Real Exchange Rate Movements, Hysteresis, and the
Decline of American Manufacturing
Douglas L. Campbell1
[email protected]
UC Davis
September, 2013
Abstract
This study investigates the decline of American manufacturing employment in the early
2000s, comparing it to the smaller decline in the 1980s. I find that US manufacturing,
and particularly sectors with greater initial exposure to trade in the 1970s and late1990s were disproportionately affected by ensuing dollar appreciations and concomitant
increases in relative unit labor costs in manufacturing. Employment losses in both
periods were due to increased job destruction and suppressed job creation, and appear
to exhibit hysteresis. Additionally, more open sectors experienced relative declines in
shipments, value-added, production worker wages, and total factor productivity when
US relative unit labor costs in manufacturing were high. I explain the persistent effects
of exchange rate movements on manufacturing using a Melitz model extension with sunk
fixed costs, which naturally leads to a dynamic gravity equation whereby shocks to trade
have persistent effects which decay over time. The appreciation of US relative prices and
wages can plausibly explain the collapse in American manufacturing employment in the
early 2000s. Theory and data imply the potential for long-lasting adverse consequences
of currency overvaluation due to government policies.
JEL Classification: F10, F16, F41, N60, L60
Keywords: Exchange Rates, American Manufacturing, Hysteresis, Trade
1
Special thanks is in order to brownbag participants at UC Davis and seminar participants at Cal
State Sacramento. I benefitted enormously from feedback from Paul Bergin, Robert Feenstra, Chris
Meissner, Katheryn Russ and Deborah Swenson. The ideas for this paper emanated from courses
taught by Robert Feenstra, Katheryn Russ, Paul Bergin, and Christopher Meissner, and I am deeply
indebted to all. This research benefitted enormously from the new and much improved v8.0 of the
Penn World Tables, so I am also indebted to Robert Feenstra, Robert Inklaar, and Marcel Timmer.
Thanks is also in order to Wayne Gray both for managing the NBER-CES manufacturing page, a
wonderful resource, and for replying graciously to my suggested improvements for the NAICs to SIC
concordance. This research project began when I was tasked to produce a report on manufacturing
for a cabinet meeting while working as a Staff Economist at the President’s Council of Economic
Advisers, and benefitted immensely from conversations I had with Chad Bown, Lee Branstetter,
Michael Klein, and Jay Shambaugh. All opinions and errors herein are my own.
1 Introduction
A central tenet of economics is that prices matter. In this paper I propose a corollary:
sunk costs imply that historical prices can affect current economic outcomes. Specifically,
I investigate the effects of large movements in exchange rates, caused by arguably exogenous changes in policy, on American manufacturing sectors with differential exposure to
international trade, and test for evidence of hysteresis.
When the dollar appreciated sharply in both the 1980s and from 1997-2002, manufacturing sectors with higher initial shares of trade suffered disproportionately compared to
other manufacturing sectors and to the rest of the economy. They also performed worse
than the same industries in other major economies. These sectors experienced large
declines in employment mostly due to an increase in job destruction but also due to a
decline in job creation. They experienced relative declines in shipments, value-added,
total factor productivity, and modest declines in hourly wages for production workers
a borderline significant decline in investment. The measured elasticity with respect to
employment and the magnitude of the appreciation in US relative unit labor costs are
large enough to explain most of the collapse in the level of American manufacturing
employment in the early 2000s.
After the dollar returned to previous levels in both cases, the employment collapses
stabilized, but the jobs lost did not return. Indeed, this pithy observation for aggregate trade in the 1980s spawned a large theoretical literature on hysteresis, with the
original progenitors of increasing returns and new trade theory, including Dixit (1989a,
1989b, 1991, 1992), Krugman (1987, 1988), Krugman and Baldwin (1987, 1989), and
Baldwin (1988, 1990) all weighing in with multiple contributions. I document that this
phenomenon holds up for both time periods, and that the hysteresis was broad-based
and focused on more tradable industries.
Traditionally, it has been thought that unit labor costs are the best price-based measure of international competitiveness. However, real exchange rate indices computed
from unit labor costs produced by the IMF and OECD have a number of drawbacks.
The first is that manufacturing output is deflated using country-specific deflators which
are constructed idiosyncratically which can lead to bias over time without the use of
multiple benchmarks. This is the same problem which afflicted older vintages of the
Penn World Tables which predated version 8.0. Second, these measures are computed
as indices-of-indices, and as such do not properly account for compositional changes in
trade toward countries, such as China, that have systematically lower unit labor costs.
Additionally, China and many other developing countries are not even included in the
1
IMF’s relative unit labor cost (RULC) index, which also uses fixed trade weights that
have become outdated. In this paper, I address all of these concerns by using a weighted
average of relative unit labor costs (WARULC) index computed for the manufacturing
sector using data from all six ICP benchmarks including developing countries such as
China.2 I show that these indices do a remarkably good job of predicting employment
declines in US manufacturing, and in particular do much better than CPI-based real
exchange rate indices or the RULC indices created by the Federal Reserve, the IMF
and the OECD. I also show that either of the other “weighted average relative” (WAR)
exchange rates such as the WARP index created by Fahle et al. 2008 or the BalassaSamuelson adjusted WARP index created by Campbell (2013b) yield very similar results
as WARULC. The details of the construction of these indices are included in a companion
paper, Campbell (2013b).
Just as today, much blame in the public and media is focused on China, in the 1980s
much negative attention, which many observers feel was misplaced, revolved around the
key US trading partner Japan. Japan is an especially useful case study since the Bank of
Japan heavily managed the Yen with capital controls until 1980, with full liberalization
coming following the Yen-Dollar Accord in 1984. Following the 1985 Plaza Accord
agreement to intervene in the currency markets, the Yen then appreciated sharply vs.
the US dollar. In the 1980s, Japan was thought to be dominating the marketplace due
to the uniqueness of Japanese culture and business practices, such as such as kaizen
costing and kanban scheduling. While cultural differences may have been important, I
show that while Japanese industries were able to gain US market share rapidly when the
Yen was relatively weak, Japanese industries consolidated their gains but did not make
any further inroads after the Yen appreciated. The US-Japan trading relationship was
in no way special in this regard, as the same phenomenon holds for other US trading
partners with relatively weak currencies in the mid-1980s. In addition, Japanese exports
to other markets were also sensitive to movements in relative unit labor costs. For
example, Japanese exports flooded UK markets in the early 1980s (in contrast to the
mid-1980s in the US), when the pound was relatively strong.
While the main focus of this paper is empirical, I briefly sketch a heterogenous-agent
increasing-returns trade model in the Melitz (2003) and Chaney (2008) tradition, showing that sunk fixed entry costs imply that trade patterns will depend on both present
and past trade shocks. I show that the presence of sunk costs implies that real wages are
a function of historical market access. In addition, sunk costs imply that government
2
I am greatly indebted to Professor Paul Bergin for suggesting I apply the Fahle et al. (2008) insight
to unit labor costs.
2
spending can cause dynamic crowding out, and that this effect can be extenuated in the
presence of trade – a phenomenon realized with the large twin deficits of the 1980s. In a
related paper (Campbell 2013a), I show that incorporating dynamics into gravity analysis as implied by increasing returns models can yield dramatically different estimates for
the impact of currency unions on trade as compared with static gravity specifications.
There is a solid foundation of research on the contemporaneous impact of exchange
rate movements on manufacturing. Most notably, Klein, Schuh, and Triest (2003) find
that exchange rate appreciations affect job creation and destruction in the US over the
period 1973-1993, but that dollar depreciations do not. However, while a majority of
studies find the expected impact, the question is not settled, and much of the literature
uses small data sets which are becoming dated.
This study builds on Klein et al. (2003) first by showing that the results hold when
extending the sample through 2009, important because this includes the period when
American manufacturing collapsed. It can be argued that the period in the Klein et
al. study were especially fortuitous in terms of research design, since the US dollar
appreciation in that period was especially sharp and the two main causes were arguably
“exogenous” – the structural US government deficits in the 1980s plus factors not related
to any known economic fundamentals (Meese and Rogoff, 1983). However, given that
there will never be any such thing as a perfect natural experiment, showing that results
hold out-of-sample is a crucial element of research in empirical international trade.
