In November 1931, shortly after the end of the gold
standard, the Riksbank tried to prevent the krona from rising above the old
parity of 18.16 kronor per £. The attempt failed, the exchange rate fell to
19.40 kronor per £. In late 1932, the Riksbank tried to bring the krona rate
back to this level. Again, the plan did not materialize. Thus, in the first
phase the Swedish central bank only seemingly embraced flexible exchange
rates. It is more appropriate to suppose the same policy orientation for the
whole period from September 1931 to the outbreak of the Second World War.
50
Figure 2: Nominal Exchange Rate SKr/£
56
MLJK
I
04
03
02
01
0470
0475
0477
0478
0479
:;<=>?@ABCDEFGCH
0471
0472
0473
That the Riksbank gave priority to exchange rate stabilization over price
level stabilization (Fregert and Jonung, 2004) is revealed in a number of
letters written by governor Ivar Rooth. In late September 1933, he explained
to O.M.W. Sprague, Harvard professor of economics and temporary assistant
to the United States Secretary of the Treasury:
“My personal opinion is that it is of the utmost importance to the
whole economic life of a nation which like Sweden for its standard
7
of living is to such a great extent depending upon foreign trade,
to have fairly stable quotations. I think that I dare say that also
in order to get a rising price-level, stable foreign exchanges are
better than the erratic movements of these rates which the world
has suffered from ever since September 1931.”1
In early 1936, Roth argued that Sweden pegged the krona rate pegged to
sterling because foreign business was for the most part invoiced in sterling:
“It was particularly important for a small country like Sweden
depending so strongly on its foreign trade to inspire trade and
industry with trust in our currency.”2
In a further statement in February 1938, Rooth wrote to Randolph Burgess,
Vice-President of the Federal Reserve Bank of New York:
“Some American professors, e.g. Professor Irving Fisher, believe
that it is an achievement by us in the Riksbank that prices have
been fairly steady up to the middle of 1936. I have told Professor
Fisher before and I am sorry to have to tell you now that what we
have done is merely that we have carried out a fairly conservative
central banking policy. In fact we have never tried to do anything
directly with regard to prices.”3
In short, the Riksbank was determined to serve the interests of the exporting
sectors by keeping the exchange rate as stable as possible right after the
suspension of the gold standard. Stabilising the exchange rate is one thing.
But did the Riksbank, as Lundberg argues, also maintain the krona rate at
an undervalued level? The steep rise of reserves suggests that this may have
been the case. In the first six months of 1929, that is before the beginning
of the world economic crisis, gold and foreign exchange reserves amounted
to 350 mio. SKr.; by the end of 1935, they had reached 1000 mio. SKr.
1
Archives Bank of England, OV 29/26 (26 September 1933).
Archives Bank of England, OV 29/4 (January 1936). The text is in German, the
English summary attached to the document is not very accurate.
3
Archives Sveriges Riksbank, Rooth papers, Box 129 (10 February 1938).
2
8
SONN
Figure 3: Gold and Foreign Exchange Reserves of Riksbank
SNNN
RNN
mj
l
kj QNN
hi
g
PNN
ONN
N
STOT
nopqrstquvwxtsqpqyqpzqy
{o|}pqyqpzqy
STUN
STUS
STUO
STUU
STUP STUV
XYZ[\]^_`abcdeaf
STUQ
STUW
STUR
(Figure 3). The increase of reserves was so steep that the Bank of England
observed in a memo in January 1936 that “the Riksbank is still holding an
abnormally large share of Sweden’s abnormally large foreign reserves”.4 The
dramatic improvement of the trade balance after 1931 is another sign that
the Swedish currency was presumably undervalued. In 1931/32, the trade
balance was highly negative, since Sweden as a small open economy was
immensely suffering from the collapse of world trade (figure 4). By 1933/34,
the trade deficit had disappeared, partly because exports had recovered. In
1932, the worst year of the depression, they amounted to 947 million kronor,
two years later to 1302 million kronor. In the same period, imports rose from
1155 to 1305 million kronor.
4
Archives Bank of England, OV 29/26: ’Purchases of gold by Sveriges Riksbank’.
9
Figure 4: Current Account, Visibles and Invisibles of Sweden (in mio. kronor)
400
300
200
Mio kronor
100
0
-100
-200
Overall balance
Trade balance
-300
-400
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
Source: Mitchell (2003).
As British prices were rising in 1936/37 and Sweden was concerned about
importing inflation, the exchange rate policy was put to a test. Swedish
economists such as Gustav Cassel and Bertil Ohlin were publicly suggesting
a revaluation of the krona in order to contain the import of inflation, and
investors began exchanging their pounds and dollars in kronor. As a result,
the foreign exchange reserves of the Riksbank dramatically increased starting
in the summer of 1936. The Swedish authorities, however, maintained the
parity in order to keep the competitive advantage of their exporting sectors.
