Manufacturing Restructuring and the Role of Real Exchange
Rate Shocks
Karolina Ekholm, Sveriges Riksbank and CEPR
Andreas Moxnes, Dartmouth College
Karen Helene Ulltveit-Moe, University of Oslo and CEPR
We are grateful to Jonathan Eaton, Simon Evenett, Ann Harrison, Philippe Martin, Thierry Mayer, Victor
Norman, Andrew Bernard, Stephen Redding, David Weinstein and Mary Amiti and two anonymous referees
for helpful comments, and thank participants at the European Research Workshop in International Trade (ERWIT/CEPR) in Madrid in June 2009, the CEPR/MNB workshop "Productivity, Trade and Development" in
Budapest, the Empirical Investigations in International Trade (EIIT) – 15th Annual Conference in Colorado in
October 2008, the European Trade Study Group (ETSG) in Warsaw in September 2008, the Nordic International
Trade Conference (NOITS) in Bergen in May 2008, the Passau New Economic Geography Conference 2008 as well
as seminar participants at the Paris School of Economics and the Norwegian Central Bank for valuable discussion.
Ekholm gratefully acknowledges …nancial support from the Jan Wallander and Tom Hedelius Foundation and
the Swedish Council for Working Life and Social Research.
1
Abstract
Using a new and extensive micro data set we investigate the impact of a change in international competitive pressure on industrial performance and restructuring. Unlike previous
studies we are able to account for the heterogeneity across …rms in their exposure to foreign
competition. We focus on a situation akin to a natural experiment, and examine the impact
of a sharp real appreciation of the Norwegian Krone in the early 2000s on Norwegian manufacturing …rms which di¤er substantially in their trade orientation. A change in the real
exchange rate (RER) a¤ects a …rm through three di¤erent channels: (i ) …rm’s export sales,
(ii ) …rm’s purchases of imported inputs, and (iii ) import competition faced in the domestic
market. Unlike previous studies, we are able to examine all three channels. Both net exporters and import-competing …rms were exposed to increased competition due to the real
appreciation. Both groups reacted by shedding labor, but only the …rst group experienced
increasing labor productivity. Partly, the productivity improvements came from measured
TFP gains, while capital deepening does not appear to have been a¤ected by the shock.
JEL Classi…cation: F14, F31, F41
2
1
Introduction
Real appreciations are typically feared by export-industry representatives and governments for
their potential negative in‡uence on pro…tability, investment and employment. However, real
appreciations may potentially also strengthen competitive pressure and force industrial restructuring that in turn gives productivity gains. A long-standing issue in economics is whether
competitive pressure matters for industrial performance. Based on theory we would expect
there to be gains from increased competition. But the empirical evidence to support this view
is still not overwhelming.1
Our objective is to investigate the impact of a change in international competitive pressure
due to a distinct real appreciation on employment, production, investment, and productivity.
In doing so we seek to shed light on the role played by competitive pressure on industrial
performance and restructuring along the intensive as well as the extensive margin. We follow
a route similar to that of Galdón-Sánchez and Schmitz (2002), focusing on a situation akin to
a natural experiment, and examine the impact of a change in competitive pressure following a
sharp real appreciation of the Norwegian Krone in the early 2000s.
The extent to which a real exchange rate (RER) shock changes the competitive pressure faced
by a …rm is determined by its exposure to trade. Recent theoretical and empirical contributions
stress the importance of taking intra-industry …rm heterogeneity into account when studying
structural adjustment to changes in the trading environment and competitive pressure (see e.g.
Melitz, 2003 and Melitz and Ottaviano, 2008). In recent years, increasing evidence has emerged
that …rms’exposure to trade varies signi…cantly even within export industries (see e.g. Bernard
et al., 2007). This implies that …rms within the same industry may be hit very di¤erently by a
RER shock. A real appreciation will tend to increase the competitive pressure for …rms that sell
a large share of their output in foreign markets. However, …rms with a large share of exports
in total sales often import a large share of their intermediate inputs. Since a real appreciation
tends to make these inputs cheaper, the RER shock has an ambiguous e¤ect on pro…tability
and …rm performance.2
Most of the literature on the impact of RER changes is based on industry-level analysis,3
while some of the more recent contributions have used …rm-level data.4 We apply a new and
extensive micro data set for Norwegian manufacturing with detailed information on …rms’exports as well as imports of intermediates. This allows us, in contrast also to previous …rm-level
studies, to calculate precise measures of trade exposure. In doing so, we are able to account for
the heterogeneity across …rms with respect to their net currency exposure –taking into account
the share of exports in total output as well as the share of imported inputs in total costs –
1
A few recent studies have found evidence of productivity gains from increased competition due to developments in commodity markets, Recent important contributions have been made by Schmitz (2005) and GaldónSánchez and Schmitz (2002) through their work on detailed data for the iron ore industry. There is also a growing
literature on the impact of trade liberalization and thus increased foreign competition, see, e.g., Pavcnik (2002)
and Tre‡er (2004).
2
The correlation between export and import share for Norwegian …rms with positive exports and imports was
0:71 in 2004.
3
See e.g., Burgess and Knetter (1998), Branson and Love (1988), Campa and Goldberg (1995, 2001), Goldberg,
Tracy and Aaronson (1999), Goldberg (1993), and Klein et al. (2003).
4
See, e.g., Gourinchas (1999), Fung (2008) and Fung and Liu (2008).
3
and thus to overcome one severe shortcoming of previous analyses of RER shocks; the lack of
detailed, …rm-speci…c measures of trade exposure.5
Several conclusions emerge from the analysis. The real appreciation took place over a period
of two years. Over this period and the subsequent two years employment fell by 11 percent.
Almost half of this decline took place within ongoing …rms. The RER shock led to reduced
employment at the net exporters and the import competing …rms –i.e. among those who were
most exposed to the shock.
Over the same time labor productivity in Norwegian manufacturing rose dramatically. A
growth decomposition reveals that the productivity increase primarily can be ascribed to within…rm improvements, while reallocations between …rms and exits were less important. Our analysis shows that a substantial share of the increase in labor productivity can be ascribed to the
RER shock. We …nd that the shock led to productivity gains at the …rm level among the net
exporters,who improved their e¢ ciency in the face of tougher market conditions. But in contrast to previous studies, increased competition from imports in the domestic market did not
appear to have promoted productivity growth.6 .
The rest of the paper is organized as follows: We …rst provide a brief background to the
RER shock and industrial performance in the Norwegian manufacturing sector. We then proceed with a decomposition exercise to evaluate the relative contribution of the intensive and
extensive margin to overall changes is employment, production and productivity growth. Based
on the observation that adjustments within …rms appear do have been important drivers behind
the development at the aggregate level, we focus the empirical analysis on the contribution of
the RER shock to the industrial performance of surviving …rms. In section 3 we lay out our
identi…cation strategy, describe the estimation procedure, present the data and discuss econometric issues. In section 4 we present the empirical results, while in section 5 we discuss their
robustness. In section 6 we conclude.
2
The RER shock and industrial performance
The central bank of Norway adopted in‡ation targeting in March 2001. This was followed by
very high wage settlements. In order to comply with the in‡ation target, the response of the
central bank was to increase the interest rate, creating a large gap vis-à-vis foreign rates. This
gap was further enlarged as the Federal Reserve Bank and the European Central Bank lowered
their interest rates as the dot com bubble burst. Prior to 2000, the real exchange rate had
been rather stable, but between 2000 and 2002 the real exchange rate appreciated by around
17 percent7 (see Figure 1).
The real appreciation led to increased competitive pressure for exporters and for importcompeting …rms. We follow Galdón-Sánchez and Schmitz (2002) and de…ne an increase in
competitive pressure as an increase in the …rm’s probability of closure. The probability of
5
A somewhat related paper that uses a real exchange rate shock to identify …rm-level responses is the recent
contribution by Verhoogen (2008). However, his focus is on quality upgrading and wage inequality.
6
Increased import competition was found to increase productivity in, e.g., the study by Pavnick (2002).
7
Measured by relative hourly wages costs for workers in manufacturing in Norway relative to major trading
partners, denominated in a common currency. Other measures of the RER, e.g. from OECDs MEI (2008), show
very similar trends.
4
R elative hourly wag e c os ts for workers in manufac turing :
Norway vers us major trading partners ($, Index 1995=100)
125
120
115
110
105
100
95
90
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
Figure 1: The Norwegian RER shock 2000-2002
closure is determined by how the RER shock a¤ects pro…ts. According to this de…nition, an
adverse e¤ect of the RER shock on pro…ts translates into an increase in competitive pressure.
If a substantial share of Norwegian …rms experienced increased competitive pressure, we
would expect this to result in marked changes in aggregate employment, production and productivity from 2000 onwards. A …rst look at the aggregated data supports such a hypothesis.
From 2000 to 2004, manufacturing employment fell by 11 percent. In 20 out of 22 industries
employment growth declined during and just after the real appreciation relative to the period
before, 1996-2000.8 68 percent of the …rms experienced a decline in employment growth relative
to the 1996-2000 period. But over the same period gross production increased by 10:6%. Not
surprisingly, from 2000 to 2004, real labor productivity in Norwegian manufacturing thus rose
by 24 percent.9
We decompose the growth in employment, real gross production, and real labor productivity
in order to get a better grasp of the adjustment process at the intensive as well as the extensive
margin in response to the RER shock. The decomposition of employment and gross production
takes the form
Qt =
X
qit
i2C
where
Qt
Qt
Qt
4
X
i2X
qit
4
+
X
qit
i2E
denotes total change in the outcome variable,
qit
qit
qit
4
is
the corresponding …rm-level variable, and C; X and E are the sets of continuing, exiting and
entering …rms, and are calculated for the pre-shock period 1996-2000 and the shock period
2000-2004.10 Table 1 provides an overview.
8
See Tables 21 and 20 in the Appendix
Henceforth all productivity measures refer to real productivity. Details about the de‡ators used are provided
in the data section.
10
Continuing …rms are …rms that operated in both 2000 (1996) and 2004 (2000). Exiting …rms are …rms that
operated in 2000 (1996) but not in 2004 (2000). Entering …rms are …rms that operated in 2004 (2000) but not in
9
5
Table 1: Contribution of continuing, exiting and entering …rms to the growth of total employment and
output.
Contribution of
(2) Within (3) Exit (4) Entry
(1) Annual growth. %
Employment:
1996-2000
2000-2004
Real output:
1996-2000
2000-2004
Sum
1.24
-2.35
3%
46%
-519%
-301%
616%
247%
100%
100%
5.67
2.55
41%
125%
-99%
-277%
158%
252%
100%
100%
Notes: The contribution of e.g. continuing …rms to total employment growth is calculated as the change
in employment for continuing …rms divided by total change in employment. Continuing …rms are …rms
that operated in both 2000 (1996) and 2004 (2000). Exiting …rms are …rms that operated in 2000 (1996)
but not in 2004 (2000). Entering …rms are …rms that operated in 2004 (2000) but not in 2000 (1996).
Columns 2-4 show the contribution, in percent, of continuing, exiting and entering …rms,
P
respectively. E.g. column 2 shows i2C qit = Qt , the change in employment among continuing
…rms relative to the total change in employment. We see that almost half of the labor shedding
from 2000 to 2004 took place within ongoing …rms. But within the same …rms real output rose.