In terms of the research design, I add an additional dimension to the difference-indifference approach using international data, showing that when the dollar appreciates,
while the more exposed sectors in the US are adversely affected, the same sectors in other
major economies are not.3 I also run a reduced-form regression using changes in defense
spending and changes in the US government budget deficit ex-automatic stabilizers, and
show that both of these are correlated with adverse performance of more tradable sectors.
Lastly, I include a bevy of additional robustness checks, some of which are new to this
literature. These include estimation using a quantile regression which is less sensitive
to outliers, clustering at the sectoral level, clustering at the two-digit SIC level in crosssectional regressions, and including an additional 20 controls. The controls include
real interest rates and energy prices (and interactions between these variables and the
industries that borrow the most and have the highest energy intensity), domestic demand
by industry, productivity growth, tariffs, changes in insurance and freight costs, controls
for low-markup industries interacting with exchange rate movements, and the share of
imported intermediate inputs.
3
I thank Thomas Wu for giving me this idea as a robustness check.
3
In general, if some sectors become more productive, or if US financial assets become more popular, leading to a dollar appreciation, while workers and firms in lowproductivity sectors may suffer due to trade, the overall economy-wide gains from trade
are likely more than large enough to compensate the losers and make everyone better
off. In this case, while documenting how relative price movements impact more tradable
sectors would still be interesting, it may not have straightforward policy relevance.
However, the logic suggesting benign impacts of trade competition does not necessarily
apply to situations in which a central bank is constrained from responding to negative
impacts of currency overvaluation due to the zero lower bound on nominal interest rates.
In this case, capital inflows can reduce aggregate demand.4 This is an important caveat
given that in the past 15 years, the rise of developing country official capital flows has
been unprecedented; in this new “Bretton Woods II” system exchange rates are actively
managed as a key element of the developing-country central bank policy mix.5 An
alternative argument to the zero lower bound as an actual constraint, is the idea of “selfinduced paralysis”. Central bankers are accustomed to controlling the economy using
the federal funds rate and are generally uncomfortable with “extraordinary” measures,
and so become passive in the face of a mental constraint.6 In this case, as well, more
capital inflows can reduce aggregate demand and lead to more debt accumulation7 , and
so a finding that exchange rate movements have large and persistent effects is relevant
for policymakers wishing to understand how to respond to the ongoing Great Reserve
Accumulation while operating in a liquidity trap. In addition, the surprisingly swift
and persistent decline of American manufacturing, and the secular stagnation of the
American economy since 2000 have not been fully accounted for.
Theory and evidence provide a cautionary tale for currency overvaluation due to
government policy, including running large-scale, structural deficits while at full employment. On the other hand, the results do not sanction currency undervaluation as
industrial policy, as any non-market distortions in exchange rates could create losses
for consumers, which may be dynamic, and have other damaging distortions.8 While
4
Paul
Krugman
has
argued
this
point
repeatedly
on
his
blog,
e.g.,
here:
http://krugman.blogs.nytimes.com/2009/05/15/china-and-the-liquidity-trap/.
5
See the series of papers by Dooley et al (2004, 2005, 2007, and 2009).
6
There is debate about whether monetary policy can be effective right now (e.g., Vissing-Jorgenson and
Krishnamurthy 2011 for the US and Ugai 2007 for Japan both find impacts of QE). It is true that
surprise Federal Reserve policy pronouncements since 2009, including the recent talk of tapering,
have continued to have an effect on markets, even while most of the QE over the past few years has
merely increased holdings of excess reserves. Thus my interpretation of the current situation is one
where the usual policy channel is not available, but where other channels can still be effective.
7
See, for example, Bergin (2000).
8
For example, many poor countries today are net food importers and have issues with nutrition and
4
economists have long been skeptical of deficit spending outside of recessions on theoretical and intuitive grounds, this paper measures tangible, long-lasting negative effects due
to the dynamic crowding out of the tradables sector.
The next section provides an overview of the recent decline and non-recovery of US
manufacturing. I then introduce a slight variation of the Melitz (2003) model, similar
to Chaney (2008), discuss interesting dynamic implications of the model, and use it to
motivate the empirical sections which follow.9
2 The Decline and Non-Recovery of US Manufacturing
In Figure 1, it can be seen that from the mid-1960s to 2000, the level of American
manufacturing employment varied considerably with the business cycle, but showed only
a slight overall decreasing trend. From 2000-2003, the manufacturing sector suddenly
lost 3 million jobs. As the economy grew from 2003-2007, the jobs lost did not return. In
the aftermath of the financial crisis in 2008, the manufacturing sector lost an additional
2.3 million jobs. By 2012, the economy had regained 500,000 of these jobs, but in the
past year, even as the economy has continued to grow, manufacturing employment has
been roughly flat, indicating that many of these job losses are likely to be permanent.
It has generally been thought by wise economists that the foolish public’s concern
with trade and offshoring as an explanation for the decline of American manufacturing employment was entirely misplaced. The real culprit was thought to be productivity
gains in manufacturing and a structural shift in production toward services, which would
imply that the decline in manufacturing is a sign of advancement. However, in Figure
4(a) it can be seen that productivity growth in terms of value-added per worker in manufacturing has been roughly constant since 1947.10 In addition, there is evidence that
perhaps one-fifth to one-half of the measured growth in value-added per worker from
1997-2007 reflects upward bias due to the dramatic increase of imported intermediate
inputs, according to Houseman et al. (2010). Even ignoring potential bias in the productivity statistics, while the pace of productivity growth has been faster in manufacturing
hunger – very serious problems which could affect growth and be exacerbated by an undervalued
currency leading to higher food prices and more food being exported out of the country.
9
For interested parties, in the Not-for-publication version of the Appendix I present a version of the
model with additional assumptions such as consumer dynamics similar to Ghironi and Melitz (2005),
which additionally assumes sunk costs. The key findings are little changed, so the following section
applies Occam’s razor.
10
For comparison, Figure 26 in the appendix also provides a chained output index, which yields similar
results although it does show slightly faster growth in the 2000s than before, which is easily visible
in the five and 15 year moving averages in panel b.
5
American Manufacturing Employment
Millions
22
Share
12%
20
employment
(left axis)
10%
18
8%
16
6%
14
share of population
(right axis)
4%
12
10
2%
1960
1970
1980
1990
2000
2010
Figure 1: American Manufacturing Employment, 1965-2012
than in the economy as a whole, the consistency of measured manufacturing productivity growth would ostensibly make it a strange explanation for a sudden employment
collapse.
And while services today are a much larger share of the overall economy than they
used to be, in Figure 4(b) it can be seen that the services share of exports has not
increased much over the past several decades. Thus it is not the case that the decline
in manufacturing has been replaced by tradable services. In fact, the trade surplus
in services actually shrank by one-third over the period 1997-2004, while the goods
deficit ex-manufacturing increased substantially. Thus, the decline in manufacturing
was actually a broad-based decline in tradables sectors, including in agricultural produce,
animal husbandry, forestry, and fish, with natural gas and metal ores being two notable
exceptions. Productivity gains and a structural shift toward services would seem better
candidates to describe the gradual long-term decline in manufacturing employment as a
share of the population as in Figure 1 rather than a sudden collapse.
By contrast, a simple accounting approach to determine the source of jobs lost suggests that the increase in the manufacturing trade deficit from 1995-2005 is actually
more than enough to explain the full decline in manufacturing employment over that
period, given the realized path of productivity growth. An alternative phrasing is that
given consumption of manufactured goods over the period 1995-2005, had there been
6
Services Exports, Share of Total
Manufacturing Value Added Per Worker
Log Scale
13
100%
12
80%
11
60%
10
40%
9
20%
8
7
0%
1947
1957
1967
1977
1987
1997
2007
1992
(a) Value-Added per Worker
1996
2000
2004
2008
(b) Services Share of Exports
Figure 2: The Usual Suspects: Explanations for the Decline of Manufacturing
no increase in the manufacturing trade deficit, the path of productivity growth actually
implies that manufacturing employment should have increased. One problem with such
an approach is that the decline in manufacturing employment over this period was likely
concentrated in low-productivity workers, so that the decline itself may have caused an
increase in observed manufacturing productivity. A second problem is that the collateral
damage from tradable sectors to the rest of the economy likely lowered the overall pace
of consumption growth, including of manufactured goods.