When British prices began to fall in the second half of 1937, investors began
selling their krona assets.
10
Figure 5: Change in real exchange rate
“”•–”†‘…—”˜’…†—”™š›œœ”•”†–”
ƒ‚
€
‚
€
‚
~€
~‚
~€ ‡ƒ ‰‡ƒ ‡‡ ‰‡‡ ‡Š ‰‡Š ‡ ‰‡ ‡‹ ‰‡‹ ‡ ‰‡ ‡ ‰‡ ‡Ž ‰‡Ž
„…† „ˆ „…† „ˆ „…† „ˆ „…† € „ˆ € „…† „ˆ „…† Œ „ˆ Œ „…† „ˆ „…† „ˆ
†‘’
¡¢›†…‰£¤–’…†—”¥…‘”ž¦§•¨™© †œ‰…‘›†š›œœ”•”†‘›…‰žŸ“ Source: Riksbank
A straightforward approach would be to calculate the real exchange rate
of the krona against sterling on the basis of foreign and domestic prices
and the nominal exchange rate. Yet, as it is hardly possible to detect the
equilibrium of the real exchange rate, we do not pursue this approach any
further. Instead, for a first approximation, we consider only the percentage
changes of the real exchange rate. The results in figure 5 of this simple PPP
exercise are quite clear. In 1933 and 1935/36, when the nominal exchange
rate was fixed, the Swedish inflation rate was lower than the British one,
implying a strong devaluation of the Swedish real exchange rate.
4
The Basic New Open Economy Model
To support the evidence from Section 3 we estimate a small-scale, structural
general equilibrium model of a small open economy. Our theoretical framework is based on the New Open Economy Macroeconomics (NOEM). This
11
strand of the literature can be regarded as an extension of the New Keynesian
paradigm, which has been used extensively in recent theoretical and applied
work exemplified for instance, by Clarida et al. (2000) or Woodford (2003).5
These models start from the optimizing behaviour of representative agents
and feature monopolistic competition and nominal rigidities. The basic New
Keyenesian DSGE model has been adopted to the small open economy setting
by Galı́ and Monacelli (2005) and Monacelli (2005). The following section
briefly describes the basic NOEM model.
4.1
Households
Consider a small open economy populated by a infinitely-lived continuum
of households. The households try to maximize their utility defined over a
composite consumption good Ct and leisure (1 − Nt ),
U = E0
∞
X
β
t
t=0
Ct1−σ
Nt1+η
−
1−σ 1+η
,
(1)
where β is the discount factor, σ −1 is the intertemporal elasticities of substitution and η is the labor supply elasticity. The composite good Ct consists
of a Dixit-Stiglitz aggregate of domestic goods Cth and of foreign goods Ctf ,
a
a−1
a−1
a−1
1
1
a
f
h
Ct = (1 − γ) a Ct a + γ a Ct
and
Cth
=
Z
1
C(j)ht
θ−1
θ
dj
0
θ
θ−1
, j ∈ [0, 1].
(2)
(3)
The elasticity of substitution between domestic and foreign goods is measured
by the parameter a, and γ measures the degree of home bias in consumption
and is therefore a natural indicator for the openness of the economy. Choosing the optimal allocation between foreign and domestic goods for each level
5
An overview over the New Open Economy literature starting with Obstfeld and Rogoff
(1995, 1996) can be found in Lane (2001).
12
of the composite consumption goods implies the following demand functions
Cth = (1 − γ)
Pth
Pt
−a
Ct ,
Ctf
=γ
Ptf
Pt
!−a
Ct ,
(4)
1
1−a
1−a 1−a
where Pt = (1 − γ)P h
+ γP f
denotes an appropriate defined
consumer price index. The optimal allocation for domestic intermediate
goods is given by
!−θ
h
P
jt
(5)
Cth ,
Cth (j) =
h
Pt
1
1−θ
R
1−θ
1
where Pth = 0 Pjth dj
is the appropriate price index for the domestic
good. The period budget constraint can be written as
Pt Ct + Et (Qt,t+1 Bt+1 ) = Wt Nt + Bt + Tt .
(6)
Note that Pt Ct = Pth Cth + Ptf Ctf . The households have also access to a
full set of state contingent claims traded internationally denominated in the
domestic currency. Bt is the nominal payment in period t from a portfolio of
assets held at the end of period t − 1. Therefore, Et Qt,t+1 Bt+1 corresponds to
the price of portfolio purchases at time t. The nominal wage is given by Wt ,
and Tt is a lump-sum transfer or tax. The remaining optimality conditions
can be rewritten in the convenient form
Ntη
Wt
,
−σ =
Pt
Ct
−σ
Ct+1
Pt
Qt,t+1 = β
.