These …rms’ contribution to aggregate production growth tripled compared to the previous
four-year period. This happened despite the fact that capital deepening (measured by change
in capital intensity)11 slowed signi…cantly from 1996-2000 to 2000-2004. Over the …rst four
years capital intensity rose by 35%, while over the last four years only with 12.4%.
Our decomposition of real labor productivity growth follows Foster, Haltiwanger and Krizan
(2001), which is ultimately based on the BHC index, see Baily, Hulten and Campbell (1992).
The decomposition takes the form
=
t
X
si;t
it
k
+
i2C
X
X
sit (
it k
i2C
sit
k
(
t k)
it k
i2X
+
X
t k)
sector at time t (
t
it
= yit =lit ), and
X
sit
it
i2C
sit (
t k)
it
(1)
i2E
where sit denotes labor share for …rm i at time t (sit = lit =Lt ),
of …rm i at time t (
+
t
it
denotes labor productivity
denotes labor productivity in the manufacturing
= Yt =Lt ). C; X; E again denote the …rms that survive, exit and enter
respectively between t
4 and t. The …rst column of Table 2 shows the average aggregate
labor productivity growth, while column 2 represents the percentage contribution to growth
P
of the …rst term in the BHC decomposition, namely i2C si;t k it =
t : Columns 3 and 4
represent the between and cross components.12 Columns 5 and 6 represent the contribution of
exit and entry respectively. We see immediately that the within component dominates. Hence,
the lion’s share of the productivity growth can be attributed to adjustments on the intensive
2000 (1996).
11
See the appendix for details on calculation.
12
Compared to other studies the cross component here is very large. One important reason for this is the fact
that we are working with the entire population of manufacturing …rms, which includes a large number of small
…rms. A robustness test where we exclude all …rms with less than 20 employees supports this argument.
6
margins –within the …rms.13 Reshu- ing of resources due to between-…rm reallocations, entry
and exits play a much less signi…cant role. For the post-shock period we even observe a negative
contribution from exit to productivity growth. This contrasts with the results in related studies
such as Pavcnik (2002) and Disney, Haskel and Heden (2003).14
Table 2: Decomposing labour productivity growth
Contribution of
1996-2000
2000-2004
Annual
productivity growth, %
2.06
6.03
Within
Between
Cross
Exit
Entry
Sum
63%
73%
27%
35%
-81%
-27%
32%
-4%
59%
23%
100%
100%
Note: The contribution of e.g. within …rm growth is calculated as the weighted change in labor
productivity for continuing …rms divided by the aggregate change.
What role – if any – did the RER shock play for the fall in employment and surge in
production and productivity between 2000 and 2004? Potentially, the …ercer international
competition resulting from the real appreciation may have a¤ected employment and industrial
performance through two channels: (i ) by forcing restructuring in the surviving …rms and (ii )
by triggering the exit of less pro…table …rms. Given the signi…cant contribution to aggregate
developments from adjustments on the intensive margin, our focus will be on the impact of the
RER shock on the …rms that continued to operate throughout the period.15
3
3.1
Assessing the impact of an RER shock
Estimation Strategy
In order to investigate the in‡uence of the RER shock and thereby increased competitive pressure on industrial performance, we primarily rely on the heterogeneity of exposure to the shock
across …rms. This is the key feature that distinguishes our analysis from other studies attempting to quantify the e¤ect of real exchange rate movements. It is also the feature that
distinguishes the analysis from related studies on the impact of a change in competitive pressure due to trade liberalization (see, e.g., Pavcnik, 2002, and Tre‡er, 2004) or commodity prices
(see Galdón-Sánchez, 2002, and Schmitz, 2005). Previous analyses have used variation in sectorspeci…c RERs to achieve identi…cation (see, e.g., Klein et al., 2003; Gourinchas, 1999), and have
constructed sector-speci…c RERs by weighting bilateral RERs by sector-level trade shares.
13
Bernard et al (2003) also …nd a dominant intensive margin e¤ect of a RER shock on labor productivity in a
counterfactual simulation. In their model, the reason is that imports keep intermediates prices from rising by as
much as the wage, so that plants substitute intermediates for workers.
14
The exit and entry rates may respond di¤erently to a RER shock than to trade liberalization, as in Pavcnik
(2002), since RER shocks are more transitory in nature. However, Baggs et al. (2009) …nd that the e¤ect of tari¤
reductions and exchange rate changes on …rm survival are similar in magnitude. Moreover, there is evidence that
retail establishment turnover rates are higher than those in manufacturing and entry into them is relatively easy,
suggesting that entry and exit contribute more to their aggregate ‡uctuations, see e.g. Cambell and Lapham
(2004).
15
In the appendix, we provide evidence that …rm’s exposure to the RER shock was not correlated with …rm’s
exit probability.
7
We believe that our approach has clear advantages. First, there is mounting empirical
evidence on the heterogeneity of …rms with respect to trade activity. Previous studies investigating the impact of changes in foreign competition have not taken this into account as they
use industry- or sector-based measures of trade exposure. Access to …rm-level data allows us
to develop …rm-speci…c trade exposure measures. Campa and Goldberg (2001) point out that
there are three distinct channels of exchange rate exposure: (i ) the …rm’s export sales, (ii ) the
…rm’s purchases of imported inputs, and (iii ) import competition faced in the domestic market.
Unlike previous studies, we are able to consider all three channels.
Second, when using sector-speci…c RERs, one makes the stark assumption that goods are
priced in the same currency as the one used in the destination country for exports and in the
source country for imports. This may not always be the case. A recent survey shows that 30
percent of Norwegian manufacturing …rms use USD as the settlement currency in international
transactions (Federation of Norwegian Industries, 2008). However, the US share of exports and
imports in Norwegian trade in 2005 was only 8 and 5 percent, respectively.16
Finally, the Norwegian RER shock was quite uniform across destinations and the change in
the nominal exchange rate was very similar vis-à-vis all currencies. From an econometric point
of view, this means that the variation in the change in sector-speci…c RERs was quite small,
leading to weak identi…cation of any possible e¤ects of the shock.
We start by de…ning the net currency exposure of a …rm, which measures the extent to which
the real exchange rate shock leads to increased competitive pressure. Consider optimal revenue
of a …rm i: Ri = pi xi + V pi xi , where pi and pi are prices in local currency set at home and
abroad, respectively, xi and xi are sold quantities at home and abroad, respectively and V is
the nominal exchange rate expressed as units of domestic currency per unit of foreign currency.
We can rewrite revenue as Ri = (xi + xi =Pi ) pi , where Pi is the real exchange rate (RER),
Pi
pi = (V pi ). An increase in Pi implies a real appreciation. Considering a small change in Pi ,
holding output constant, we get the elasticity of revenue with respect to Pi ,
@Ri Pi
=
@Pi Ri
V pi xi
Ri
i
(2)
i.e. it is equal to the …rm’s export share. For given output and prices, a one percent real
appreciation decreases total revenue with
i
percent.
Symmetrically, we can de…ne …rm i’s costs as Ci = qi vi +V qi vi , where qi and qi are prices in
local currency of domestic and imported inputs, respectively, and vi and vi are used quantities
of domestic and imported inputs, respectively. We can rewrite costs as Ci = (vi + vi =Qi ) qi ,
where Qi = qi = (V qi ). Considering a small change in Qi , holding inputs constant, we get the
elasticity of costs with respect to Qi ,
@Ci Qi
=
@Qi Ci
V qi vi
Ci
ei
(3)
i.e. it is equal to the share of imported inputs in total costs.17 For given inputs and prices, a
16
Data from Statistics Norway, http://www.ssb.no/uhaar/tab-20.html.
More formally, we could de…ne the cost function C (qi ; qi ; Xi ) = minvi ;vi f(vi + vi =Qi ) qi g s.t. f (vi ; vi ) = Xi ,
where Xi = xi + xi . Applying Shepard’s lemma, we get that (@Ci =@Qi ) (Qi =Ci ) = ei .
17
8
one percent real appreciation decreases total costs by ei percent.
Suppose Pi = Qi , implying that the RER measured by output prices is equal to the RER
measured by input prices. Then the elasticity of pro…ts (
i)
– revenues minus costs – with
respect to a change in the real exchange rate can be expressed as:
@ i Pi
@Pi i
=
Pi
ei Ci
Pi
i
=
(
i
i
Ri
Pi
ei )
i
i =Ci
=
(4)
ei
i
ei
i =Ri
:
De…ne the net currency exposure as the di¤erence between the export share and the share of
ei . A positive net currency exposure ( i ) implies that the e¤ect on
imported inputs, i
i
pro…ts of a real appreciation (a rise in Pi ) is negative.18 The greater the net currency exposure,
the more negative the impact on pro…ts of an increase in Pi , the higher the probability of closure,
and, thus, the greater the increase in competitive pressure resulting from the RER shock. This
will be our main identifying assumption.
To check the robustness of our results, we exploit the fact that we are able to distinguish
between exposure related to the export share and exposure related to the share of imported
inputs in total costs. We calculate the separate exposures on the export and import sides, and
refer to them as gross currency exposure.
The currency exposure measures used to identify the e¤ect of a RER shock do not capture
the third channel through which a RER shock works, i.e. the import competition channel.
We deal with this by including measures of import competition as controls in the estimations.
Import competition in the domestic market cannot be measured at the …rm-level, however, but
only at the industry level. We choose this approach recognizing that a real appreciation means
tougher foreign competition for import-competing …rms, while assuming that pro…ts of …rms
with higher net currency exposure, all else being equal, will be more negatively a¤ected by a
real exchange appreciation than …rms with lower net currency exposure.
There are a few potential problems with our identi…cation method: (i ) Even if
i
= 0,
the …rm may be exposed because revenues and costs are denominated in di¤erent currencies.
However, for our analysis this is probably less of a problem, since the RER appreciated similarly
against most countries. (ii ) We do not observe the use of …nancial derivatives, i.e. to what extent
…rms hedge currency risk. However, we believe that this will not seriously bias our measure
because survey evidence suggests that long-term (> 3 years) currency hedging is relatively
uncommon.19 Secondly, …rms can only hedge against nominal currency shocks, not relative
output or input price movements.
18
Three features of (4) are worth noting: (i ) Net exposure is divided by pro…t relative to revenue or sales. The
pro…t e¤ect of high-pro…t …rms are, all else being equal, less sensitive to the net currency exposure to a real
appreciation.(ii ) Pro…ts are a¤ected by RER movements even for a …rm with zero net exposure. This is because,
as long as pro…ts are positive, revenue is higher than costs. So, a one per cent depreciation will have a larger
e¤ect on revenues than on costs.(iii ) The elasticity of pro…ts is zero when ei (Ci =Ri ) = i , so ei > i for a …rm
with positive pro…ts. Again, this is related to the point above, that the optimal import share is higher than the
export share because revenue is higher than costs.
19
See Norges Bank (the Norwegian Central Bank): "Penger og Kreditt" 1/2005.