More recently, much research has focused on the role of trade liberalization and the
rise of China. For example, Pierce and Schott (2012) argue that China’s ascension to
the WTO removed trade policy uncertainty and led to a large increase in imports from
China, reducing US manufacturing employment. Autor, Dorn, and Hanson (forthcoming) find that import competition from China explains one-quarter of the aggregate loss
in manufacturing employment through 2007, while Acemoglu et al. (2013) argue that
import competition from China suppressed overall U.S. employment growth directly via
its impact on manufacturing and indirectly via input-output linkages with the rest of
the economy. Ebenstein et al. (2012) also document a series of facts consistent with the
idea that Chinese import competition reduced US manufacturing employment.
Interestingly, the exchange rate does not typically feature prominently in these stories,
perhaps because the most commonly used real exchange rate index for the US, the
Federal Reserve’s broad trade-weighted real exchange rate index, shows much less of
a dollar appreciation in the early 2000s than it did in the 1980s, when manufacturing
job losses were much less pronounced. However, the Fed’s measure was created using
Divisia, a weighted average of changes in bilateral real exchange rate indices in which
7
all countries, including China and Switzerland, are assigned the same base year price
levels. Thus this method does not account for compositional changes in trade toward
.8
1
1.2
1.4
Goods Trade Balance ex−Oil, GDP Share
−.04
−.02
0
.02
countries, such as China, with systematically lower price levels (see Fahle et al. 2008).
1950
1960
1970
1980
1990
Goods Trade Balance ex−Oil, GDP Share
RER Index (Fed’s Method)
2000
2010
WARP v8.0
Figure 3: Real Exchange Rate Measures vs. the Current Account
Figure 3 plots an extension of the Fed’s measure back to 1950 using the Fed’s methodology and an updated version of the weighted-average relative price (WARP) index
introduced in Fahle et al using PWT v8.0 vs. the US Goods trade balance, ex-Oil. This
new version of WARP is much closer to being a near mirror image the goods trade deficit
since 1980, and shows that the early 2000s shock to relative prices was actually substantially larger than the dollar appreciation episode of the 1980s. Additionally, after the
exchange rate shocks in the 1980s, trade returned to balance much more slowly than the
exchange rate. After the 2000s shock, although it appears as though the trade balance
improved with the decline in the WARP index after 2006, much of this was actually
the result of a depressed economy reducing demand for imports. As the economy has
(slowly) recovered, the goods trade balance ex-oil has continued to worsen, lagging the
improvement in relative prices.
Although much emphasis has been placed on the decline of manufacturing since 2000,
8
there was actually a first period of decline in manufacturing employment not associated
with the business cycle, from 1979 to 1986. This era coincided with a 45.4% appreciation of the dollar according to the Fed’s Broad Trade-Weighted Dollar index and the
emergence of “twin” deficits, fiscal and trade. In 1979, a business cycle peak, the government’s deficit had been .5% of GDP, but this increased to 4.4% of GDP by 1986,
while the current account deficit increased from 1.1% in 1979 to 3.2% in 1986.
The second period of anomalous employment behavior begins in the late 1990s with
the WARP index appreciation induced by rising trade with China and other developing
countries, devaluations following the Asian Financial Crisis, the unprecedented ensuing
dollar reserve accumulation by developing countries (see Meissner and Taylor, 2006),
and the structural fiscal deficits starting in 2001. Textbook theory, as does the theory
presented in the next section, suggests that an increase in fiscal deficits should lead to
a real exchange rate appreciation and a shift of resources out of the tradables sector.
Aggregate data appear to confirm theory in the 1980s and in the early 2000s.
As manufacturing employment has historically been strongly procyclical, it is helpful to plot the “structural” level of manufacturing employment with the cyclical portion
deducted vs. the Fed’s broad, trade-weighted real exchange rate (Figure 4.a). The structural level of manufacturing employment was derived by regressing log manufacturing
on log real GDP lagged one period and a trend up until the exchange rate appreciated,
and then subtracting out the predicted level of manufacturing employment based on
realizations of GDP and adding back in the conditional average, with the levels plotted
for the entire period. I find that manufacturing seems to respond to both GDP and the
real exchange rate with a lag (as have others), but the key results do not hinge on this.
In Figure 27 in the appendix, I do the same for the late 1990s. Note also that there is no
apparent correlation between the small quarterly gyrations in the broad, trade-weighted
RER and manufacturing employment, but when the RER appreciates substantially, there
is a large movement in manufacturing employment and the CA balance.
Figure 5 plots the import share of trade vs. a one year lag of the Fed’s RER index.
When a simple trend is fit to the data in panel (a), there appears to be only a slight
correlation that is not statistically signficant. When the observations are connected
by year, in panel (b) however, the relationship appears to be much more systematic.
Dollar appreciations lead to movements to the northeast. Dollar depreciations move the
relationship southwest. Shifts in the curve to the northwest indicate relative declines in
ability to produce tradable goods. Sharp dollar appreciations are followed by movements
to the northwest, as happened after the 1980s dollar bubble and 2002. The weak dollar
period in the late 1970s was followed by a slight shift in the curve to the southeast,
9
.002
130
Explaining Log Changes in Structural Manufacturing
80
-.004
90
-.002
100
110
RER
0
120
Structural Log Manufacturing Employment
9.7
9.75
9.8
9.85
US Manufacturing and the RER
1970
1975
1980
year
1985
1990
Structural Log Manufacturing Employment
1970
RER
1980
1990
year
2000
Prediction using RER
(a) Structural Man. Employment vs. the RER
2010
Log Change in Employment
(b) Predictions
.65
.65
Figure 4: Structural Manufacturing Employment vs. the RER
1999
2008
2010
1988
2009
1986
2003
2000
1985
2001
1984
1998
1995 1994
1989
1996
1997
1993
1990
1972
1983
1992
1991
1971
1978
1977
1973
1980
1982
1974
1976
1981
2007
2011
2009
.9
2000
1999
2004
1986
2003
2002
1987
1985
2001
1988
1984
1972
1983
1971
1978
1977
1973
1982
1979
1976
1974
1981
1980
1975
.8
2008
2010
1998
1994
19951989
1996
1997
1993
1990
1992
1991
.45
.45
1979
2005
2006
2002
US Import Share of Total Trade
.5
.55
.6
US Import Share of Total Trade
.5
.55
.6
2004
2005
2006
1987
2007
1975
1
Divisia
1.1
1.2
.8
(a) Correlation (not statistically significant)
.9
1
1.1
Fed’s Divisia Index, lagged one year
1.2
(b) Movement by Year
Figure 5: Fed’s RER Index and the Import Share of Trade
and while the curve was stable following a weak dollar period in the 1990s, this was
the period which saw rising trade with China and other developing countries. In sum,
the relationship of the US dollar since the end of Bretton Woods seems to validate
Krugman’s (1988) thesis – large exchange rate movements have persistent effects.
3 Theory
3.1 The Model
In this section, I motivate the empirics in the next section using a slight variation of the
Chaney (2008) model with sunk costs as in Melitz (2003). In this model, households in
10
the home country consume from a continuum of goods, ω, from a set of goods in H+1
sectors, Ωh , determined in equilibrium. There is a freely traded homogenous numeraire
good q0 as in Chaney (2008), with one unit of labor producing w units of the good.
Ut =
µ0
q0t
H Z
Y
(
(σh −1)
σh
qh (ω)t
σh µ h
dω) (σh −1) , σh > 1∀h .
(3.1)
h=1 Ω
h
This leads to the solution each period for variety ω, with total income in the home
1
(1−σh )
dω) (σh −1) :
ω∈Ωh ph (ω)t
R
country of Yt , and the CES price index Pht = (
qh (ω)t =
h
µh Yt ph (ω)−σ
t
.
1−σh
Pht
(3.2)
Firms maximize profits each period after paying a sunk fixed cost to receive a productivity draw (output per unit of labor, ϕ) and begin producing for the home market,
and then choose whether to pay a sunk entry cost to enter the foreign market. Profits
per period for an existing firm from sales at home are11 are thus
Πh (ω)t = qh (ω)t ph (ω)t −
qh (ω)t wt
− fht wt .
ϕh (ω)
(3.3)
Where p is price, q is output sold at home, q ∗ and p∗ denote quantities and prices of
goods produced at home and sold abroad, w is the wage, τ is an iceberg trade cost, f is the
per-period overhead cost and ϕh (ω) is the output per unit of labor, supplied inelastically
by households. Firms have an exogenous probability of death δ, yet otherwise will always
choose to stay in a market they have previously entered, as expected profits are strictly
positive going forward. Maximizing profits, firms choose prices marked up over marginal
cost
ph (ω)t =
σh
wt
,
σh − 1 ϕh (ω)
ph (ω)∗t =
σh
wt τt
.