Ct
Pt+1
(7)
where the first describes the optimal labor/leisure choice and the second can
be rewritten as a conventional stochastic Euler relation by taking conditional
expectations on both sides and rearranging
1 = βRt Et
Ct+1
Ct
13
−σ
Pt
,
Pt+1
(8)
where Rt = Et Q1t,t+1 is a riskless nominal return of a one period bond paying
one unit of the domestic currency in period t + 1.6
4.2
Domestic and CPI inflation, Terms of Trade and
Real Exchange Rate
We assume that the law of one price holds at all times
Ptf = St Pt∗ ,
(9)
where Pt∗ is is foreign currency price of the foreign produced good and St the
exchange rates expressed as foreign currency in terms of domestic currency.
The terms of trade (the price of foreign goods in terms of domestic goods)
are defined as
∆t ≡
St Pt∗
Ptf
=
.
Pth
Pth
(10)
Note that ∆ is equal to 1 in the steady state (PPP holds). To study the effect
of an undervalued currency, we treat log ∆t as an unobservable exogenous
AR(1) process which allows us to estimate the time path of the terms of
trade, log ∆t = ρδ log ∆t−1 + ǫδt , ǫδ ∼ N (0, σδ ). The real exchange rate is
defined as
Pf
St Pt∗
Φt ≡ t =
.
(11)
Pt
Pt
4.3
Domestic Firms
There exists a continuum of monopolistic competitive firms. Each firm produces a differentiated good with an identical constant returns to scale production function
Yt (j) = Zt Nt (j),
(12)
where Zt is a autoregressive technology shock that is assumed to be identical
across firms, log Zt = ρz log Zt−1 + ǫzt . ǫz ∼ N (0, σz ). We analyze the optimal
6
For more details see Galı́ and Monacelli (2005) and Gali (2008).
14
behavior of the firms in two steps. First note that minimizing the cost of
production PWht Nt (j) subject to (12) implies
t
M Ct =
Wt /Pth
,
Zt
(13)
where M C is the real marginal cost of production and Wt /Pth the real wage.
In the second step we describe the optimal price setting behavior. We
assume staggered price setting as in e.g. Calvo (1983) and Yun (1996). A
proportion (1 − ω) of firms per period receive the random signal that they
can reset prices. The probability of a newly chosen price being effective in
period t + k is ω k , which means an average duration of a price of (1 − ω)−1 .
There is domestic and foreign demand for the intermediate good j. Firm j
faces the the overall demand
Cth (j)
=
Pth (j)
Pth
−θ
Yt ,
(14)
∗
with Yt = Cth + Cth , where the superscript ’∗’ denotes foreign demand for domestic goods. Since all firms face identical demand curves and have the same
production technologies, all firms which are allowed to reoptimize, choose the
same price P̄th in order to maximize the current value of the profits generated
while the price stays effective:
P̄th
= arg max Et
Pth (j)
∞
X
ω
τ
Qt,t+τ (Pth (j)
−
h
Pt+τ
M Ct+τ )
τ =0
Pth (j)
h
Pt+τ
−θ
Yt+τ , (15)
subject to the demand curves (14). Note that Qt,t+τ is the appropriate discount factor. The first order condition for the choice of P̄th is
∞
X
(ω)τ Qt,t+τ Yt+τ
Et
τ =0
(1 − θ)
P̄th
h
Pt+τ
−θ
+θ
P̄th
h
Pt+τ
−θ−1
M Ct+τ
!
= 0.
(16)
The other firms have to keep the price from the last period. Using the
15
definition of the domestic price index implies
h
.
Pth = (1 − ω)P̄th + ωPt−1
(17)
Combining the log-linearized version of the last to equations yields the forward looking version of the new Keynesian Phillips-curve
h
πth = βEt πt+1
+ κmct ,
(18)
with κ = (1 − ω)(1 − βω)/ω.7
4.4
Market clearing and international consumption risk
sharing
To aggregate over the firms, equate the demand and production functions
for Cth (j) and integrate both sides, which yields the following aggregate production relation
Yth =
where ςt =
R 1 Pth (j) −θ
0
Pt
Zt
Nt ,
ςt
(19)
dj. Market clearing requires
∗
Yth = Cth + Cth .
(20)
The foreign country is assumed to be large relative to the home country.