9
3.2
Empirical model and identi…cation
To analyze the in‡uence of increased competitive pressure on various …rm outcomes, we use a
di¤erences-in-di¤erences approach. The approach is similar to Tre‡er (2004) who studied the
response of Canadian …rms to trade liberalization through the NAFTA agreement. We de…ne
the years 1996-2000 as the pre-RER-shock period and the years 2000-2004 as the RER shock
period. Let
yit be the average annual log change in the outcome variable of …rm i in period
t. The average annual log changes in the two periods are:
yi1 = (ln Yi2004
ln Yi2000 ) = (2004
2000) ;
yi0 = (ln Yi2000
ln Yi1996 ) = (2000
1996) ;
(5)
where t = 0 denotes the pre-RER shock period, and t = 1 denotes the RER shock period. The
RER was relatively stable in the years 1996 to 2000, as can be seen in Figure 1. From year
2000, however, the RER increased sharply.20 The choice of year 2000 as the start of the RER
shock period allows us to compare the outcome variable in the end year with its baseline level,
i.e. before the shock started. Although the shock reached its peak in 2002, the years 2003 and
2004 are also included as there may be some time between RER changes and …rms’response.21
We propose the following linear model for explaining the impact of the RER shock on the
change in …rm outcomes:22
yit =
i
+
t
+
(
i0
pt ) +
SE
yit
+ "it ;
The e¤ect of the RER change is assumed to be determined by the interaction term (
where
i0
is net currency exposure in the base year
where higher pt indicates an
at the level of the …rm,
(macro shocks)
t,
i,
appreciation.24
(1996)23
(6)
i0
pt ),
and pt is the log of the RER,
The speci…cation includes a growth-…xed e¤ect
a time-speci…c e¤ect capturing economy-wide business conditions
and idiosyncratic industry demand and supply shocks, proxied by changes
in Swedish manufacturing employment
SE .
yit
The model allows for time-invariant heterogeneity in growth rates across …rms (the
i ).
How-
ever, there may be variation in growth rates which coincides with our measure of net exposure.
For example, it may be that some …rms experience worsening worldwide business conditions,
and that these conditions are correlated with exposure. To control for idiosyncratic industry
20
We also investigate whether nominal exchange rate volatility increased during the shock years. We compute
the volatility of the exchange rate by following Tenreyro (2007). For each year in the sample, we compute the
standard deviation of the monthly variation of the nominal exchange rate, volt = stddev [(et;m et;m 1 ) =et;m 1 ],
where months m = 1:::12 and et;m is the e¤ective Norwegian krone exchange rate (I-44). By this measure, average
volatility in the two periods was quite similar, although volatility increased slightly in 2003 and 2004.
21
It is reasonable to assume that adjusting to a RER shock takes time, so that …rms’response will be characterized by a time lag. Labor market and …ring regulations impede immediate adjustment of the labor stock, while
exports and intermediates imports are typically bound by contracts that cannot be immediately re-negotiated.
22
In the appendix, we also formulate a model where the e¤ect of competitive pressure on the outcome variable
is allowed to be non-linear.
23
We choose the base year 1996 for both net exposure and other control variables. This ensures that the
covariates are pre-determined, which should minimize concerns about reverse causality. The correlation in net
exposure from t to t + 1 is 0:89. In the appendix, we also report results using year 2000 as the base year for all
covariates. The results remain largely unchanged.
24
The RER can be treated as an exogenous variable from the …rm’s perspective.
10
shocks – applying worldwide – we use Swedish industry-level manufacturing employment data
SE . The variable y SE will control for underlying worldwide changes in
(in logs) represented by yit
it
supply and demand, changes in pricing-to-market behavior, changes in the degree of competition from low-cost countries such as China, and other time-varying industry characteristics. We
choose to use Sweden as our control because (i ) Sweden’s RER was relatively stable during the
period under study and (ii ) Sweden is Norway’s largest trading partner with an economy that
is highly integrated with the Norwegian one, not only regarding goods and capital markets, but
regarding the labor market as well.
Di¤erencing (6) across periods yields our baseline di¤erence-in-di¤erence …rm-level speci…cation:
yi1
where
yi0 = +
0,
1
vi
will pick up the change in
i0 (
p1
SE
p0 ) + ( yi1
"i and the …rm …xed e¤ect
i
SE
yi0
) + xi0 + vi
(7)
is di¤erenced out: The estimated
yit which is due to the business cycle. The variable ( p1
p0 )
is de…ned as the di¤erence in the economy-wide change in the real exchange rate between the
pre-shock and the shock period and will just be a positive constant across all …rms. However,
variation in
i
will enable us to make inferences about
. If
> 0, the appreciation had a
positive impact on e.g. productivity growth, with exposed …rms experiencing a larger increase,
or smaller decrease, in productivity growth than similar non-exposed …rms.25
Note that if there were just two groups of …rms, exposed and non-exposed, then equation (7)
would amount to a triple-di¤erences strategy:
would re‡ect the di¤erence in change in average
growth rates between the exposed and non-exposed. Also note that the estimating equation
does not su¤er from serial correlation in the errors, since the averaging over periods ignores
time-series information.26 If we were to use many sequential years of pre- and post-shock data
without correcting for serial correlation, the standard errors on the coe¢ cient estimates would
be too small.
Following Tre‡er (2004), we also add a vector of controls, xi0 ; from the base year, 1996.
The …rm level controls include number of employees, labor productivity (both in logs) and
dummy variables indicating whether the …rm is exporting and importing. The industry level
controls include a measure of skill intensity (see the appendix for details), the Her…ndahl index
and its interaction with the exposure variable, and a measure of import penetration. In the
results section, we also present estimates using industry …xed e¤ects. Naturally, in this case the
industry controls will be subsumed by the …xed e¤ects.
The Her…ndahl index: Campa and Goldberg (1995) observe much weaker e¤ects of exchange
rates in industries with high price-over-cost markups. To account for the role of markups for
RER-e¤ects we include the Her…ndahl index, calculated in the base year at the 2-digit level in
the regression model (see Table 21 in the Appendix for Nace2-digit). In addition we include a
variable where the Her…ndahl index is interacted with net currency exposure.27 A higher value
25
If exposed …rms shift sales toward the home market, and if this increases the competitive pressure among
non-exposed …rms, then our estimates become biased, as the competitive pressure also changes for the control
group. As we show in the appendix, we do not …nd evidence for this mechanism in our data.
26
Bertrand, Du‡o and Mullainathan (2004) …nd that the strategy of ignoring time-series information performs
reasonably well in a Monte Carlo study.
27
In alternative speci…cations using gross exposure instead, the interaction term is excluded.
11
of the Her…ndahl index indicates stronger market power in the industry and should therefore be
positively correlated with mark-ups. A coe¢ cient of the interaction variable with an opposite
sign to beta would indicate that …rms operating in industries with relatively high markups are
less a¤ected by the RER shock.
Import competition: As discussed above, our identi…cation strategy implies that RER e¤ects
working through import competition are not identi…ed explicitly. Hence, to address the e¤ect
of the RER shock on import competing …rms, as well as to avoid biased estimates, we include a
variable measuring import penetration. Import competition is de…ned as the value of industry
imports relative to the value of industry absorption in our base year. In cross-industry studies,
the interaction between import competition and the RER shock would be the proper variable
to include. However, due to the long-di¤erencing of the data, the RER is constant across all
…rms, and as a consequence becomes part of the constant term. We report details about the
construction of the import competition variable as well as other variables in the appendix.
Imported capital goods: As pointed out in Eaton and Kortum (2001), it is possible that a
real appreciation of the home currency can make imported capital goods cheaper, causing a
di¤usion of advanced technology and subsequent impact on industrial performance. This e¤ect
will, however, a¤ect all …rms identically. That is, since the price decline on imports is the
same across all …rms, the change in incentives to invest in new imported machinery does not
depend on trade exposure. As a consequence, this e¤ect is subsumed in the constant term in
equation (7). Therefore, our framework can only identify the e¤ect of a change in competitive
pressure (through the RER shock, a¤ecting …rms di¤erently) on …rm outcome variables, and
not complimentary e¤ects of the RER shock, such as cheaper capital goods (a¤ecting …rms
identically).28
3.3
Data
We analyse …ve key …rm outcomes: employment, production, capital intensity and productivity.
The construction of the …rst three is detailed in the appendix. To measure productivity, we
look at real labor productivity as well as a revenue based TFP index. The calculation of the
TFP index is described below, where we also discuss the relationship between measured TFP
and true e¢ ciency.
We employ an exhaustive …rm-level data set for the Norwegian manufacturing sector. The
data set is based on several data sources. To begin with, we use …rm data from Statistics
Norway’s capital database, which is an unbalanced panel of all non-oil joint-stock companies
spanning the years 1996 to 2004, with approximately 8; 000 …rms per year.29 The panel provides
information about total revenue, value added, employment, man hours, capital, total operating
costs and intermediate costs. In 2004 the data set covered about 90 percent of manufacturing
output in Norway. All joint-stock companies are sampled with certainty in the panel. The
econometrics as well as descriptive statistics are based on a balanced panel, i.e. those …rms
28
In the appendix, we explore whether cheaper capital goods increased machinery imports during the shock
period.
29
Statistics Norway’s capital database is described in Raknerud et al. (2004).
12
operating in 1996, 2000 and 2004, comprising about 3; 700 …rms.3031
Information about exports and imports at the …rm level is assembled from customs declarations.32 These data make up an unbalanced panel of all yearly exports and imports values by
…rm. 39 percent of the unbalanced set of …rms were exporting in 2004, while 64 percent were
importing. Total manufacturing exports and imports amounted to 117 and 72 billion NOK in
2004, corresponding to $17:4 billion and $10:7 billion using the average exchange rate. The
trade data have then been merged with the capital database, based on a unique …rm identi…er. The weighted average export and import exposures were 25:6 and 16:5 percent in 1996.33
Weighted average net exposure was 9:2 percent in 1996, indicating that on average …rms would
face increased competitive pressure as a consequence of a real appreciation. Further details on
the data sets and the construction of variables are provided in the appendix.
3.3.1
Price indices
In most studies measuring productivity, output is de‡ated using industry-level price indices.
A potential problem is that heterogeneous prices within sectors may bias the productivity
measures and therefore lead to measurement error. For example, one might hypothesize that
prices charged by highly-exposed …rms decline in response to the shock, and this will incorrectly
show up as a fall in output and productivity in the data, if our de‡ators are incorrectly measured.
In this paper we correct for potential measurement error by constructing de‡ators that
re‡ect …rms’exposure to foreign markets. We use three di¤erent price indices. The …rst two are
producer price indices, one for exported goods (pejt ) and one for goods sold domestically (pdjt ),
both for all 3 digit NACE industries j. The third is an index for import prices pijt , by 2 digit
NACE. Given information about the …rm’s total sales, exports,h total intermediate icosts and
imports, value added in constant prices has been constructed as sit sit
vit
vit Pjt1 100,
where sit and vit represent domestic sales and costs of domestically-sourced intermediates in
current prices, and sit and vit represent foreign sales and costs of imported intermediates.34
i0
h
Pjt is the corresponding (4x1) vector of price indices, P = pdjt pejt pdjt pijt .35 Hence, value
added in constant prices controls for the impact of the RER shock on prices for …rms with a
high export share.
30
We also delete observations if value added is missing or negative, or if exports/(total revenue) or imports/(total operating costs) are larger than 1. Among the unbalanced panel of 8; 000 …rms per year, this
amounted to eliminating 285 …rms per year.
31
Continuing …rms are larger, less capital intensive, and more likely to export and import, compared to other
…rms. Average employment/capital intensity/export likelihood/import likelihood for continuing …rms relative to
non-continuing …rms is 1.48, 0.82, 1.36 and 1.03 in 2004.
32
The trade data are originally at the …rm-year-product (HS8)-destination level. In most of the analysis here,
except for a section in the appendix, we aggregate up to the …rm-year level.
33
These measures refer to the weighted average over the balanced set of …rms (including …rms with zero
exposure), where revenue is used as weights.