σh − 1 ϕh (ω)
(3.4)
A home firm which has previously paid to receive a productivity draw will pay a sunk
fixed cost to export, f x , if it is less than the expected discounted present value of future
profits.12
F oreign Entry : Et Π(ω)∗P V,t = Et
∞
X
x
(1 − δ)s Π(ω)∗t+s − fht
wt ≥ 0.
(3.5)
s=0
q (ω)∗ w τ
t t t
And similarly for exports abroad:
Πh (ω)∗t = qh (ω)∗t ph (ω)∗t − h ϕh (ω)
.
12
And similarly, firms will pay a fixed cost to receive a productivity draw and enter the domestic market
if the
expected profits, home and abroad, are greater than the fixed cost of entry: Et Π(ω)tot,P V,t =
P∞
Et [ s=0 (1 − δ)s Π(ω)t+s + Π(ω)∗P V,t ] − fe,ht wt ≥ 0.
11
11
The baseline empirical approach in the next section will be to use relative price indices
to explain the behavior of sectoral manufacturing employment. Thus, we can write
sectoral labor demand as:
Z
Lht =
ω∈Ω
|
qh (ω)t
+
ϕh (ω)t
{z
}
Home Production
ω∈Ω∗
∞
X
qh∗ (ω)t
e
e
x x
e
+ Mht
(fht
+ fht
pht ) +
Mh,t−s
fht ph,t,−s (3.6)
|
{z
}
ϕh (ω)t
s=0
|
{z
Z
Entry
|
}
Export Production
{z
Overhead
}
e is the mass of potential entrants at time t, M
Where Mht
ht is the total mass of all
previous entrants, pxht = 1 − G(ϕ̄x ) is the share of new firms in sector h with productivity
greater than the cutoff productivity for exporting, ϕ̄x , and ph,t−s = 1 − G(ϕ̄f,t,−s ) is
the share of continuing firms with productivity greater than the maximum cutoff for
continuing to produce for the home market, ϕ̄f,t,−s in between years t-s and t. The mass
of entrants in Chaney (2008) is assumed to be exogenous, and based on country factors
(proportional to output).
The cutoff productivity for entering into the export market at time t can be derived
from equation (3.5), assuming that firms know the productivity distribution when they
decide to invest to receive a productivity draw, and then have perfect foresight of market
conditions for the upcoming period when they decide to invest. However, firms make
their investment decisions using rules-of-thumb, taking the form of simple expectations
about the future which they believe will be like today, conditioned on not receiving a
“death” draw with probability δ. This assumption implies that the cutoff productivity
for exporting is
∗(1−σ)
ϕ̄xt =
Pht
wtσh
x
λ0 δfh,t
µh Yt∗
!σ
1
h −1
τt ,
(3.7)
σ
where λ0 =
σhh
.
(σh −1)σh −1
When labor costs, trade costs, or the sunk fixed costs of exporting rise, or the foreign
market either becomes more competitive or has a fall in demand in sector h, the cutoff
productivity for exporting will be higher which means that fewer firms will enter.
Additionally, existing firms will exit and stop producing if revenue fails to cover perperiod fixed costs. The cutoff productivity for staying in business for purely domestic
12
firms is given by:13
(1−σ)
ϕ̄f t =
Pht
wtσh
λ0 fht
µh Yt
!σ
1
h −1
.
(3.8)
This equation tells us that when labor costs or fixed costs rise, or when the domestic
market becomes more competitive or domestic demand in sector h shrinks, fewer firms
will be around to employ labor in overhead activities. To the extent that it is the case that
more productive firms export (as it is in this model), then relative price appreciations,
denoted by a rise in wages, or a rise in domestic vs. foreign GDP, would imply that
import-competing industries might be more adversely affected than relatively exportintensive industries along the extensive margin, since industries with many firms that
do not export may have a more difficult time covering the fixed overhead costs.
The first term in the sectoral labor demand equation (3.6) is the total labor requirement for home production. Plugging in the solutions from above and integrating assuming Pareto-distributed productivity with parameter γh , the first term becomes
(σh −γh )
−σh
e
λ1 ϕ̄mh,t,−s
s=0 µh,t Yt Mh,t−s wt
P∞
(1−σh )
P∞
e
s
s=0 ρ Mh,t−s wt
λ2 (ϕ̄mh,t−s )σh −γh +
P∞
∗e
∗ ∗ (1−σh ) λ (ϕ̄∗
s
σ −γ
2 mxh,t−s ) h h
s=0 ρ Mh,t−s (wt τht )
(3.9)
Where λ0 and λ1 are parameters14 , ρ = 1 − δ for brevity, ϕ̄mh,t,−s is the maximum cutoff
productivity to remain in the market for a firm that entered s periods previously in the
intervening years, and variables with an asterisk denote foreign variables. Thus ϕ̄∗mxh,t,−s
is the maximum cutoff productivity for a foreign firm which entered s periods previously
to export and remain producing during the intervening years, and variables with an
asterisk denote foreign variables. The denomenator of this equation is the solution
1−σh
to Pht
. Thus, along the intensive margin, labor demand for domestic production
depends positively on domestic sectoral demand (µht Yt ), negatively on domestic wages,
∗ . The extensive margin operates via current
and positively on importing trade costs, τht
and lagged cutoff productivities, which negatively impact home sectoral labor demand.
Higher home wages, a more competitive home market, higher fixed costs or smaller
domestic demand will all potentially trigger firm exits (via equation 3.8), which will
not necessarily be reversed immediately when these variables return to previous levels.
The sole discordant note is that, due to the CES preferences, which serve as a modeling
13
The
constraint
+
14
for
µ∗
Y ∗τ
µh Yt
h t t
(1−σ)
∗(1−σ)
P
P
ht
ht
(σh /(σh −1)−σh
λ0 =
γh −(σh −1)
staying
σ−1−1
h
in
σ
(λ0 wt h fht )
and λ1 =
business
1
σh −1
for
.
1
γh −(σh −1)
13
firms
which
also
export
is
ϕ̄f xt
=
.
convenience rather than as a statement about the way the world operates, growing
productivity in a sector will not imply decreased labor demand as both intuition and
data would suggest.
The second term on the right-hand side of equation (3.6) is analagous, as labor devoted to production for exports will be a positive function of foreign demand along the
intensive margin, and a negative function of home wages and trade costs for exporting.
Additionally, there can be movements along the extensive margin, which will depend on
the cutoff productivity for existing firms, equation (3.8). If wages, fixed overhead costs
(fht ), iceberg trade costs, or more foreign firms enter, the cutoff productivity for making
a profit will rise, and some existing firms will be forced out of the market.
∗(−σh ) 1−σh
∗
∗
∗e
h −γh
λ1 ϕ̄σmh,t,−s
τt
s=0 µh,t Yt Mh,t−s wt
.
P∞ s e
P∞ s ∗e
∗(1−σh )
∗(σh −γh )
(1−σh ) λ ϕ̄σh −γh +
λ2 ϕ̄mh,t−s
2 mh,t−s
s=0 ρ Mh,t−s (wt τht )
s=0 ρ Mh,t−s wt
P∞
(3.10)
While there is no explicit “exchange rate” in this model, one could proxy it in several
ways. One is to stipulate that both wages and output are denomenated in local dollars,
and to then treat an exchange rate appreciation as local wages and output rising relative
to foreign. A second approach, used by Eichengreen et. al. (2012), is to proxy exchange
rate movements using the iceberg trade costs. Either way, the implications are similar.
3.2 Implications
Proposition 1: Trade is a Function of History
To simplify matters, the fixed overhead costs will now be set to 0. Total exports in
industry h at time t are the sum of exports of each cohort of past entrants, where I
borrow Chaney’s notation and set the mass of entrants in industry h at time t equal to
αht Yt .