Therefore, we do not need to distinguish between foreign changes in consumer prices and overall inflation and foreign consumption and production.8
Consumption of the domestic good in the foreign country (export demand)
7
The Keynesian Phillips-curve is one of the main building blocks of the New Keynesian
Synthesis, see e.g. Clarida et al. (2000).
8
More precisely think of the domestic country as one of a continuum of infinitesimally
small countries which make up the world (foreign) economy. This means that the domestic
economy has zero mass in the foreign economy, see Galı́ and Monacelli (2005) for a more
detailed description.
16
is given by
∗
Cth
=γ
Pth
St Pt∗
−a
Ytf ,
(21)
if the foreign households have the same preferences as the domestic households. We assume that the log Ytf follows an exogenously given stable AR(1)
process:
f
log Ytf = log(1 − ρf )Y f + ρf log Yt−1
+ ǫft , ǫf ∼ N (0, σ f ).
The existence of complete financial markets has implications for movement
of the marginal utilities in the two countries and real exchange rate. Noting
that the internationally traded securities are denominated in the domestic
currency, the optimal portfolio choice of the foreign households can be characterized by the following Euler equation
Qt,t+1 = β
∗
Ct+1
Ct∗
−σ Pt∗ St
∗
Pt+1
St+1
.
(22)
Combining (22) with the optimal choice of the domestic households (7) the
following condition can be derived:9
C0∗
C0
−σ 1
Φ0
Ct∗
Ct
−σ
= µΦt ,
(23)
where µ =
is a constant that depends on the initial distribution of wealth across countries. Note that in the case of symmetric initial
conditions µ equals 1. Hence the existence of complete security markets leads
to a simple relation linking the level of domestic consumption to the level of
foreign consumption and the real exchange rate. This is an alternative way
to state the uncovered interest parity condition, which can also be derived
combining the first order conditions of foreign and domestic consumer for the
optimal portfolio choice.
For the further analysis we log linearize the system around the determin9
See also Chari et al. (2002) and Galı́ and Monacelli (2005).
17
istic steady state values. The log linearized equations can be found in the
Appendix.
5
Empirical Analysis
To gain deeper insight we estimate the structural small new open economy
model outlined above for the Swedish data in the 1930s. We add to a recent
strand of the literature that deals with the estimation new open economy
models. Different empirical strategies have been applied, for example Ghironi (2000) applied non-linear least squares at the single-equation level to
estimate the structural parameters of such a model. Smets and Wouters
(2002) based their estimation on matching model implied impulse responses
to those of an an identified VAR, while Bergin (2003, 2006) and Dib (2003)
applied maximum likelihood estimation. Following the contributions of Lubik
and Schorfheide (2005, 2007), Ambler et al. (2004), Justiniano and Preston
(2004, 2006) and Adolfson et al. (2007a,b), we use Bayesian methods for the
estimation.10 This allows to base our estimation not only on the likelihood
function. It also allows to incorporate additional information in coherent way
via prior distributions, which can help to add curvature to the posterior in
regions where the likelihood is flat. To arrive at the posterior distribution of
the parameters, we use Gibbs sampling to generate draws from the posterior.
First we solve the linearized system with the technique of King and Watson
(2002). The following state-space representation mapping the unobservable
states into the observable data vector vt is derived:
!
yt
,
vt = L
xt
yt = Lyy yt−1 + Lyx xt−1 ,
(24)
xt = Axt−1 + ǫt .
10
The programs used to solve and estimate the model were written from first principles
in Matlab Version R2007a.
18
The vector xt is a collection of the structural shocks of the model consisting
of the technology shock zt , the terms of trade shock δt and the foreign output
shock ytf and yt contains the remaining state variables. Hence the matrix
A is equal to diag([ρz , ρδ , ρf ]) and the covariance matrix of ǫ is given by
Σ = diag([σz2 , σδ2 , σf2 ]). To account for potential measurement error and
model misspecification11 we add a vector autoregressive measurement error
et for the estimation as proposed by Ireland (2004). The empirical model is
thus given by
yt
vt = L
xt
!
+ et ,
(25a)
yt = Lyy yt−1 + Lyx xt−1 ,
(25b)
xt = Axt−1 + ǫt ,
(25c)
et = Det−1 + ξ t ,
(25d)
where D is a matrix of VAR parameters of the measurement error and ξ t is
a zero mean vector of disturbance with covariance matrix Υ, ξ t ∼ N (0, Υ).
5.1
Data and Prior Choice
To estimate the empirical implications of the model, we use monthly data
on industrial production and inflation. Seasonally adjusted industrial production is from the League of Nations, and covers the period January 1928
- December 1939. Following Rabanal and Rubio-Ramirez (2005) we calculated log deviation from a quadratic trend. Inflation is based on a consumer
price index (League of Nations), and covers January 1928 - December 1939.