34
sit and vit are calculated as total sales-exports and total intermediate costs-imports respectively. In a few
cases sit and vit become negative due to data inconsistencies. These observations are dropped from the dataset.
35
The price index for domestic intermediates equals the index for domestic output since price indices for
intermediates are not available. Also, the price index for intermediate imports may not fully re‡ect the import
content of sector j. However, we believe that this approximation is satisfactory since input-output tables show
that most industries supply a major share of output to themselves. A more basic de‡ator using only price indices
d
for the output side, constructed as def latorit = pejt
(1
it + pjt
it ) yields similar results.
13
3.4
Productivity measurement
Revenue based total factor productivity (TFP) is estimated sector by sector using the following
gross production function: yit =
0
+
k kit
+
m mit
+
l lit
+ ! it +
it ;
where yit is real gross
output, kit is real capital services, mit is real material expenditure, lit is man hours (all in logs),
! it is unobserved productivity and
is either measurement error or a shock to productivity
it
which is not forecastable during the period in which labor can be adjusted (we omit industry
subscripts for notational convenience).
Measurement error in yit due to the use of industry-level de‡ators is a well known problem.
This can be especially harmful in studies comparing the productivity growth of …rms with
di¤erent degrees of exposure and therefore facing di¤erent prices. We describe how we minimize
this problem in the Section 3.3.1.
OLS estimates of the production function are in most cases biased. First, if ! it is observed
by the …rm before it chooses the optimal amount of labor and capital, ! it will be correlated
with lit and kit . Second, there may be selection bias because unobserved productivity may be
correlated with the …rm’s exit decision.36 We control for these two e¤ects using Olley-Pakes
(1996) techniques. We brie‡y describe the procedure, which consists of three stages. First,
consider the investment (in logs) function iit = f (! it ; kit ): Given that f is strictly increasing in
! it ; ! it = ft 1 (iit ; kit ): The production function can then be written
yit =
where
t (kit ; iit )
=
0
+
k kit
l lit
+
m mit
+
t (kit ; iit )
+
(8)
it
+ ft 1 (iit ; kit ): We approximate
it
by a 4th order polynomial
expansion with a full set of interactions. We allow the polynomial to vary over time by including
year dummies.37 OLS for every industry on equation (8) yields an unbiased estimate of
the polynomial,
t (kit ; iit ).
k
l
and
is unidenti…ed as capital appears both linearly and in ft 1 :
Second, we …nd survival probabilities Pit by estimating a probit model of exit. Again, the
regressors are a 4th order polynomial expansion along with year dummies. The third step of
the estimation procedure uses the estimates of
l,
t (kit ; iit )
and Pit and substitute them into
the following equation
yit+1
l lit+1
m mit+1
where g() approximates E ! it+1 j! it ;
=
order polynomial expansion in Pit and
that ! it+1 = g() +
it+1 ,
so
it+1
k kt+1
it+1
t
+ g (Pit ;
k kt )
t
+
it+1
+
it+1
= 1 , which we choose to approximate by a 3rd
k kt
= ! it +
0,
including year dummies. Note
represents the unanticipated part of …rm productivity, which
is mean independent of kt+1 (but not necessarily lt+1 ). Since capital enters both linearly as
separate input and inside g(),
k
is estimated by non-linear least squares.
Following the previous literature, e.g. Pavcnik (2002) and Klette (1996), in every industry,
we calculate the TFP index by subtracting predicted output from actual output, and then
comparing it relative to a reference plant. This ensures that the productivity index has the
36
For example, a …rm’s productivity and capital stock may jointly increase the probability of survival. Then,
! it and kit are negatively correlated in the selected sample. This creates a downward bias in the estimate of k :
37
We choose the following periods: 1996-1998, 1999-2000, 2001-2002 and 2003-2004.
14
desired properties such as transitivity and insensitivity to the units of measurement. Speci…cally,
we subtract productivity of an average reference plant in the base year from an individual …rm’s
productivity,
tf pit = yit
b lit
l
b mit
m
b kit
k
(yr
ybr )
where yr = y i0 and ybr = b l li0 + b m mi0 + b k k i0 , and the bar over a variable indicates a mean
over all …rms in the base year. Hence, tf pit represents a logarithmic deviation of a …rm from
the mean within an industry in the base year.
3.4.1
Revenue based productivity and e¢ ciency
Revenue based TFP confounds e¢ ciency (! it ) with …rm level prices. Although we take extra care
to construct …rm-speci…c input and output de‡ators, based on a …rm’s imports and exports (see
Section 3.3.1), some measurement error in real output will remain as long as we do not observe
physical output of the …rms.38 Denote the log production function qit = f (lit ; mit ; kit ) + ! it ,
where qit is physical quantities. Revenue based TFP is then
tf pit = (pit
= (pit
pt ) + qit
f (lit ; mit ; kit )
pt ) + ! it ;
(9)
where pit is the …rm price and pt the industry-level de‡ator.
Consider the case where ! it increases. First consider a model with monopolistic competition
and CES preferences, e.g. Melitz (2003). In this case, prices are a constant mark-up over
marginal costs, so any increase in e¢ ciency yields a proportional decline in prices, so that
revenue based TFP remains constant. Second, consider a model with monopolistic competition
and a linear demand system, as in Melitz and Ottaviano (2008). In such a case, prices are
a variable mark-up over marginal costs. In particular, an increase in e¢ ciency yields lower
prices but also higher mark-ups, so that revenue based productivity is positively correlated
with e¢ ciency. Third, consider the Ricardian model with Bertrand competition by Bernard
et al (2003). Here, …rms who become more e¢ cient increases their cost advantage over their
closest competition, set higher mark-ups, and therefore increases their revenue based TFP.39
Finally, consider a model where …rms face constant world market prices. In this case, revenue
based TFP is perfectly correlated with e¢ ciency, since prices are by assumption exogenous at
the level of the …rm.
Consider instead the case where the RER appreciates and ! it is constant. First, in a
Melitz (2003) model, revenue based TFP remains constant. Second, in Melitz and Ottaviano
(2008) revenue based TFP decreases - due to lower mark-ups in export markets - but less
than proportionally than the appreciation. Third, in Bernard et al (2003), revenue based TFP
will also decrease, since limit pricing may force the …rm to charge lower markups in export
markets. Finally, if world market prices are …xed, revenue based TFP falls proportionally with
38
In the following, we assume that inputs are homogeneous and measured without error. Katayama et al.
(2009) provide a comprehensive overview of these issues.
39
If markups are not limited by the costs of potential competitors, pricing is identical to Melitz (2003), so an
increase in ! it has no impact on tf pit .
15
the appreciation. In sum, with ! it constant, none of these models can explain our …nding that
the RER shock resulted in measured TFP gains.
The fact that we do …nd evidence for measured productivity gains therefore suggests that
! it increased, even though ! it and tf pit are not perfectly correlated. More generally (in a wider
range of models), equation (9) shows that, in principle, higher prices can explain measured
productivity gains. However, price hikes among a group of …rms experiencing a large increase
in competitive pressure does not seem very likely. Also note that the correlation between tf pit
and ! it is likely to be higher when using within-…rm time series variation in tf pit (as in this
study), compared to when using cross-…rm variation in tf pit , since variation in prices is most
likely lower within …rms (across time) than across …rms.
4
Empirical results
Can …rms’industrial performance be linked to …rms’currency exposure and thus be regarded
as a response to the increase in competition? Looking at employment, a …rst quick look at the
data suggests "no". Whether …rms had a high or low exposure does not seem to matter. But
once we start controlling for other …rm characteristics, the picture changes. Table 3 provides
an overview of …rms’employment growth, in terms of hours, before and during the RER shock
depending on their size, skill intensity and net currency exposure. As is clear from the table,
employment fell for all subgroups during the shock (from 2000 to 2004). However, comparing
the growth rates in period 0 and 1, we see that among the small …rms, the employment decline
was stronger among the exposed …rms compared to the non-exposed (both in terms of
y i1
y i0 ). The picture is less clear for large …rms, but we observe that the fall in
y i1 and
y i1 is
higher among exposed than non-exposed …rms.
Table 3: Firm characteristics and employment growth
Firm
size
Small
Small
Small
Small
Large
Large
Large
Large
Note:
Exposed
No
Yes
No
Yes
No
Yes
No
Yes
Skillintensity
Low
Low
High
High
Low
Low
High
High
yi0 : growth 1996-2000,
Firm size: Small:
Exposed: No:
No. of
…rms
1729
330
698
113
529
285
187
55
y i1
y i0
-3.4
-4.4
-2.9
-6.7
-1.9
-2.2
-2.1
-3.6
0.1
0.4
1.6
0.6
0.1
-1.8
2.5
0.7
y i1
y i0
-3.5
-4.7
-4.5
-7.4
-2.0
-0.4
-4.6
-4.3
yi1 : growth 2000-2004;
20 employees, Large: > 20 employees;
<0, Yes:
>0; Skill intensity: Low: <.18, High:
.18
Turning to productivity, we repeat the exercise above. Table 4 shows that …rms with a
positive net exposure on average had a higher increase in real labor productivity growth. The
increase in productivity growth was most pronounced among small …rms.
16
Table 4: Labor productivity, skill intensity, size and net exposure
Firm
size
Small
Small
Small
Small
Large
Large
Large
Large
Skillintensity
Low
Low
High
High
Low
Low
High
High
Exposed
No
Yes
No
Yes
No
Yes
No
Yes
yi0 : growth 1996-2000,
Note:
Firm size: Small:
Exposed: No:
No of
…rms
1729
327
698
112
529
285
187
55
yi1
yi0
yi1 - yi0
4.5
10.2
2.8
10.8
2.8
11.3
6.3
11.4
5.8
4.9
-1.3
-3.7
3.6
6.2
-2.7
0.3
-1.3
5.4
4.1
14.5
-0.8
5.1
9.0
11.1
yi1 : growth 2000-2004;
20 employees, Large: > 20 employees;
<0, Yes:
>0; Skill intensity: Low: <.18, High:
.18
Figure 2 provides an illustration of the di¤erence between exposed and non-exposed …rms’
productivity growth during the sample period. It is clear that while labor productivity growth
was relatively similar among the two groups in the pre-shock period, productivity grew much
faster for the exposed …rms in the shock period. However, highly exposed …rms may have
improved their labor productivity for other reasons than increased competitive pressure. In
other words, unobservable …rm characteristics may be correlated with exposure. We therefore
turn to econometric analysis.
Exposed
800
600
400
Labor productivity, 1000 NOK
1000
Non-exposed
1996
1998
2000
2002
20041996
1998
2000
2002
2004
Year
Graphs by eshare
Figure 2: Real labor productivity and net exposure.
Table 5 reports the results from estimating (7) on employment, production, real labor productivity, TFP, and capital intensity. We estimate the …ve equations simultaneously using
seemingly unrelated regressions (SURE), as we anticipate that the residuals might be correlated across outcome variables. Our baseline speci…cation includes …rm controls and industry
variables.
17
Employment: The results on net exposure suggest that the increased competitive pressure
in the export markets led to reduced employment measured in hours worked. We also observe
that increased import competition in the domestic market had a signi…cant negative impact on
employment. Hence, it appears that increased competitive pressure led to reduced employment
among export-oriented …rms as well as import-competing …rms.
Production: Higher competitive pressure in export markets contributed to increased production among the highly exposed. On the other hand, real output declined in sectors that
experienced higher import competition.