Xht =
∞
X
s
(1 − δ) αh Yt−s
Z ∞
ϕ̄t−s
s=0
xh,t (ϕ)µ(ϕ)dϕ
(3.11)
Substituting in the solutions for x = pq, plugging in the pricing rules, assuming Pareto
distributed productivity (G(ϕ) = 1 − ϕ−γh ) and integrating, I arrive at:
Xht =
t
µ∗h Yt∗ (wt τt )1−σh X
λ
(1
3
∗(1−σh )
Pt
s=0
 −γh +σh −1
∗(1−σ) σh
Ph,t−s wt−s
σh −1 
s
x

λ0 δfh,t−s
τt−s
δ) (αh Yt−s )
∗
µh,t−s Yt−s

−
σh −1
(3.12)
1−σ
Where λ3 =
σh h
γh
γh −σh +1 (σh −1)1−σh ,
and where
(3.10).
14
Pt1−σ
is the denomenator of equation
The key underlying insight of this equation is that trade today depends on the past
history of trade costs, such as from both entry and iceberg trade costs, in addition to
market sizes and contemporaneous variables. Even with the simplifying assumptions,
this equation is still fairly complex, so for purposes of expository clarity, I have summarized the sign of the impact of key variables on exports from home to abroad (foreign
variables denoted by an *) at time t
∗
x
x
Xt = f ( Yt , Yt−s , Yt∗ , Yt−s
, wt , wt−s , τt , τt−s , fht
, fh,t−s
), s > 0.
(3.13)
|{z} | {z } |{z} | {z } |{z} | {z } |{z} |{z} |{z} | {z }
+
+
+
+
-
-
-
-
-
-
The main idea that the empirical section of this paper tests is that if trade costs
rise from a shock, then the negative impact of this shock will decay over time. Note
that if we were in a one-period world, then, as in Chaney (2008), the elasticity of
substitution will not magnify the impact of iceberg trade costs, but that with multiple
periods of firm entry, this result no longer follows. How general is this dynamic gravity
formulation? In the Appendix, I prove a similar transition dynamics arise when moving
from autarky to free trade for assumptions similar to key models in the new trade theory
cannon, including Krugman (1980) and Melitz (2003). Recent related research includes
Burstein and Melitz (2011), who provide impulse response functions for shocks to trade
costs, and Bergin and Lin (2012), who focus on the dynamic impact of future shocks,
and Alessandria and Choi (2012), who find that sunk exporting costs are substantially
larger than the costs of remaining in a market. The large aforementioned literature on
hysteresis in the 1980s also carried the same core insight, that trade shocks can have
lagged effects, as equation (3.12). The contribution here is that this is the first paper
which shows that the logic of sunk costs naturally leads to a “dynamic gravity” equation
which can be derived explicitly, even with heterogenous firms.
Empirically, incumbent firms dominate most sectors in terms of market share, which
means that the current trade relationship could be determined, in part, by historical
factors as emphasized by Campbell (2010), Eichengreen and Irwin (1998), and Head,
Mayer and Ries (2010).
Corollary to Proposition 1: Real Wage a Function of Historical Market Access
A key insight from New Trade Theory is that the real wage is a function of market
access. Krugman (1992) argues that new trade theory can help explain higher wages
in the northern manufacturing belt of the US, and Meissner and Liu (2012) show that
market access can help explain high living standards in northwest Europe in the early
20th century. An important corollary is that sunk costs imply that the real wage is also
15
a function of historical market access. This follows from the dynamic gravity equation,
as utility is increasing in the number of varieties and the extensive margin increases over
time after a decline in trade costs. Figure 29 in the Appendix shows a chloropleth map
of per capita income by county, which can be compared to the distribution of importcompeting manufacturing in Figure 30. It is immediately obvious that both are highly
correlated with access to sea-navigable waterways – and that the US north was still
much richer than the south in 1979. I posit that this owes more to the past history of
trade costs than it does to low shipping costs on Lake Erie today.
Proposition 2: Government Spending Implies Dynamic Crowding Out
Matching the US experience of the 1980s and 2000s, the basic logic of sunk fixed costs
implies that government spending can cause dynamic crowding out. On page 77 in the
Mathematical Appendix, I show that government spending can cause dynamic crowding
out for the autarkic case, and derive an expression for the transition dynamics whereby
it takes private agents time to adjust to a cut in government spending. The extension
for the symmetric two-country case is trivial.
4 Estimating the Impact of Exchange Rate Movements on US
Manufacturing
4.1 Research Design
In the US, there has been a strong correlation between real exchange rate movements,
trade deficits, and manufacturing employment. In general, exchange rates cannot be
treated exogenously. However, the most obvious form of endogeneity, that a larger, more
productive tradable sector should lead to a currency appreciation, and that large job
losses and market share declines should lead to currency depreciations, should bias our
results against finding that exchange rate appreciations reduce trade balances and reduce
manufacturing employment. Additionally, trade flows tend to respond to exchange rate
movements with a one period lag. Thus while the dollar reached its peak in 1985,
the trade deficit and manufacturing employment had their worst year in 1986. An
employment collapse in 1986 would not be expected to cause a currency appreciation,
much less one year earlier.
Besides endogeneity, there could also be a third factor causing both exchange rates
and manufacturing employment to move. The most obvious candidate is high real interest rates, which could cause currency appreciation and also have a direct impact on
16
manufacturing. However, during the 2000-2003 employment collapse, real interest rates
were very low. During the collapse in 1985-1986, real interest rates were relatively high,
but they had actually fallen substantially from 1982. In Figure 4, we saw that the large
decline in structural manufacturing employment came in the 1985-1986 period, while
structural employment was fairly steady in 1982. Additionally, real interest rates only
continued to decline slowly from 1986-1989, while manufacturing employment stabilized.
Later, I will examine this question in more detail using sectoral data, which also strongly
suggests that high real interest rates are not what caused the decline in manufacturing
employment in the mid-1980s, although they do appear to have impacted export sectors
disproportionately.
Nevertheless, it remains important to understand what drove the dollar appreciation
in both periods. In general, it is difficult to explain short-term movements in real
exchange rates based on fundamentals (see Meese and Rogoff, 1983). If all exchange rate
movements were a random walk, this would greatly aid research design. However, in the
1980s, and in the early 2000s, part of the dollar appreciation was almost certainly caused
by large structural fiscal deficits. In both periods, the fiscal deficits were caused by the
relatively random outcomes of the presidential elections in 1980 and 2000. Additionally,
the dollar’s rise in the late 1990s was caused in part by the Asian Financial Crisis, which
crippled the industrial sectors of several East Asian nations at the same time that their
currencies depreciated. Financial crises would lead to an endogeneity problem as we
would not necessarily expect the US to have had a larger trade deficit with Korea in
1998 than in 1996. The Asian Financial Crisis ushered in the “Bretton Woods II” period
and the Great Reserve Accumulation by developing nations which became large after
the dollar began to weaken in 2002. This system of managed exchange rates can also
be argued to be a relatively exogenous source of dollar strength after 2002, particularly
using the WARP index as many countries with managed exchange rates such as China
saw their share of US imports rise.
As others have written (e.g., Gourinchas, 1999), there is a very good case to be made
that the 1980s dollar appreciation was a random event, which makes it an unusually
good experiment given the sharp swings in the dollar. Larger appreciations are likely
to have a higher signal-to-noise ratio and more likely to have a measurable impact on
trade and employment than are small year-to-year fluctuations, which are less plausibly a
key determinant of manufacturing employment. While the factors causing the exchange
rate appreciation in both periods alone gives good reason to believe that exchange rate
movements are the cause of the aggregate trade balances and decline in manufacturing,
in fact both can be disaggregated, providing additional tests for the theory that large
17
movements in relative prices can have persistent effects.
Disaggregated data allow the possibility for a difference-in-difference setup, as one can
test whether industries with higher initial levels of openness tend to be more affected by
currency appreciations. In addition, one has more degrees of freedom to test competing
hypotheses, such as whether more open sectors are also more sensitive to real interest
rate fluctuations (they are not). Thus, if the results are spurious, it would have to be
because some factor other than real interest rates is causing both the dollar appreciation
and a differential impact on employment of more-open sectors. Disaggregated data also
allow one to use quasi-experiments from policy-led changes in real exchange rates, such
as the case of Japan.