Inflation is calculated as seasonal first differences of the price index in logs
and is demeaned. The vector of observable variables for which we derive the
state-space representation consists of vt = [yt , πt ]′ .
As already emphasized by Sims (1980) strong a priori restrictions are
necessary to identify rational expectation models. To overcome identification
11
Del Negro et al. (2004) show that misspecification is also an issue for larger scale
models. See also Schorfheide (2000) for a loss function based comparison of potentially
misspecified models.
19
problems a number of parameters is usually calibrated (infinitely strict priors
are used).12
We set the discount factor to the conventional 0.99. Adolfson et al.
(2007b) have estimated a small open economy model for Sweden from 19802004. We follow their lead and calibrate the preference parameters σ and η
to one. Adolfson et al. (2007b) find a posterior mode for the substitution
elasticity between local and foreign goods of 22. Using macro level data the
estimated values are typically lower, but the micro evidence is in the range
of 5 to 20, see the evidence cited in Obstfeld and Rogoff (2000). We adopt
a more conservative value of 5, which conforms to their prior mean and the
value used by Adolfson et al. (2007a). The Calvo pricing parameter is set to
their posterior mode of 0.82. Finally, we set the long run share of imports in
consumption γ to the sample (yearly data) average of 0.25.
The remaining priors are chosen among others to ensure stability of the
transition equation. The persistence parameters of the structural shocks are
normally distributed with mean 0.6 and fairly large standard deviations. The
prior means for all autoregressive parameters of the measurement error are
all set to 0.4 with standard deviations of 0.2. All nuisance parameters are
centered over 0.1. The chosen degree of freedom parameters ensure that
the conditional posterior distributions of all variance/covariance parameters
have proper pdfs. All priors are assumed to be independent and the choice
of prior is summarized as
dij ∼ N (0.4, 0.22 ), i, j = 1, 2,
Υ ∼ IW (4 × 0.12 I2 , 7),
ρk ∼ N (0.6, 0.22 ), k ∈ (z, δ, f ),
(26)
σk2 ∼ IG(0.12 , 3), k ∈ (z, δ, f ).
5.1.1
Estimation Strategy
Given the choice of our calibrated parameters we can apply the Gibbs sampling algorithm. Defining a new state vector αt = (yt′ , x′t , e′t )′ the model in
12
For more on identification problems associated with the estimation of New Keynesian
models, see e.g. Beyer and Farmer (2004) or Lubik and Schorfheide (2005).
20
(25) has the following state-space representation:
 
yt
 
vt = L I2 xt  ;
et

 


 
!
0 0
Lyy Lyx 0
yt−1
yt
ǫ
 t
 


 
.
A 0  xt−1  + I3 0 
xt  =  0
ξt
0 I2
et−1
0
0 D
et
This allows to apply multi-move Gibbs Sampling as introduced by Carter and
Kohn (1994). In the first step the state vector αT is generated conditional
on A, Σ, D, Υ and the calibrated parameters.
Given that αT is observable eT is also observable and (25d) is a simple VAR model. Using the independent normal-inverse Wishart prior from
above, draws for d = vec(D′ ) and Υ can be generated as described e.g. in
Hamilton (1994, chapter 12)13 .
A similar argument holds for the structural shocks. Given that αT is
observable xT is known and (25c) consists of three independent univariate
autoregressive processes. Note that in the univariate specialization of the inverse Wishart distribution is called inverse Gamma distribution.14 Using an
independent inverse Gamma prior, analogue to the multivariate case above,
draws for the free elements of A and the variance parameters can be generated.
We use the steps above repeatedly to generate 50 000 draws and discard
the first 10 000 for burn-in. All performed convergence diagnostics were
satisfactory.
13
14
See also Canova (2007, chapter 4).
For the inverse Gammma the standard parameterization, (see e.g. Greenberg, 2008)
is
f (y|α, β) ∝ y −(α+1) exp (−β/y)
We use analog to the multivariate (inverse Wishart) case υ = 2α and T = 2β.
21
5.2
Empirical results
The parameter estimates are summarized in table 1. The table shows prior
and posterior means together with 90 percent probability intervals. Overall
the priors are considerably updated by the information in the likelihood.
The terms of trade shock turns out to be be most persistent shock. The
variance parameters are estimated precisely, the posterior mean being 70 to
80 percent smaller than the prior mean. The posterior distribution of the
VAR parameters is largely dominated by the sample information. In one case
the 90 percent interval of the posterior and prior do not even overlap.