Labor productivity: The estimated coe¢ cient of net exposure is positive and signi…cant,
indicating that the real appreciation had a positive e¤ect on the most exposed …rms’ labor
productivity The positive impact of the RER shock was greater the larger the share of the …rm’s
sales that were exported, and the smaller the share of its intermediates that were imported. On
the other hand, labor productivity was not a¤ected in sectors that experienced higher import
competition.
Total factor productivity: The results are similar to those for labor productivity, although
the magnitudes are smaller. The RER appreciation resulted in increased revenue based TFP.
Capital intensity: Capital deepening is a potential source of productivity gains, and it may
accelerate during a RER shock. But the results show that there is little evidence of capital
deepening in response to the shock.40 A high degree of net exposure in the base year was
not associated with signi…cant changes in capital intensity, nor did import competition in the
domestic market appear to play a role.
All in all, we …nd that most exposed …rms on the one hand side shed more labor, while on
the other hand they also increase productivity (and production) the most. Moreover, whereas
increased competition in export markets enhanced productivity, increased import competition
had no e¤ect on productivity. Hence, while higher competitive pressure appears to have led to
decreased employment among both export-oriented and import competing …rms, only among
the …rst group of …rms was the employment decline associated with productivity improvements.
A Breusch-Pagan test of independence of the equations, that is, whether the disturbance correlation matrix is diagonal, strongly rejects the null hypothesis of independence. Table 6 reports
the correlation matrix, which points to a close positive relationship between labor productivity
and TFP, and a strong negative relationship between labor productivity and employment.
40
The results are similar when using de‡ated investment or investment intensity as the dependent variable.
18
Table 5: Increased competition and industrial performance. Baseline speci…cation.
Net exposure
Firm size
Labor productivity
Export dummy
Import dummy
Skill intensity
Import competition
Her…ndahl
Her…ndahl
Swe control
Industry dummies
No. of obs.
Employment
(in hours)
SURE
-1.284c
(.724)
-.014a
(.003)
-.135a
(.007)
.013
(.008)
.007
(.008)
.047a
(.007)
-.027a
(.006)
.002
(.093)
17.172c
(9.684)
.024
(.019)
No
3701
Production
SURE
.982c
(.581)
.010a
(.002)
.027a
(.006)
.007
(.007)
-.009
(.007)
.016a
(.006)
-.031a
(.005)
-.005
(.075)
-16.116b
(7.771)
.005
(.015)
No
3701
Labor
productivity
SURE
2.542a
(.673)
.021a
(.003)
.200a
(.007)
.001
(.008)
-.019a
(.007)
-.034a
(.007)
-.005
(.005)
-.022
(.087)
-9.185
(8.992)
.009
(.018)
No
3701
TFP
SURE
.486b
(.240)
.008a
(.001)
.048a
(.002)
.001
(.003)
-.006b
(.003)
-.004
(.002)
-.002
(.002)
.013
(.031)
.204
(3.209)
-.002
(.006)
No
3701
Capital
intensity
SURE
2.029
(1.382)
-.005
(.006)
.157a
(.014)
-.036b
(.016)
.022
(.015)
-.041a
(.014)
-.004
(.011)
-.191
(.178)
-24.126
(18.483)
.019
(.036)
No
3701
Notes: Independent variables in rows, dependent variables in columns.
All independent variables from the base year, 1996. The dep. variables take the form
yi1
yi0 :
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level
Table 6: Correlation matrix of residuals from SURE regression.
Employment
Labor prod.
TFP
Capital int.
Output
Employment
1.00
-.51
-.25
-.30
.52
Labor prod.
TFP
Capital int.
Output
1.00
.71
.36
.21
1.00
-.07
.25
1.00
-.04
1.00
Notes: Correlation matrix of residuals based on SUR estimates from Table 5.
Breusch-Pagan test of independence: chi2(10) = 5299.
Robustness: To check the robustness of the results, we estimate two other speci…cations: (i)
no …rm controls but including industry dummies and (ii) …rm controls and industry dummies.
We also estimate a modi…ed speci…cation with all covariates referring to the year 2000, i.e.
immediately before the shock. Results are reported in Tables 11 - 15 in the Appendix. The
robustness checks con…rm the impact of the RER shock on productivity, and especially so with
respect to labor productivity. As for production and employment, the in‡uence of the RER
19
shock appears less signi…cant. However, as we shall return to in the next section, this might be
related to the selection bias embedded in our econometric strategy (see Section 5.1).
We also check the robustness of the results by estimating the baseline model with gross
exposure instead of net exposure, i.e. we estimate coe¢ cients for export and import shares
separately, see Tables 16 - 18 in the Appendix. If our empirical model is correct, then the
export exposure coe¢ cient should be equal to the import exposure coe¢ cient, with opposite
signs (since i = i ei ). The results show that this is indeed the case for variables that
are signi…cant in the baseline regressions. The positive impact of export exposure on labor
productivity and production is con…rmed, and so is also the negative impact of import exposure
on labor productivity and TFP (Table 16).
Finally, in Table 19 we use a value added production function to calculate TFP, instead of
the gross output version used in the baseline speci…cation. Calculating TFP in this way tends
to increase the measured TFP gains from the RER shock, making it comparable to the e¤ect
on labor productivity.
5
Discussion of results
Before concluding, there is a set of aspects related to our estimation strategy that should be
discussed. First, we explore the validity of using non-exposed …rms as a control group. Second,
we investigate whether the impact of a change in competitive pressure on the outcome variables
is non-linear. Third, we assess the robustness of the assumption that …rm imports are in fact
used as intermediates in production, and go on to ask whether falling prices of imported capital
goods can explain the surge in labor productivity during the shock years. Finally, we aim at
assessing the economic signi…cance of our results.
5.1
Selection
Our econometric strategy precludes using data on …rms entering or exiting the sample, so …rms
which failed during the sample period are dropped. But balancing the panel is not a random
process. Firms staying in business may respond di¤erently to shocks than those who are driven
out of business, and this could potentially bias our results.
To deal with this selection problem, we follow the two-step Heckman (1979) procedure.41
First, we estimate a reduced form probit model of the probability of a …rm being in the continuous sample. Second, a variable is constructed using the inverse Mills ratio from the probit
and used as an additional regressor in the estimation of (7) to correct for selection bias.
The dependent variable in the …rst stage si is a dummy variable taking the value 1 if the
…rm is present from the beginning to the end of the sample. The dependent variable is set
to 0 if the …rm was present in the base year but exited in 2004 or earlier. The probability
of surviving will generally depend on the same …rm and industry characteristics that a¤ect
the RER response. However, if the set of explanatory variables are identical in both stages,
the model is identi…ed based on the functional form alone, which can lead to multicollinearity
problems. Hence, we choose a di¤erent set of variables in the …rst stage. Speci…cally, we posit
41
Wooldridge (2001) describes selection models for attrition problems in panel data.
20
that operating pro…ts in the base year enter the exit decision, but are excluded from the main
estimating equation. Although it is hard to …nd a completely satisfying exclusion restriction,
it seems reasonable that operating pro…ts in the base year have limited in‡uence on double
di¤erenced productivity, while being correlated with …rm exit. We also limit the covariates in
the entry equation to variables that are signi…cant in a stand-alone probit estimation. These
include the following variables: number of employees, labor productivity, export and import
status (dummy variables). All variables in the selection equation refer to values in the base
year.
Formally, we model the decision to stay in business as si = 1 [wi0 + ! i > 0], where 1[ ] is an
indicator function and wi is a vector of covariates. Assuming joint normality of errors vi and
! i , the errors in the second step vi now have expectation E [vi jsi = 1] = E [vi j! i >
12 E
[! i j! i >
wi0 ] =
12
(wi ), where
12
wi0 ] =
is the covariance between ! i and vi and () is the
inverse Mills ratio.
Estimating the baseline speci…cation of the empirical model using maximum likelihood with
…rm and industry controls (ML-HS), the main results from the SURE estimation are con…rmed,
see Table 7. The results are if anything stronger both with respect to statistical and economic signi…cance. Worth noting is especially the results on the impact of the RER shock on
employment.
Table 7: Increased competition and industrial performance: Controlling for selection bias.
Net exposure
Firm controls
Industry controls
No. of obs.
Employment
(in hours)
ML-HS
-.1.896a
(.781)
Yes
Yes
3747
Production
ML-HS
1.449b
(.637)
Yes
Yes
3747
Labor
productivity
ML-HS
2.740a
(.733)
Yes
Yes
3747
TFP
ML-HS
.625a
(.262)
Yes
Yes
3701
Capital
intensity
ML-HS
2.090
(1.397)
Yes
Yes
3701
Note: All independent variables are from the base year, 1996.
yi1
yi0 :
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level
The dependent variable takes the form
5.2
Domestic market shares
Our econometric methodology relies on the assumption that exporters and non-exporters are
not in direct competition in the home market. Suppose that, as a response to the RER shock,
that exporters shifted sales to the home market. This would e¤ectively increase the competitive
pressure facing non-exporters. This would create a bias in our estimates, since non-exposed
…rms act as a control group in our analysis. One way to check whether this is an issue, is to
examine export intensity among non-exposed (
i0
0) and exposed (
i0
> 0) …rms across time.
Figure 3 shows average export intensity (unweighted) from 1996 to 2004 for the two groups.42
42
Export intensity is calculated as de‡ated export value relative to real gross output. Note that average export
intensity among the negatively net exposed is positive. This simply re‡ects the fact that these …rms have higher
21
For both groups, export intensity is relatively stable. Importantly, there is no evidence that
the export share among exposed …rms decreased signi…cantly after the RER shock, compared
to the pre RER shock period. This suggests that shifts in competition in the home market do
0
Export intensity, average
.1
.2
.3
not bias our results.
1996
1998
2000
Year
2002
Non-expos ed
2004
Exposed
Figure 3: Average export intensity.
We have also augmented the empirical baseline model by adding log change in the export
share from 2000 to 2004 as an additional control variable. This allows us to compare changes
in the dependent variable for …rms with similar changes in their export share. The coe¢ cient
estimates did not change signi…cantly when including this control.43
5.3
Non-linear e¤ects of a change in competitive pressure
In this section, we investigate whether the impact of a change in competitive pressure on the
outcome variables is non-linear. For example, one might hypothesize that a small increase
in competitive pressure may not lead …rms to increase e¢ ciency, whereas a large increase in
competitive pressure may lead to changes in work practices, lay-o¤s, etc. We divide …rms into 6
mutually exclusive groups, based on their level of initial net exposure
Firms with
and
i0
i0
<
:17, ( 2) …rms with
< 0, ( 4) …rms with
and ( 6) …rms with
i0
i0
:17 and
i0
0 and
i0
i0
<
i0 .
The groups are ( 1)
:03, ( 3) …rms with
< :03, ( 5) …rms with
i0
:03 and
:03
i0
i0
< :17
:17. The thresholds roughly correspond to the 10, 25, 50, 75 and 90
percentile of initial exposure. We then estimate the following model:
yi1
yi0 = +
1 Di
2
+ :: +
5 Di
6
SE
+ ( yi1
SE
yi0
) + xi0 + vi
where Di k , k = 2; ::6, is an indicator variable taking the value 1 if …rm i belongs to group
Table 8 shows the results using …rm-level controls and industry dummies.
import intensity than export intensity.
43
Results available upon request.
22
k.
Table 8: Increased competition and industrial performance: Testing for non-linear e¤ects.