4.2 Industry-Level Hysteresis
The basic approach to identification in this paper is to use a difference-in-difference
methodology to identify the impact of large exchange rate movements across manufacturing sectors with differing initial exposure to international trade over the period
1972-2009. In periods such as 1972-1979, when the dollar fell, the entire distribution
of log changes in US exports by sector and destination is significantly larger than the
distribution of changes in log imports (Figure 6). When the dollar spiked in the mid1980s, the distribution of log changes in imports is then shifted far to the right of the
distribution of exports, with the median log change in imports close to one vs. slightly
greater than zero for exports. This corresponds to a 72% increase in imports relative
to exports, which implies that sectors which began the period with proportionally more
import penetration in 1979 experienced a larger magnitude shock. The same pattern
holds up over the period of dollar weakness from 1986-1996, and dollar strength from
1996-2005.
The four panels of Figure 7 confirm that during periods of sharp dollar appreciations,
sectors with higher initial trade shares were disproportionately harmed. During periods
of dollar depreciation, there was no meaningful difference in performance, but the previous period’s losses appeared to be locked in. Openness predicts employment declines
over the period 1979-1986, but there is no relationship between openness in 1979 and
employment growth over the period 1986-1996 (Appendix Figure 31). These periods
were all chosen at similar points in the business cycle, and in particular, not in recession
years, to control for the fact that some manufacturing sectors are much more cyclical
than others. Additionally, these results are robust to controlling for domestic sectoral
demand growth, productivity growth, lagged capital-per-worker ratios, changes in tariffs, and sectoral changes in the cost of insurance and freight. They are also robust to
18
.6
.4
.6
.4
.2
Density
Density
0
.2
0
−3
−2
−1
0
1
Log Change Imports
2
3
−3
−2
Log Change Exports
−1
0
1
Log Change Imports
3
Density
0
0
.1
.2
.2
Density
.3
.4
.4
.6
(b) Dollar Appreciation: 1979-1986
.5
(a) Dollar Depreciation: 1972-1979
2
Log Change Exports
−3
−2
−1
Log Change Imports
0
1
2
3
−2.5
Log Change Exports
−2
−1.5
−1
−.5
Log Change Imports
(c) Dollar Depreciation: 1986-1996
0
.5
1
1.5
2
(d) Dollar Appreciation: 1996-2005
Figure 6: Distribution of Changes in Trade, by Sector and Country
19
2.5
Log Change Exports
1
.5
1
0
.5
−.5
0
−1.5
−1
−.5
−1
0
.2
.4
.6
0
.2
Openness, 1972
Log Change in Employment, 1972−1979
Slope not Significant
.4
Openness, 1979
Log Change in Employment, 1979−1986
.8
Slope Significant
(b) Dollar Appreciation: 1979-1986
−1.5
−2
−1
−1.5
−.5
−1
0
−.5
.5
0
1
.5
(a) Dollar Depreciation: 1972-1979
.6
0
.2
.4
Openness, 1986
Log Change in Employment, 1986−1996
.6
.8
0
Slope not Significant
.2
.4
.6
Openness, 1996
Log Change in Employment, 1996−2005
(c) Dollar Depreciation: 1986-1996
.8
1
Slope Significant
(d) Dollar Appreciation: 1996-2005
Figure 7: The Evolution of Manufacturing Employment
running a quantile regression to reduce the significance of outliers (see Tables 20 and 23
in the Appendix).
The magnitude of the slopes are large and economically significant. The slope in the
1979-1986 period is -.74 (with a standard error clustered on the 2-digit SIC level of .26),
which implies that for every 10% increase in trade an industry had in 1979, it approximately lost an additional 7.4% of manufacturing employment during that period, when
the Federal Reserve’s broad trade-weighted dollar index appreciated 45.4%. In Appendix
Figure 31, I show that import exposure in 1979 is uncorrelated with employment growth
from 1986-1996, suggesting that there was no rebound effect after the dollar depreciated. The slope in the 1996-2005 period was similarly large, at -.75, with a standard
error clustered at the 2-digit SIC level of just .17. However, since the mean amount
of trade was much larger in 1996, the later period accounted for a much larger overall
decline in employment.
20
Since 2005, the dollar has continued to fall, and as of 2011 is now close to its long-run
average. In 2005-2007, when the dollar was still fairly elevated (as measured by weighted
average relative prices), there was again a statistically significant relationship between
openness in 2005 and employment losses, even though overall manufacturing employment
was roughly flat. From 2007-2008, when the dollar fell, the relationship disappeared.
From 2009-2011, as the dollar declined further, there was also no statistically significant
relationship (using NAICs data through 2011). In these later years, of course, there was
much else going on beside exchange rate movements, but the fact that there appears
to have been no rebound in these sectors in 2007-8 or in 2009-2011 suggests that the
employment losses in the high-dollar years had been locked in, just as they had been
after the dollar appreciated in the 1980s.
In Figure 8, the evolution of employment indices by categories of tradability in 1972
is plotted. The employment index for each sector is given a base year value of 100 in
1979, and then the changes in the employment indices not due to changes in demand or
shipments, or by general movements in all sectors by year, are plotted over time with
error bounds. Comparing the top 25% of sectors by openness vs. the bottom 50%, we
see that there was no statistically significant difference in employment growth during
the 1970s, but that when the dollar appreciated in the 1980s, the more open sectors
lost roughly 10% of their employment relative to other sectors, and that after the dollar
fell in the late 1980s, this differential impact seems to have disappeared very modestly.
The appreciation in the late 1990s and early 2000s seems to suggest a similar story –
steep losses in the early 2000s which were then not reversed. While the magnitudes
appear smaller here, this is in part a function of the fact that both of these categories
of industries contain a large variation in their respective degrees of openness.
Previous research is mixed on the question of the impact of exchange rate movements
on trade; Klein, Schuh, and Triest (2002) provide an overview of an array of qualitiy
studies to that point. The average study in the sample spans less than 15 years, 40
manufacturing industries, and one country; i.e., Grossman (1986) finds a negative impact of the dollar’s appreciation on blast furnaces over the period 1976 to 1983. Among
these studies are three papers by Branson and Love (1986, 1987, and 1988) who used
2-digit SIC data to estimate that the real exchange rate appreciation of the 1980s destroyed 1 million manufacturing jobs. Klein, Schuh, and Triest (2003) study 442 4-digit
manufacturing industries over the 1974-1993 period, and find that real exchange rate
appreciations increase job destruction and slightly reduce job creation, while smaller appreciations merely increase both. Campa and Goldberg (2001) find that exchange rate
movements affect low-markup industries but not sectors with higher markups.
21
110
105
100
95
90
85
1972
1976
1980
1984
Most Open: Top 25%
1988
1992
1996
Least Open: Bottom 50%
80
90
100
110
120
Figure 8: Employment Growth by Degree of Tradability in 1972 (SIC)
1989
1992
1995
1998
2001
Most Open: Top 25%
2004
2007
2010
Least Open: Bottom 50%
Figure 9: Employment Growth by Degree of Tradability in 1989 (NAICs)
22
Much work has been done for RER movements in Europe. Berman et. al. (2012)
confirms a heterogeneous impact by markups on 10 years of French data, with highmarkup firms reacting to depreciations by increasing their markup more than volumn,
and low-markup firms vice-versa. Pozzolo and Nucci (2008) find a larger impact of RER
movements on low-markup industries using 16 years of Italian data. Moser et al. (2010)
find that real exchange rates affect job creation and job destruction in Germany, but that
due to Germany’s regulated labor market, most of the impact happens via job creation.
Belke et al. (2013) find that large exchange rate movements in Germany have hysteretic
effects.
None of these studies use weighted-average relative prices, nor am I aware of empirical
studies which study hysteresis in American manufacturing for both the 1980s and 2000s.
This is in stark contrast to the 1980s, when the apparent persistence in the US trade
deficit gave rise to the aforementioned large, mostly theoretical literature seeking to
explain what a generation of trade economists, including Dixit, Krugman and Baldwin,
clearly saw as an important phenomenon.
4.3 Data
Data on employment, shipments, value-added, wages, investment, and capital, and the
prices of shipments, materials, and energy all come from the BEA’s Annual Survey of
Manufactures via the NBER-CES manufacturing page with data at the 4-digit SIC level
from 1958-2009. An additional NAICs version of the data was also coded up from 19892011 directly from the BEA. Trade data from 1991-2011 are from Comtrade WITS where
available, and this data was augmented with trade and the cost of insurance and freight
data from Feenstra, Romalis, and Schott (2002) from 1972-2005. Sectoral tariff data
for 1974-2005 come from Schott (2008) via Feenstra, Romalis, and Schott (2002). Data
on intermediate imports are from the BEA’s Input/Output tables for the year 1997.