Table 1: Parameter Estimation Results
Prior
Posterior
Parameter
Median
90% Interval Median
90% Interval
Technology
ρz
0.60 [0.344 , 0.856]
0.54 [0.336 , 0.792]
Terms of trade
ρδ
0.60 [0.344 , 0.856]
0.60 [0.446 , 0.918]
Foreign output
ρf
0.60 [0.344 , 0.856]
0.50 [0.362 , 0.714]
sd technology
σz
0.10 [0.041 , 0.131]
0.03 [0.024 , 0.046]
sd terms of trade σδ
0.10 [0.041 , 0.131]
0.02 [0.013 , 0.038]
sd foreign output σf
0.10 [0.041 , 0.131]
0.03 [0.024 , 0.082]
ME par
d11
0.40 [0.144 , 0.656]
0.89 [0.570 , 0.980]
ME par
d21
0.40 [0.144 , 0.656]
0.18 [-0.132 , 0.444]
ME par
d12
0.40 [0.144 , 0.656]
-0.04 [-0.170 , 0.061]
ME par
d22
0.40 [0.144 , 0.656]
0.48 [0.207 , 0.626]
sd ME output
υy
0.10 [0.061 , 0.135]
0.15 [0.144 , 0.182]
sd ME inflation
υπ
0.10 [0.061 , 0.135]
0.02 [0.021 , 0.033]
cov ME
υyπ
0.00 [-0.005 , 0.005]
0.00 [0.000 , 0.001]
Our main interest concerns the role of the exchange rate in the strong Swedish
recovery after 1932. Figure 6 depicts the estimated terms of trade for Sweden in the 1930s. As for the crucial period from 1932 to 1934, when Swedish
22
exports increased and introduced the recovery, we can observe a strong depreciation of the Swedish real exchange rate. From early 1934 until 1938 the
mean and nearly all probability mass of the estimated terms of trade lies in
region of undervaluation, indicating a lasting undervalued Swedish krona up
to 4 percent. These results strongly suggest that Sweden’s exporting sectors
were profiting to a large extent from the exchange-rate policy of the Riksbank. They are also highly compatible with the narrative evidence presented
in section 3. Repeatedly, the Riksbank expressed its concern over exchange
rate stability. Its goal was to keep the krona undervalued in order to boost
exports.
Figure 6: Estimated Terms of Trade
0.1
Terms of Trade − Deviations from Steady State
0.08
0.06
0.04
0.02
0
−0.02
−0.04
−0.06
−0.08
−0.1
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
Year
Notes: Posterior mean, 16-th and 84-th percentiles of estimated terms of trade in percentage deviations from the steady state.
23
Variance decomposition can be use to infer the role of the the different
shocks in driving output and inflation. The forecast error decompositions
shown in figure 7 confirms our claim. At all horizons the terms of trade shock
plays a dominant role in explaining output. It accounts for about half of the
overall forecast error variance. The other most important factors associated
with variation in output turn out to be the measurement error and the foreign
demand shock, each contributing about 10 percent to the explained variance.
Unfortunately, inflation is mostly captured by the measurement error, and to
a smaller degree by the terms of trade shock. This may be due to the strong
persistence of the series. The other shocks do not have a lot of explanatory
power.
Figure 7: Forecast Error Decomposition
Forecast Error Decomposition Output
100
Percent of FEV
80
60
40
20
0
0
2
4
6
8
10
12
14
16
18
14
16
18
Horizon
Forecast Error Decomposition Inflation
100
Percent of FEV
80
60
40
20
0
0
2
4
6
8
10
12
Horizon
Technology
Terms of trade
Foreign output
24
ME Inflation
ME Output
6
Conclusion
In their influential essay on the revived Bretton Woods System, Dooley et al.
(2003) claim that since 1945 the periphery countries have always chosen the
same development strategy. This strategy consists of four elements: undervalued currencies, controls on capital flows and trade, reserve accumulation,
and the use of the center region as a financial intermediary that lent credibility to their own financial systems. In the immediate postwar years, Western
Europe and Japan pursued such a strategy, since 1990 Eastern Europe and
Asia.
In this paper we ask whether we can use this insight for a better understanding of Sweden’s economic policies in the 1930’s. Building on Lundberg
(1983) who claimed that Sweden consciously kept its currency undervalued
vis-à-vis the British pound we apply a DSGE model for a small open economy
to study the effects of an undervalued currency.
Our results suggest that the insight of Dooley et al. (2003) is in fact useful
to understand the 1930’s. Our finding also suggests that the development
strategy as they described it was not an invention of the postwar world, but
came out of the discussion on the optimal exchange-rate policy in the wake
of the Great Depression.
The Dynamic System
In the following, log deviations from steady state values are denoted by small
letters H = log HHt .