Dependent
variable
2
3
4
5
6
Firm controls
Industry dummies
No. of obs.
Employment
(in hours)
SURE
.004
(.013)
-.013
(.013)
-.034c
(.019)
-.026
(.019)
-.001
(.017)
Yes
Yes
3714
Production
SURE
.034a
(.010)
.022b
(.010)
.003
(.016)
.014
(.015)
.021
(.013)
Yes
Yes
3714
Labor
productivity
SURE
.026b
(.012)
.024c
(.013)
.021
(.018)
.048a
(.018)
.047a
(0.016)
Yes
Yes
3714
TFP
SURE
.005
(.004)
.001
(.005)
.010
(.006)
.008
(.006)
.010c
(.005)
Yes
Yes
3714
Capital
intensity
SURE
.006
(.025)
-.001
(.026)
.009
(.037)
.007
(.037)
.014
(.032)
Yes
Yes
3714
Note: All independent variables are from the base year, 1996.
yi1
yi0 :
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level
The dependent variable takes the form
The impact of competitive pressure on labor productivity is signi…cant for groups,
5; and
6, relative to the baseline group
strong for groups
5 and
2;
3;
1. However, the coe¢ cient is roughly twice as
6, suggesting that …rms above the 75 net exposure percentile are
particularly a¤ected. We see the same pattern for TFP, although the magnitudes are smaller.
For employment, we observe a signi…cant decline in the medium exposed group
0 and
i0
5.4
i0
4 (…rms with
< :03), but not for the other groups.
Measurement error in imported inputs
Net exposure is calculated based on the assumption that …rm imports are in fact intermediates
used in production. In some cases, however, …rms performing both manufacturing and retail
activities end up being classi…ed as manufacturing …rms in the data. As a consequence, the net
exposure variable might su¤er from measurement error since imports are only partially used as
inputs in production.
We perform the following robustness check to deal with this:
In the data, we identify …rm-level imports at the HS 8-digit product level. Using a
UN concordance, we map HS6 products to the BEC ("Broad Economic Categories")
classi…cation.44
The BEC classi…cation is an attempt to classify products based on their economic use.
Speci…cally, the BEC allows us to check whether an HS8 product is classi…ed as a consumer
good (BEC 6).
44
http://unstats.un.org/unsd/cr/registry/regcst.asp?Cl=10&Lg=1
23
We then aggregate the data by BEC and NACE, so that for each NACE 2-digit industry,
we know the share of …rm imports associated with consumer goods. On average, the share
is 14%, and the median is 3%.
We then perform the DID estimations on the industries with the highest (NACE 33, 25,
28, 21, 24, 17, 36, 22, 16, 19, and 18, sorted from lowest to highest share) and lowest share
of consumer imports.
The results are broadly similar for the two groups. Except for the employment coe¢ cient,
coe¢ cients signi…cant in the baseline speci…cation are also signi…cant here (detailed results available upon request). This suggests that retail activities are not a major source of measurement
error in the analysis.
5.5
Imports of capital goods and productivity growth
As pointed out in Eaton and Kortum (2001), an appreciation of the home currency can make
imported capital goods cheaper, causing a di¤usion of advanced technology and subsequent
productivity growth. Although this mechanism is not identi…ed by our econometric model (see
Section 3.2), it is possible that the surge in aggregate labor productivity between 2000 and 2004
can be, at least partially, explained by cheaper inputs. Here we explore some trends in imports
of machinery before and during the shock.
First, we identify …rm level imports that correspond as closely as possible to capital goods.
We choose imports of SITC Rev. 3 category 7 "Machinery and transport equipment" as our
proxy for capital goods. This group consists of sub-categories such as "Power-generating machinery and equipment", "Machinery specialized for particular industries", etc. Second, we sum
de‡ated machinery imports and total operating costs over the balanced set of …rms. Total real
machinery imports and the import intensity (relative to operating costs) are shown in Figure 4.
If anything, imports of machinery declined during the shock period, suggesting that the e¤ect
of cheaper imported inputs was not particularly strong.45
5.6
Economic signi…cance
Our empirical results point to a clear link between the real appreciation and the rise in productivity and production, and the fall in employment. Based on the results from the baseline
speci…cation, we …nd that employment fell by 1:3 percent for a one percent real appreciation
for a …rm with a net exposure equal to one. Hence, given that average net exposure of the
manufacturing sector of was 0:11 before the shock, we infer that the 14 percent real appreciation between 2000 and 2004 resulted in a 2:0 percent reduction in employment (
i0
p1 ).
From 2000 to 2004, manufacturing employment in Norway dropped by 11 percent. Based on
our calculations, around one …fth (2/11) can be attributed to a restructuring process among
the exporters that was triggered by the RER shock.
As for productivity, the manufacturing sector as a whole experienced a 24 percent increase
in labor productivity between 2000 and 2004. The growth decomposition above suggests that
45
In addition, the correlation between TFP growth between 2000 and 2004 and growth in imported machinery
between 2000 and 2004 was not signi…cantly di¤erent from zero.
24
8
Real machin er y impo r ts, bill NOK
5
6
7
1998
2000
Year
2002
2004
1998
2000
Year
2002
2004
Rea l machin er y impo r ts/o pe ra tin g co sts
.02 5
.03
.03 5
.04
.04 5
1996
1996
Figure 4: Real machinery imports.
around around 73 percent of this improvement, i.e. 17:5 percent were related to adjustments
on the intensive margin. Our analysis indicates that the …rms who faced the strongest increase
in direct competitive pressure – in terms of net exposure to the RER shock – gave the most
signi…cant contribution to this increase in productivity growth.
The estimated coe¢ cient on net exposure suggests that labor productivity increased by 2:5
percent for a one percent real appreciation for a …rm with a net exposure equal to one. Hence,
we infer that the real appreciation between 2000 and 2004 resulted in 3:9 percent higher labor
productivity. In other words, up to one quarter of the adjustment on the intensive margin can
be attributed to the enhanced competitive pressure (3.9/17.5).
6
Conclusions
We expect …rms to react to changes in competitive pressure. One potential source of changes
in competitive pressure are changes in the real exchange rate. Despite there being numerous
studies on the economic e¤ects of real exchange rate shocks, there is very little evidence on the
adjustment to such shocks at the …rm level. Since …rms within the same industry are found to
di¤er signi…cantly in size, productivity and trade exposure, …rm level analyses are required in
order to understand properly how the economy adjusts to real exchange rate shocks.
In this paper, we treat the sharp appreciation of the Norwegian Krone in the early 2000s as
a natural experiment in order to assess the impact of a real exchange rate shock on industrial
performance. To identify the impact of the shock we use …rms’ heterogeneous exposure to
international markets. Our hypothesis is that those …rms that are most exposed ex ante are
the ones most a¤ected by the shock in terms of increased competitive pressure.
Several conclusions emerge from the analysis. While both exporters and import-competing
…rms were exposed to increased competition due to the real appreciation and reacted by shedding
labor, only exporters increased their labor productivity in response to the shock. The RER
25
shock was associated with substantial within-…rm productivity gains for the net exporters.
Partly, this improvement in labor productivity came from measured TFP gains. Moreover,
the productivity gains among the most exposed …rms enabled them to increase output while
reducing employment. We do not …nd evidence of capital deepening due to the shock.
While our results suggest that changes in competitive pressure matters for industrial performance and productivity improvements, the speci…c mechanisms behind these gains remains
to be investigated. Testing hypotheses about core competence and changes in work practices
could be a fruitful avenue of future research.
26
References
[1] Baily, M. N., Hulten, C. and Campbell, D., 1992, “Productivity Dynamics in Manufacturing
Plants,” Brookings Papers on Economic Activity: Microeconomics, 187-249.
[2] Baggs, J., Beaulieu, E., and Fung, L., 2009, "Firm survival, performance, and the exchange
rate", Canadian Journal of Economics, Vol. 42, Issue 2, 392-421.
[3] Bernard, A. B., J. B. Jensen, S. J. Redding and P. K. Schott, 2007, "Firms in international
trade", Journal of Economic Perspectives 21, 105-130.
[4] Bernard, A. B., Redding, S. J., and Schott, P. K., 2006, "Multi-Product Firms and Trade
Liberalization", NBER Working Paper No. 12782.
[5] Bernard, A. B., Eaton, J., Jensen, J. B. and Kortum, S. S., 2003, "Plants and Productivity
in International Trade" American Economic Review 93, 1268-1290.
[6] Bertrand M., Du‡o, E., and Mullainathan, S., 2004, "How Much Should We Trust
Di¤erences-in-Di¤erences Estimates?" Quarterly Journal of Economics 119, 249-275.
[7] Burgess, S. and. Knetter, M. , 1998, "An International Comparison of Employment Adjustment to Exchange Rate Fluctuations", Review of International Economics 6, 151-163.
[8] Branson, W. and Love, J., 1988, "US manufacturing and the real exchange rate", in
Marston, R. C. (ed.), Misalignment of exchange rates: E¤ ects on trade and industry, National Bureau of Economic Research Project Report Series, University of Chicago Press,
Chicago and London.
[9] Campa, J. and Goldberg, L.,1995, "Investment in manufacturing, exchange rates and external exposure", Journal of International Economics 38, 297-320.
[10] Campa, J. and Goldberg, L., 2001, "Employment versus wage adjustmentand the US dollar", Review of Economics and Statistics 83, 477-489.
[11] Campbell, J. R. and Lapham, B., 2004, "Real Exchange Rate Fluctuations and the Dynamics of Retail Trade Industries on the U.S.-Canada Border", The American Economic
Review 94, 1194-1206.
[12] Colantone, I., 2006, "Trade Openness, Real Exchange Rates and Job Reallocation: Evidence from Belgium", mimeo, LICOS, Katholiek University, Leuven.
[13] Disney, R., J. Haskel and Y. Heden, 2003, "Restructuring and productivity growth in UK
manufacturing", Economic Journal 113, 666-694.
[14] Eaton, J. and Kortum, S., 2001, "Trade in capital goods", European Economic Review 45,
1195-1235.
[15] Federation of Norwegian Industries, 2008, Norsk Industri, Konjunkturraport 2008, downloadable at www.norskindustri.no/get…le.php/Dokumenter/PDF/Konjunkturraport%202008trykk.pdf.
27
[16] Foster, L., Haltiwanger J., and Krizan, C. J. , 2001, “Aggregate Productivity Growth:
Lessons from Microeconomic Evidence,” in Dean, E., Harper, M., and C. Hulten (Eds.),
New Developments in Productivity Analysis, University of Chicago Press, Chicago.
[17] Fung, L., 2008, "Large Real Exchange Rate Movements, Firm Dynamics, and Productivity
Growth", Canadian Journal of Economics 41, 1540-5982.
[18] Fung, L. and Liu, J.-T. , 2008, "The impact of real exchange rate movements on …rm performance: a case study of Taiwanese manufacturing …rms", Japan and the World Economy
21, 85-96.
[19] Galdón-Sánchez, J. E. and Schmitz, J. A, jr. 2002, "Competitive pressure and labor productivity: world iron-ore markets in the 1980s", American Economic Review 92, 1222-1235.
[20] Goldberg, L., Tracy, J., and Aaronson, S. R. , 1999, "Exchange Rates and Employment
Instability: Evidence from Matched CPS Data", American Economic Review 89, 204-210.
[21] Gourinchas, P.-O., 1998, "Exchange rates and jobs: What do we learn from gross ‡ows?",
in B. Bernanke and J. Rotemberg (eds), NBER Macroeconomics Annual, MIT Press, Cambridge, Mass.