The classification of broad industrial sectors by markups is borrowed from Campa and
Goldberg (2001).15
The measure of the real exchange rate I use is a manufacturing trade weighted average
relative unit labor cost (WARULC) measure for manufacturing introduced in Campbell
(2013b), a companion paper. Unit labor costs have long thought to be the best measure
of competitiveness, but there are substantial problems with the ULC-based RER indices
15
The Campa-Goldberg classification of low markup industries at the 2-digit SIC level includes primary
metal products, fabricated metal products, transportation equipment, food and kindred products,
textile mill products, apparel and mill products, lumber and wood products, furniture and fixtures,
paper and allied products, petroleum and coal products, and leather and leather products.
23
20
10
15
8
Density
10
6
Density
5
4
0
2
0
0
.2
.4
.6
.8
1
0
Openness
(a) Openness
.1
.2
.3
Share of Shipments
.4
.5
(b) Imported Intermediate Inputs
Figure 10: Distributions of Key Variables by Sector, 1997
created by the IMF and the OECD (plotted for comparison in Figure 11). The WARULC
index corrects some of the core issues with the construction of the IMF index, as it
includes China, time-varying trade weights, and also deflates manufacturing value-added
using Geary-Khamis reference prices from all six benchmark years of the ICP using PWT
v8.0 methodology (see Feenstra et al. 2013). In the IMF measure, manufacturing output
is deflated using country VA deflators from a single base year, which becomes problematic
when comparing two countries’ output over long time periods. This is essentially the
same problem which afflicted older vintages of the PWT (see Feenstra et al. 2013). In
addition, the WARULC index is a simple geometric weighted average of relative unit
labor costs rather than of relative unit labor cost indices, as is the IMF and OECD
measures. I then took the log of this index so that when unit labor costs are the same at
home and abroad, the index would equal zero (subtracting one instead would not change
any of the core results). Alternatively, one could also use any of the other weightedaverage relative (WAR) exchange rates from Campbell (2013b) and attain very similar
results. Using real exchange rate indices produced by the Federal Reserve, the IMF, the
BIS or the OECD would yield significantly different results, however, since these indices
do not reflect the rising role of trade with China and other developing countries.
The basic summary statistics for the main variables used in the regressions are reported in Table’s 1 and 2, with the distribution of openness and imported intermediate
inputs for the year 1997 illustrated in Figure 10. “Openness” is defined here as the
average of the share of exports in total shipments and imports as a share of domestic
consumption (imports divided by shipments plus imports minus exports). Over the period openness increased from about 7% in 1972 to 24% in 2001 and 32% by 2009. In 1974,
24
Table 1: Data Summary for Select Years
(1)
1972-2009
0.165
(0.190)
(2)
1974
0.0862
(0.112)
(3)
1985
0.115
(0.119)
(4)
1993
0.174
(0.174)
(5)
2001
0.238
(0.241)
Imports, Millions
1456.1
(5376.1)
220.0
(734.9)
851.3
(3073.8)
1349.8
(4074.1)
2593.0
(7997.6)
Exports, Millions
912.6
(2680.0)
205.2
(577.2)
396.2
(1023.3)
973.6
(2377.6)
1549.7
(3795.3)
Shipments, Millions
7100.2
(16745.0)
2551.0
(5205.4)
5496.3
(12840.4)
7457.6
(14510.6)
9829.5
(20554.1)
Value Added, Millions
3236.5
(6239.1)
1099.3
(1687.9)
2273.4
(3429.0)
3475.4
(5497.0)
4551.3
(8030.1)
Hourly Wages, Prod. Workers
11.13
(4.883)
4.366
(1.008)
9.572
(2.782)
11.99
(3.354)
15.10
(4.145)
Shipments per Worker, (1000s)
190.7
(223.5)
61.43
(60.72)
143.1
(151.5)
197.6
(165.6)
267.6
(278.9)
Weighted-Avg. Rel. ULCs
1.149
(0.115)
1.099
(0)
1.494
(0)
1.063
(0)
1.396
(0)
Duties %
0.0503
(0.114)
0.0831
(0.0709)
0.0553
(0.0564)
0.0505
(0.108)
0.0306
(0.0420)
Ins., Freight Costs %
0.0844
(0.0954)
0.0747
(0.0668)
0.0736
(0.0750)
0.0969
(0.0471)
0.0913
(0.0494)
K/L, (1000s)
87.42
(106.3)
51.04
(56.95)
78.01
(89.27)
84.43
(90.95)
115.3
(130.7)
5-factor TFP index 1987=1.000
1.031
(0.891)
0.974
(0.214)
0.974
(0.0818)
1.018
(0.131)
1.078
(1.444)
Openness
Mean coefficients; sd in parentheses.
25
Table 2: Additional Data Summary for Select Years: Prices and Inputs
(1)
1972-2009
0.380
(0.121)
(2)
1972
0.458
(0.108)
(3)
1985
0.413
(0.109)
(4)
1993
0.373
(0.118)
(5)
2001
0.364
(0.120)
Investment/Value-Added
0.0639
(0.0464)
0.0627
(0.0450)
0.0757
(0.0714)
0.0622
(0.0654)
0.0647
(0.0426)
Energy Costs/Value-Added
0.0490
(0.0750)
0.0324
(0.0453)
0.0676
(0.120)
0.0488
(0.0856)
0.0491
(0.0689)
Materials Costs/Value-Added
1.181
(0.921)
1.113
(0.893)
1.280
(1.323)
1.119
(0.716)
1.147
(0.700)
Shipments Deflator
1.086
(0.414)
0.515
(0.839)
0.979
(0.0633)
1.160
(0.121)
1.284
(0.246)
Materials Deflator
1.045
(0.317)
0.412
(0.119)
1.001
(0.0628)
1.122
(0.0797)
1.158
(0.197)
Investment Deflator
0.985
(0.247)
0.435
(0.0653)
0.940
(0.0229)
1.136
(0.0550)
1.116
(0.126)
Energy Deflator
1.070
(0.359)
0.268
(0.0903)
1.118
(0.0521)
1.126
(0.0205)
1.386
(0.0717)
Production Worker Hours
1992.6
(125.6)
1981.7
(103.3)
1958.3
(122.1)
2035.3
(125.8)
1985.0
(132.1)
Prod. Workers/Total Emp
0.723
(0.114)
0.755
(0.105)
0.723
(0.115)
0.714
(0.119)
0.714
(0.120)
Japanese Import Penetration
0.0196
(0.0433)
0.0145
(0.0379)
0.0236
(0.0461)
0.0292
(0.0520)
0.0258
(0.0503)
Chinese Import Penetration
0.0250
(0.245)
0.0000425
(0.000189)
0.00266
(0.00858)
0.0256
(0.115)
0.0808
(0.545)
Payroll/Value-Added
Mean coefficients; sd in parentheses.
26
1.6
1.4
1.2
1
.8
.6
1970
1980
1990
WARULC
2000
2010
IMF Index
Figure 11: WARULC vs. IMF RULC Index
manufacturing trade was roughly balanced and unit labor costs for US manufacturing
were on average about 10% higher than ULCs of US trading partners, but after the
dollar’s appreciation in the mid-1980s left US relative ULCs an additional 40% higher,
manufacturing imports became twice as large as exports. The average applied tariff was
about 8.2% in 1974, and fell to just 2.4% by 2005. By contrast, the cost of insurance
and freight was about 9.6% of customs costs in 1974, and was still 9.8% in 2005. The
last two entries in Table 1, capital-per-worker and the 5-factor TFP index, were also
taken from the NBER-CES manufacturing data set. The details of their creation are
described in Bartelsman and Gray (1996).
The second data table includes data on prices, costs as a share of value-added, production workers as a share of total employment, worker hours, and import penetration
from Japan and China. The number of observations in both panels is balanced at 353
for all time periods, except for customs and the cost of insurance and freight, as this
data starts later and has a slightly smaller sample in the 1970s and 1980s.