25
1. Static equations
ηnt + σct = wt − pht − γδt
mct = wt − pht − zt ,
yt = zt + nt
yt = (1 − γ)ct + γa(2 − γ)δt + γytf
2. Dynamic equations
f
σEt (ct+1 − ct ) = σEt (yt+1
− ytf ) + (1 − γ)Et (δt+1 − δt )
πt = πth + γ(δt − δt−1 )
h
+ κmct ,
πth = βEt πt+1
zt = ρz zt−1 + ǫzt
f
ytf = ρs yt−1
+ ǫft
δt = ρδ δt−1 + ǫδt
References
Adolfson, Malin, Stefan Laséen, Jesper Lindé, and Mattias Villani,
“Bayesian estimation of an open economy DSGE model with incomplete
pass-through,” Journal of International Economics, July 2007, 72 (2), 481–
511.
, , , and , “Evaluating An Estimated New Keynesian Small Open
Economy Model,” Working Paper Series 203, Sveriges Riksbank (Central
Bank of Sweden) February 2007.
Ambler, Steve, Ali Dib, and Nooman Rebei, “Optimal Taylor Rules
in an Estimated Model of a Small Open Economy,” Working Papers 04-36,
Bank of Canada 2004.
26
Bergin, Paul R., “Putting the ’New Open Economy Macroeconomics’ to a
test,” Journal of International Economics, May 2003, 60 (1), 3–34.
, “How well can the New Open Economy Macroeconomics explain the
exchange rate and current account?,” Journal of International Money and
Finance, August 2006, 25 (5), 675–701.
Bernanke, B. and H. James, “The Gold Standard, Deflation, and Financial Crisis in the Great Depression: An International Comparison,” in R.G.
Hubbard, ed., Financial Markets and Financial Crises, Chicago, London:
University of Chicago Press, 1991, pp. 33–68.
, Th. Laubach, F. Mishkin, and A. Posen, Inflation Targeting:
Lessons from the International Experience, Princeton: Princeton University Press, 1999.
Beyer, Andreas and Roger E. A. Farmer, “On the indeterminacy of
New-Keynesian economics,” Working Paper Series 323, European Central
Bank 2004.
Calvo, Guillermo A., “Staggered prices in a utility-maximizing framework,” Journal of Monetary Economics, September 1983, 12 (3), 383–398.
Canova, Fabio, Methods for Applied Macroeconomic Research, Princeton
and Oxford: Princeton University Press, 2007.
Carter, C. K. and R. Kohn, “On Gibbs Sampling for State Space Models,” Biometrika, 1994, 81, 541–553.
Chari, V V, Patrick J Kehoe, and Ellen R McGrattan, “Can Sticky
Price Models Generate Volatile and Persistent Real Exchange Rates?,”
Review of Economic Studies, July 2002, 69 (3), 533–63.
Clarida, Richard, Jordi Gali, and Mark Gertler, “Monetary Policy
Rules and Macroeconomic Stability: Evidence and some Theory,” Quarterly Journal of Economics, 2000, 115, 147–180.
27
Del Negro, Marco, Frank Schorfheide, Frank Smets, and Raf
Wouters, “On the fit and forecasting performance of New Keynesian models,” Working Paper 2004-37, Federal Reserve Bank of Atlanta 2004.
Dib, Ali, “Monetary Policy in Estimated Models of Small Open and Closed
Economies,” Working Papers 03-27, Bank of Canada 2003.
Dixit, Avinash K and Joseph E Stiglitz, “Monopolistic Competition and
Optimum Product Diversity,” American Economic Review, June 1977, 67
(3), 297–308.
Dooley, M. P., D. Folkerts-Landau, and P. Garber, “An Essay on the
Revived Bretton Woods System,” 2003. NBER Working Paper 9971.
Eichengreen, B. and J. Sachs, “Exchange Rates and Economic Recovery
in the 1930s,” Journal of Economic History, 1985, 45, 925–946.
Eichengreen, Barry, Golden Fetters. The Gold Standard and the Great
Depression 1919-1939, Oxford: Oxford University Press, 1992.
Fisher, Irving, Stabilized Money, London: Allen & Unwin, 1935.
Fregert, K. and L. Jonung, “Deflation Dynamics in Sweden: Perceptions, Expectations, and Adjustment During the Deflations of 1921-1923
and 1931-1933,” in R. Burdekin and P. Siklos, eds., Deflation: Current
and Historical Perspectives, Cambridge, New York: Cambridge University
Press, 2004, pp. 91–128.
Gali, Jordi, Monetary Policy, Inflation, and the Business Cycle: An Introduction to the New Keynesian Framework, Princeton: Princeton University
Press, 2008.