[22] Gourinchas, P.-O., 1999, "Exchange rates do matter: French job reallocation and exchange
rate turbulence, 1984-1992", European Economic Review 43, 1279-1316.
[23] Haltiwanger, J., Kugler, A., Kugler, M. , Micco, A., and Pagès, C., 2004, "E¤ects of Tari¤s
and Real Exchange Rates on Job Reallocation", Evidence from Latin America”, Journal
of Policy Reform 7, n.o 4, December
[24] Heckman, J. J., 1979, "Sample Selection Bias as a Speci…cation Error", Econometrica 47,
153-61.
[25] Katayama, H., Lu, S. and Tybout, J. R., 2009, "Firm-level productivity studies: Illusions
and a solution", International Journal of Industrial Organization 27, 403-413.
[26] Klein, M. W., Schuh, S., and Triest, R. T., 2003, "Job creation, job destruction, and the
real exchange rate", Journal of International Economics 59, 239-265.
[27] Melitz, M. J., 2003, "The impact of trade on intra-industry reallocations and aggregate
industry productivity", Economectrica 71, 1695-1725.
[28] Melitz, M. J. and Ottaviano, G. I. P., 2008, "Market Size, Trade, and Productivity", Review
of Economic Studies 75, 295-316.
[29] Nucci, F. and Pozzolo, A. F., 2004, "The E¤ects of Exchange Rate Fluctuations on Employment: An Analysis with Firm-Level Panel Data", mimeo.
[30] OECD, 2004, OECD Economic surveys: Norway 2004, OECD, Paris.
[31] OECD, 2008, SourceOECD Main Economic Indicators: Comparative subject tables Vol
2008 release 04.
28
[32] Olley, G. S. and Pakes, A., 1996, "The Dynamics of Productivity in the Telecommunications
Equipment Industry," Econometrica 64, 1263-97.
[33] Pavnick, N., 2002, "Trade Liberalization, Exit, and Productivity Improvements: Evidence
from Chilean Plants," The Review of Economic Studies 69, 245-276.
[34] Raknerud, A., Rønningen D., and Skjerpen T., 2004, "Documentation of the capital database", Statistics Norway Documents 2004/16.
[35] Scarpetta, S., Hemmings P., Tressel T., and Woo, J., 2002, "The role of policy and institutions for productivity and …rm dynamics: evidence from micro and industry data", OECD
Economics Department Working paper 15.
[36] Schmitz, J. A. jr., 2005, "What determines productivity? Lessons from the dramatic
recovery of the US and Canadian iron ore industries following their early 1980s crisis",
Journal of Political Economy, 113, 582-625.
[37] Swenson, D., 2004, "Overseas Assembly and Country Sourcing Choices", Journal of International Economics, 66, 107-130.
[38] Tenreyro, S., 2007, "On the Trade Impact of Nominal Exchange Rate Volatility", Journal
of Development Economics 82, 485-508
[39] Tre‡er, D., 2004, "The Long and Short of the Canada-U.S. Free Trade Agreement", American Economic Review 94, 870-895.
[40] Verhoogen. E., 2008, "Trade, Quality Upgrading and Wage Inequality in the Mexican
Manufacturing Sector", Quarterly Journal of Economics 123, 489-530.
[41] Wooldridge, J. M., 2001, "Econometric Analysis of Cross Section and Panel Data", The
MIT Press, Cambridge, Mass.
29
A
A.1
Appendix
Data, variables and de…nitions
The exhaustive …rm-level data set for the Norwegian manufacturing sector is based on several data sources. It includes …rm data from Statistics Norway’s capital database, which is an
unbalanced panel of all non-oil joint-stock companies spanning the years 1996 to 2004, with
approximately 8; 000 …rms per year. The panel provides information about total revenue, value
added, employment, capital, total operating costs and intermediate costs. This panel is merged
with another exhaustive …rm-level data set for Norwegian manufacturing, containing information about exports and imports assembled from customs declarations. The trade data have then
been merged with the capital database, based on a unique …rm identi…er.46
Exports denote the sum of a …rm’s export value across destinations.
Imports denote the sum of a …rm’s import value of intermediates across sourcing countries.
Export exposure is de…ned as export value relative to total revenue.
Import exposure is de…ned as import value relative to total operating costs.
Net currency exposure is de…ned as the di¤erence between Export exposure and Import
exposure.
Imported inputs are measured as the import value de‡ated by 2-digit SITC-level indices
(year 2000=100).47 Total inputs are measured as total intermediate costs de‡ated by
a corresponding index, where pdjt and pijt are used to de‡ate the domestic and foreign
component of total inputs respectively, as described in the section on data.
Import competition in the 2-digit NACE sector j is de…ned as total import value in j
relative to total absorption in j in our base year. Absorption is calculated as (production
valuej ) - (export valuej ) + (import valuej ). Data on all these variables are collected from
Norwegian input-output matrices.48
Employees and …rm size refer to the number of persons employed in the …rm.
Man hours refers to the number of hours worked per …rm per year.
Production refers to total real revenue.
Labor productivity is measured as real value added divided by total number of hours
worked.
46
Among the …rms in the capital database in 2004, 66:7 percent are matched in the customs data (either as
exporters or importers). Among the …rms in the customs database in 2004, 8:6 percent are matched in the
capital database. The unmatched customs observations are related to activities outside the manufacturing sector
(agriculture, petroleum, services).
47
http://www.ssb.no/english/subjects/08/03/40/uhvp_en
48
http://ssb.no/nr_en/input-output.html
30
Output de‡ators refer to Statistics Norway’s commodity price index for the industrial
sector at the 3-digit NACE level.49 Year 2000=100. There are separate indices for the
domestic and export markets. Firm-level output de‡ators are calculated by de‡ating
domestic and exported outputs separately.
Input de‡ators refer to Statistics Norway’s domestic commodity price index for domestic goods and the import de‡ator for imported inputs. Firm-level input de‡ators are
calculated by de‡ating domestic and imported inputs separately.
Capital intensity is measured as annualized user cost of capital (including leased capital)
k = (r +
relative to hours worked. The cost of capital is calculated as Rit
k
k )Kit ;
where Kitk
is the real net capital stock of type k, for …rm i at time t, k is either buildings and land
(b) or other tangible …xed assets (o),50 r is the real rate of return, which we calculated
from the average real return on 10-year government bonds in the period 1996-2004 (4:2
percent), and k is the median depreciation rates obtained from accounts statistics. The
b + Ro :
total cost of capital is de…ned as Rit
it
Relative hourly wage costs for workers in manufacturing is a trade-weighted measure of
relative wages measured in a common currency. The index is produced and updated
annually by the Technical Calculating Committee on Income Settlements (Teknisk Beregningsutvalg, TBU).51 We use this measure proxying for
P1
P0 in the econometric
analysis. Note that our identi…cation strategy is completely invariant to the choice of
RER. The RER measure will, however, a¤ect the magnitude of the estimated :
Skill intensity is de…ned as the number of highly-skilled (tertiary education, i.e. +12
years) employees relative to total employment in each NACE 2-digit sector in 2000, source:
Statistics Norway.
The Her…ndahl index is calculated in the base year (1996) at the 2-digit level.
Swedish employment refers to the number of persons employed in a given NACE 3-digit
sector. The data is collected from Statistics Sweden’s webpages and then manually linked
to the Norwegian data set.52
A …rm exits if employment is positive in year t and zero or missing in year t + 1, 1996
t < 2003: A …rm enters if employment is zero or missing in year t and positive in year
t + 1, 1996
t < 2003:
The industry classi…cation is 5-digit NACE. A …rm’s NACE code is based on the …rm’s
main activity. Further details on the variables in the …rm database are provided by Raknerud,
Rønningen and Skjerpen (2004).53
49
http://www.ssb.no/english/subjects/08/02/20/ppi_en
The latter group consists of machinery, equipment, vehicles, movables, furniture, tools, etc.
51
http://www.regjeringen.no/nb/dep/aid/tema/Inntektspolitikk/rapporter-fra-tbu.html
52
The Swedish data is available at: http://www.ssd.scb.se/databaser/makro/Produkt.asp?produktid=NV0109
53
http://www.ssb.no/english/subjects/10/90/doc_200416_en/doc_200416_en.pdf
50
31
390 400
250
.14
220
350 360 370 380
Hours, 1 mill
Employment, 1000
230
240
.13
.12
.11
.1
.09
1996
1998
2000
Year
2002
1996
1998
2000
Year
Employment, 1000
2002
2004
Hours, 1 mill
160
40
100
60
120
80
140
100
180
Exit rate (t to t+1)
120
Entry rate (t to t+1)
2004
1996
1998
2000
Year
2002
Exports, bill NOK
2004
1996
Imports, bill NOK
1998
2000
Year
Capital/worker, 1996=100
2002
2004
Value added, bill NOK
Figure 5: Entry/exit rate, employment, exports/imports and capital intensity, 1996-2004.
Figure 5 displays some key trends in our data, from 1996 to 2004.54 The top left graph
shows the entry and exit rate; the top right graph shows total manufacturing employment (in
persons and hours); the bottom left graph shows total manufacturing exports and imports (in
current prices); and the bottom right graph shows capital intensity - measured as real total
capital cost relative to total employment - and real value added.
A.2
Firm entry and exit
The use of a di¤erence-in-di¤erences methodology in this paper means that we can only analyze
the impact of the RER shock on continuing …rms. The productivity decomposition in Table
2 suggests that the within-…rm productivity improvements were the most important source of
aggregate productivity gains during the shock years. Furthermore, the aggregate entry and exit
rates presented in Figure 5 also suggest that there was no systematic relationship between the
RER shock and entry/exit. In this section, we …rst investigate the e¤ect of the RER shock on
the likelihood of …rm exit formally. Speci…cally, we estimate the following probit
Exiti = 1 [ +
where vi is i.i.d. normal,
i0
is an intercept term,
+ xi0 + vi ]
i0
is net exposure in the base year, and xi0 is
the same vector of controls used in the baseline speci…cation. Exiti = 1 if the …rm has positive
54
The graphs are based on the unbalanced set of …rms.
32
employment in 2000 but not in 2004, and Exiti = 0 if employment is positive in both 2000
and 2004. Table 9 reports the results, showing that net exposure did not have a signi…cant
impact on the exit rate during the shock years. Firm size and labor productivity, however, are
positively correlated with lower exit rates.
Table 9: The probability of exit. 2000-2004.
Net exposure
.092
(.116)
-.079a
(.017)
-.161a
(.038)
.018
(.049)
-.060
(.046)
Yes
5313
Firm size
Labor productivity
Export dummy
Import dummy
Industry dummies
No. of obs.
Note: All independent variables are from the base year, 1996.
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level
Robust standard errors in parentheses.
Second, we check the average productivity of entrants and exitors relative to continuing
…rms before and during the shock. The hypothesis here is that although the exit/entry rate
did not change signi…cantly during the shock years, the composition of exitors/entrants might
have changed due to the shock. Comparing relative productivity for the three groups - exitors,
entrants and survivors - reveals that exitors were always less productive than the survivors, while
entrants were more productive than the survivors in the pre-shock period (1996-2000), but not
so during the shock (2000-2004). All in all, these descriptives suggest that the contribution of
net entry to productivity was relatively small.55
Table 10: Average labor productivity of Survivors, exitors and entrants
t
2000
2004
Relative to the whole economy
manufacturing labor productivity in period t-4
Entrants,t Exitors,t-4 Survivors,t
1.16
0.90
1.05
1.22
1.04
1.25
Note: Labor productivity is averaged using employment shares as weights.