Galı́, Jordi and Tommaso Monacelli, “Monetary Policy and Exchange
Rate Volatility in a Small Open Economy,” Review of Economic Studies,
07 2005, 72 (3), 707–734.
28
Ghironi, Fabio, “Towards New Open Economy Macroeconometrics,”
Boston College Working Papers in Economics 469, Boston College Department of Economics 2000.
Greenberg, Edward, Introduction to Bayesian Econometrics, Cambridge:
Cambridge University Press, 2008.
Hamilton, James D., Time Series Analysis, Princeton, New Jersey:
Princeton University Press, 1994.
Ireland, Peter N., “A method for taking models to the data,” Journal of
Economic Dynamics and Control, March 2004, 28 (6), 1205–1226.
Jonung, Lars, “Knut Wicksell’s Norm of Price Stabilization and Swedish
Monetary Policy in the 1930’s,” Journal of Monetary Economics, 1979, 5,
459–496.
, “Swedish Experience under the Classical Gold Standard, 1873-1914,”
in Michael D. Bordo and Anna J. Schwartz, eds., A Retrospective on the
Classical Gold Standard, 1821-1931, Chicago: University of Chicago Press,
1984, pp. 361–399.
Justiniano, Alejandro and Bruce Preston, “Small Open Economy
DSGE Models - Specification, Estimation, and Model Fit,” Manuskript,
International Monetary Fund and Department of Economics, Columbia
University 2004.
and , “Can Structural Small Open Economy Models Account for the Influence of Foreign Disturbances?,” Technical Report, Columbia University
2006.
Kindleberger, Charles P., The World in Depression. 1929-1939, London:
Allen Lane The Penguin Press, 1973.
King, Rober G. and Marc W. Watson, “System Reduction and Solution Algorithms for Singular Linear Difference Systems under Rational
Expectations,” Computational Economics, 2002, 20, 57–86.
29
Lane, Philip R., “The New Open Economy Macroeconomics: A Survey,”
Journal of International Economics, 2001, 54, 234–266.
Lubik, Thomas A. and Frank Schorfheide, “Do central banks respond
to exchange rate movements? A structural investigation,” Journal of Monetary Economics, May 2007, 54 (4), 1069–1087.
Lubik, Thomas and Frank Schorfheide, “A Bayesian Look at New Open
Economy Macroeconomics,” in Marc Gertler and K. Rogoff, eds., NBER
Macroeconomics Annual, number 521, MIT Press, 2005.
Lundberg, E., Ekonomiska kriser förr och nu, Stockholm: SNS, 1983.
Monacelli, Tommaso, “Monetary Policy in a Low Pass-Through Environment,” Journal of Money, Credit and Banking, 2005, 37, 1047–1066.
Obstfeld, Maurice and Kenneth Rogoff, “Exchange Rate Dynamics
Redux,” Journal of Political Economy, June 1995, 103 (3), 624–60.
and , Foundations of International Macroeconomics, Cambride, Massachusetts: The MIT Press, 1996.
and , “The Six Major Puzzles in International Macroeconomics: Is
There a Common Cause?,” NBER Macroeconomics Annual, 2000, 15
(0012003), 339–390.
Rabanal, Pau and Juan F. Rubio-Ramirez, “Comparing New Keynesian models of the business cycle: A Bayesian approach,” Journal of
Monetary Economics, 2005, 52 (6), 1151–1166.
Reinhart, C. and K. Rogoff, “This Time is Different. A Panoramic View
of Eight Centuries of Financial Crises,” 2008. NBER Working Paper 13882.
Schön, Lennart, A Modern Swedish Economic History: Growth and Transformation in Two Centuries, Stockholm: SNS, 2000.
Schorfheide, Frank, “Loss function-based evaluation of DSGE models,”
Journal of Applied Econometrics, 2000, 15 (6), 645–670.
30
Sims, Christopher A., “Macroeconomics and Reality,” Econometrica,
1980, 48, 1–48.
Smets, Frank and Raf Wouters, “Openness, imperfect exchange rate
pass-through and monetary policy,” Journal of Monetary Economics, July
2002, 49 (5), 947–981.
Straumann, Tobias and Ulrich Woitek, “A Pioneer in Monetary Policy?
Sweden’s Price Level Targeting of the 1930s Revisited,” European Review
of Economic History, 2009, 13, 251–282.
Woodford, Michael, Interest and Prices: Foundations of a Theory of Monetary Policy, Princeton, NJ: Princeton University Press, 2003.
Yun, Tack, “Monetary Policy, Nominal Rigidity, and Business Cycles,”
Journal of Monetary Economics, 1996, 37, 345–370.
31