Numbers for entrants and survivors refer to productivity in t relative to total manufacturing productivity in t-4.
Numbers for exitors refer to productivity in t-4 relative to total manufacturing productivity in t-4.
55
In contrast to e.g. Disney, Haskel and Heden (2003) and Scarpetta et al. (2002).
33
A.3
Industry versus …rm-level de‡ators
In this section we ask to what extent our de‡ation procedure a¤ects the measurement of labor
productivity. Figure 6 shows aggregate labor productivity for two subgroups of …rms, those
with negative net exposure in the …rst year (left panel) and those with positive net exposure in
the …rst year (right panel). The solid line represents total value added relative to total hours
worked (in logs), where value added is de‡ated using the method described in the main text.
The dotted line represents labor productivity de‡ated using only NACE 3-digit output price
indices, and the right panel clearly shows that this declined in 2002, the year the RER was
the strongest. The …rst method, however, does not exhibit the same pattern. This pattern is
consistent with the hypothesis that our …rm-level de‡ators successfully remove relative price
movements from our preferred productivity measure.
Positive net exposure
-1.2
-1
-.8
-.6
Negative net exposure
1996
1998
2000
2002
20041996
1998
2000
2002
2004
Year
LP, firm-level deflators
LP, industry-level deflators
Graphs by netexp
Figure 6: Measurement of labor productivity.
A.4
Tables
Table 11: Increased competition and industrial performance. No …rm controls. Industry dummies.
Dependent
variable
Net exposure
Firm controls
Industry dummies
No. of obs.
Employment
(in hours)
SURE
-.891
(.617)
No
Yes
3714
Production
SURE
1.163a
(.471)
No
Yes
3714
Labor
productivity
SURE
3.129a
(.619)
No
Yes
3714
TFP
SURE
.824a
(.208)
No
Yes
3714
Note: All independent variables are from the base year, 1996.
yi1
yi0 :
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level
The dependent variable takes the form
34
Capital
intensity
SURE
.654
(1.137)
No
Yes
3714
Table 12: Increased competition and industrial performance. Firm controls and industry dummies.
Dependent
variable
Net exposure
Firm controls
Industry dummies
No. of obs.
Employment
(in hours)
SURE
-.376
(.602)
Yes
Yes
3714
Production
SURE
.729
(.481)
Yes
Yes
3714
Labor
productivity
SURE
2.118a
(.562)
Yes
Yes
3714
TFP
SURE
.474a
(.200)
Yes
Yes
3714
Capital
intensity
SURE
1.181
(1.149)
Yes
Yes
3714
Note: All independent variables are from the base year, 1996.
yi1
yi0 :
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level
The dependent variable takes the form
Table 13: Increased competition and industrial performance. Independent variables from 2000. Firm
and industry controls.
Dependent
variable
Net exposure
Firm controls
Industry controls
No. of obs.
Employment
(in hours)
SURE
.378
(.641)
Yes
Yes
3701
Production
SURE
1.174a
(.532)
Yes
Yes
3701
Labor
productivity
SURE
1.775a
(.574)
Yes
Yes
3701
TFP
SURE
.238
(.212)
Yes
Yes
3701
Capital
intensity
SURE
1.239
(1.256)
Yes
Yes
3701
Note: All independent variables are from 2000.
yi1
yi0 :
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level
The dependent variable takes the form
Table 14: Increased competition and industrial performance. Independent variables from 2000. No …rm
controls. Industry dummies.
Dependent
variable
Net exposure
Firm controls
Industry dummies
No. of obs.
Employment
(in hours)
SURE
-1.435a
(.597)
No
Yes
3714
Production
SURE
1.215a
(.456)
No
Yes
3714
Labor
productivity
SURE
4.252a
(.597)
No
Yes
3714
TFP
SURE
1.050a
(.201)
No
Yes
3714
Note: All independent variables are from 2000.
yi1
yi0 :
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level
The dependent variable takes the form
35
Capital
intensity
SURE
1.909c
(1.100)
No
Yes
3714
Table 15: Increased competition and industrial performance. Independent variables from 2000. Firm
controls and industry dummies.
Dependent
variable
Net exposure
Firm controls
Industry dummies
No. of obs.
Employment
(in hours)
SURE
.336
(.558)
Yes
Yes
3714
Production
SURE
1.215a
(.461)
Yes
Yes
3714
Labor
productivity
SURE
2.472a
(.502)
Yes
Yes
3714
Capital
intensity
SURE
1.140
(1.092)
Yes
Yes
3714
TFP
SURE
.583a
(.007)
Yes
Yes
3714
Note: All independent variables are from 2000.
yi1
yi0 :
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level
The dependent variable takes the form
Table 16: Increased competition and industrial performance. Gross exposure. Firm and industry controls.
Dependent
variable
E exposure
I exposure e
Firm controls
Industry controls
No. of obs.
Employment
(in hours)
.738
(.872)
-3.929a
(.964)
Yes
Yes
3701
Production
1.216c
(.701)
-.677
(.776)
Yes
Yes
3701
Labor
productivity
1.377c
(.811)
-4.065a
(.897)
Yes
Yes
3701
TFP
.344
(.290)
-.674b
(.320)
Yes
Yes
3701
Capital
intensity
.835
(1.669)
-3.591b
(1.845)
Yes
Yes
3701
Note: All independent variables are from 1996. The dependent variable takes the form
yi1
yi0 :
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level.
Table 17: Increased competition and industrial performance. Gross exposure. No …rm controls, industry
dummies.
Dependent
variable
E exposure
I exposure e
Firm controls
Industry dummies
No. of obs.
Employment
(in hours)
-1.747a
(.710)
-.0645
(.884)
No
Yes
3714
Production
1.773a
(.542)
-.071
(.675)
No
Yes
3714
Labor
productivity
5.419a
(.709)
.975
(.882)
No
Yes
3714
TFP
1.566a
(.239)
.506c
(.297)
No
Yes
3714
Capital
intensity
1.131
(1.310)
.199
(1.629)
No
Yes
3714
Note: All independent variables are from 1996. The dependent variable takes the form
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level.
36
yi1
yi0 :
Table 18: Increased competition and industrial performance. Gross exposure.Firm controls and industry
dummies.
Dependent
variable
E exposure
I exposure e
Firm controls
Industry dummies
No. of obs.
Employment
(in hours)
.568
(.594)
-.989
(.738)
Yes
Yes
3714
Production
1.283b
(.648)
-3.173a
(.807)
Yes
Yes
3714
Labor
productivity
1.169c
(.692)
-3.643a
(.860)
Yes
Yes
3714
TFP
.360
(.247)
-.660b
(.307)
Yes
Yes
3714
Capital
intensity
.178
(1.417)
-2.794
(1.760)
Yes
Yes
3714
Note: All independent variables are from 1996. The dependent variable takes the form
yi1
yi0 :
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level.
Table 19: Increased competition and industrial performance. TFP based on value added production
function.
Dependent
variable
Net exposure
Firm controls
Industry controls
No. of obs.
Employment
(in hours)
-1.297c
(.725)
Yes
Yes
3714
Production
.980c
(.581)
Yes
Yes
3714
Labor
productivity
2.547a
(.672)
Yes
Yes
3714
TFP
2.158a
(.596)
Yes
Yes
3714
Capital
intensity
2.019
(1.383)
Yes
Yes
3714
Note: All independent variables are from 1996. The dependent variable takes the form
a signi…cant at the .01 level, b signi…cant at the .05 level, c signi…cant at the .1 level.
37
yi1
yi0 :
Table 20: Currency exposure, employment and productivity growth by industry
Exposure
Nace
15
16
17
18
19
20
21
22
24
25
26
27
28
29
30
31
32
33
34
35
36
37
0.17
0.05
0.22
0.21
0.08
0.11
0.67
0.00
0.48
0.26
0.18
0.46
0.13
0.27
0.25
0.26
0.49
0.36
0.46
0.15
0.22
0.31
e
0.12
0.19
0.34
0.25
0.17
0.13
0.18
0.03
0.22
0.34
0.29
0.27
0.10
0.20
0.12
0.25
0.20
0.25
0.27
0.09
0.21
0.19
is export exposure,
change in interval
0.05
-0.14
-0.12
-0.04
-0.09
-0.02
0.49
-0.03
0.26
-0.08
-0.10
0.19
0.03
0.08
0.13
0.02
0.30
0.12
0.19
0.06
0.01
0.12
Hours worked
0
1
9.2
4.8
-1.2
0.7
-5.6
10.4
2.5
3.7
7.1
9.4
4.1
6.8
13.7
8.8
-2.1
6.0
8.2
12.8
24.4
10.7
4.9
-3.2
-2.2
-3.9
-4.4
-8.1
-4.1
0.0
-6.3
-3.2
-5.5
-4.8
-4.6
-3.7
-6.9
-3.7
-5.8
-8.7
-4.0
9.4
-4.8
-6.0
-7.8
-1.5
e import exposure and
1
Labor productivity
0
0
1
-11.4
-8.7
-3.2
-8.8
1.5
-10.4
-8.8
-6.9
-12.5
-14.2
-8.7
-10.5
-20.6
-12.5
-3.7
-14.7
-12.2
-3.4
-29.2
-16.7
-12.6
1.7
8.1
4.6
0.5
3.1
13.7
2.4
-5.1
-6.2
11.0
-6.5
-0.2
7.9
-2.4
-2.4
-1.0
0.2
2.1
2.2
4.5
12.6
1.7
7.1
0.5
2.5
6.3
3.8
3.1
5.6
11.1
4.4
12.8
3.3
3.4
12.8
5.1
2.4
12.8
-1.8
6.5
13.1
6.0
5.6
1.6
18.4
net exposure.
1
0
-7.7
-2.1
5.8
0.7
-10.6
3.1
16.2
10.5
1.7
9.8
3.6
4.9
7.5
4.7
13.8
-2.0
4.4
10.9
1.5
-7.0
-0.1
11.3
i refers to the percentage
i (1996-2000 or 2000-2004): Exposure is calculated in the base year.
38
Table 21: The NACE Rev. 1 industrial classi…cation.
Nace
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Manufacture of food products and beverages
Manufacture of tobacco products
Manufacture of textiles
Manufacture of wearing apparel, dressing and dyeing of fur
Tanning and dressing of leather, manufacture of luggage, handbags, saddlery, harness and footwear
Manufacture of wood and of products of wood and cork, except furniture (..)
Manufacture of pulp, paper and paper products
Publishing, printing and reproduction of recorded media
Manufacture of coke, re…ned petroleum products and nuclear fuel
Manufacture of chemicals and chemical products
Manufacture of rubber and plastic products
Manufacture of other non-metallic mineral products
Manufacture of other non-metallic mineral products
Manufacture of fabricated metal products, except machinery and equipment
Manufacture of machinery and equipment n.e.c.
Manufacture of o¢ ce machinery and computers
Manufacture of electrical machinery and apparatus n.e.c.
Manufacture of radio, television and communication equipment and apparatus
Manufacture of medical, precision and optical instruments, watches and clocks
Manufacture of motor vehicles, trailers and semi-trailers
Manufacture of other transport equipment
Manufacture of furniture, manufacturing n.e.c.
Recycling
39