4.4 Estimation
The estimation approach is to regress the log change in employment on lagged relative
openness, controlling for a host of variables from equation (3.6), including sectoral demand, changes in trade costs and changes in relative prices, in addition to intuitive other
controls such as productivity in terms of value added-per-production worker by sector,
27
and a host of other controls.
ln(Lht /Lh,t−1 ) = αt + β1 (ln(W ARU LC) ∗ Rel.Openness)h,t−1 + β2 ln(Dh,t /Dh,t−1 )+
(4.1)
β3 ln((V A/L)h,t /(V A/L)h,t−1 +
Pn
i=4 βi Ci,t
+ αh + νt + ht ,
∀h = 1, ..., 353, t = 1973, ..., 2009.
Where Lht is employment in sector h at time t, Dh,t is real sectoral demand, (V A/L)h,t
is real value-added per production worker and the C’s are other various controls. Each
regression also includes sectoral fixed effects αh , year fixed effects νt , standard errors
clustered at the industry level, and is weighted on initial period value-added (these latter
three elements are new to this literature). The results do not appear to be sensitive to the
choice of weights, as qualitatively similar results attain when weighting on overall average
value-added, employment, or shipments, although the key coefficient is the largest when
weighting on employment or when not weighting. Additionally, one could get very similar
results by simply using openness rather than relative openness.16
The baseline regression results are featured in Table 3. It can be seen that openness is
not generally a significant predictor of employment losses, but that the interaction term
between lagged openness and high RULC periods is consistently larger and highly significant. When controls for sectoral demand and value-added-per-worker are included, the
coefficient increases substantially, while the estimates become only slightly less precise
despite a sizeable increase in the R-squared. The results in this table include 353 sectors
with complete, balanced data, and exclude all sectors in the 2-digit SIC category publishing, which the NAICs classification system does not include in manufacturing (this
sector is added back in later for robustness). The coefficient of -.083 suggests that in
1985, when US ULCs were 50% (or 1.5, for a log value of .4) above a weighted-average
of ULCs of US trading partners, an industry with an openness twice than that of the
average industry would have lost an additional 6.5% of employment from 1985-1986
(=exp(-.083*1.4*2)-1) as compared with a completely closed industry, and 3.2% more
than an industry with average openness. Over the entire 1982-1986 period, this industry
would have lost a cumulative 23% of employment relative to a closed industry, and 11%
more than an industry with average openness.
In the third column I include additional controls, and find that the results are little
changed. First, I include a term for the interaction between WARULC and the industries
16
These additional controls, and others, are contained in the additional not-for-publication appendix.
28
Table 3: Exchange Rates, Openness, and Manufacturing Employment
(1)
ln∆ L
-0.00475
(0.00338)
(2)
ln∆ L
-0.00468
(0.00402)
(3)
ln∆ L
-0.00500
(0.00414)
(4)
ln∆ L
0.000899
(0.00396)
-0.0567***
(0.0113)
-0.0829***
(0.0137)
-0.0819***
(0.0140)
-0.0975***
(0.0153)
ln∆ VA-per-Production Worker
-0.216***
(0.0276)
-0.214***
(0.0279)
-0.231***
(0.0333)
ln ∆ Demand
0.447***
(0.0525)
0.445***
(0.0524)
0.444***
(0.0638)
Campa Low Markup*L.ln(WARULC)
0.0212
(0.0197)
0.00709
(0.0207)
Imported Intermediates*L.ln(WARULC)
0.0718
(0.156)
0.0542
(0.172)
L.K/L
0.0454*
(0.0243)
0.0600***
(0.0223)
L.(K/L)*Real Interest Rate
-0.680**
(0.281)
-0.548*
(0.307)
L.Relative Openness
L.ln(WARULC)*Rel. Openness
Change in Tariffs
-0.000920
(0.00479)
Change Ins. & Freight Costs
-0.00453
(0.00314)
L.Rel. Openness*Real Interest Rate
0.0107**
(0.00412)
353
10165
0.496
0.351
0.483
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
353
12963
0.234
0.0507
0.168
353
12963
0.522
0.320
0.491
353
12963
0.523
0.293
0.489
Standard errors clustered on 4-digit SIC industries in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01.
All regressions weighted by initial sectoral value-added, and include 4-digit SIC industry and year
fixed effects over the period 1973-2009. The dependent variable is the log change in sectoral
manufacturing employment. Tariff and CIF data span 1974-2005.
29
classified by Campa and Goldberg (2003) as being low markup. The coefficient is of
the wrong sign and not signficant. Secondly, I include an interaction term between an
industry’s share of imported intermediate goods in 1997 and the degree of overvaluation.
The coefficient is of the expected sign, as it follows that imported intermediate inputs
should become relatively cheaper as the dollar appreciates, ameliorating some of the
impact of more expensive domestic labor and inputs. However, the coefficient is smaller
than the standard error. Additionally, I control for the lagged capital-per-worker ratio
and capital-per-worker interacted with the real interest rate, defined as the rate on 30year mortgages minus the core CPI. This coefficient is fairly large and significant at 95%,
as industries which are more capital intense do comparatively worse when real interest
rates are higher, but the significance disappears in some specifications in later tables.
In column (4) I introduce controls for changes in tariffs and costs of insurance and
freight charges at the sectoral level (with data available from 1974-2005), and lagged
openness interacted with the real interest rate. This latter control was included to
ensure that the core result is not caused by more tradable sectors happening to be more
sensitive to movements in real interest rates. In fact, more tradable sectors appear to
be less capital intensive and not significantly more sensitive to interest rate fluctuations.
The signs on all three of these variables are as expected, but none of the coefficients are
significant. By contrast, the magnitude of the variable of interest, the interaction term
between WARULC and openness, increases while also becoming more precise.
Since previous research in this vein used divisia-based real exchange rate indices, in
which the levels have no economic meaning, log changes in the indices were typically
used. A puzzle that emerged from this literature was why dollar appreciations seemed to
adversely affected manufacturing while dollar depreciations did not seem to help. The
levels of the WARULC index, by contrast, tell us how much higher domestic unit labor
costs are relative to a trade-weighted average of foreign labor costs. Using the level of
this index instead of the log change allows one to solve the earlier puzzle and directly test
for hysteresis, since more open industries suffer employment declines when this index is
higher than one, and yet experiences no bounceback when this index is near one. This
implies that relative price shocks have persistent effects.
Figure 12 includes the results from running the regression in equation 4.1 by year,
substituting relative openness for the interaction between relative openness and the
WARULC index, with controls for demand, productivity, and capital intensity lagged
one period. It is plotted vs. the WARULC index and the real interest rate (divided
by 4 for comparability). It can be seen that Klein et al. results appear to hold out
sample for the strong dollar period in the late 1990s and early 2000s, as the impact
30
appears similar. The WARULC index also does a suprisingly good job of predicting the
decline in relatively more open manufacturing sectors in 2002 (much better than Divisia,
WARP or BSWARP indices). However, it also does less well in the 2005-7 period, which
is one period when the WARULC index falls more quickly than does the PPP based
exchange rates. In addition, although capital-intensity was included as a control, high
real exchange rates do appear to play a partial role in the decline of more open industries
in the early 1980s. As I will discuss later, export-intense sectors in the US also tend to
be more sensitive to movements in real interest rates, while import-competing sectors
appear less sensitive. An alternative version of this graph using import penetration and
1.5
1
−.04
1.1
−.02
0
1.2
1.3
WARULC
.02
1.4
.04
the import-weighted WARULC is featured in Appendix Figure 28.
1975
1980
1985
1990
Coeff. on Rel. Openness
2 s.e. Error Bound
WARULC
1995
2000
2005
2010
2 s.e. Error Bound
Real Interest Rate/4
Figure 12: Impact of Relative Openness by Year
In Table 4 I include various other controls found in Klein et al. (2003), including lagged
log change in the prices of shipments, materials, energy, and investment. Additionally I
include interaction terms between lagged intermediate input price changes (for materials,
energy, and investment) and the share of these inputs over shipments. Of the price
variables, the only one that appears to be consistently significant is the lagged log change
in the price of shipments, which ostensibly does predict subsequent employment growth,
but leaves the overall R-squared little changed. Lagged changes in hourly wages, lagged
changes in materials costs, energy costs, and investment costs alone seem to have little
31
predictive power.
The interaction terms between energy costs (including oil) and the ratio of energy
costs to total shipments is surprisingly not consistently significant. The interaction term
between lagged increases in the price of investment and the investment-to-shipments
ratio is, by contrast, large and significant at 95%. I find that the interaction term
between overvaluation and relative openness is robust to all of these controls, and also
robust to including all controls from Table 3 and Table 4 combined.
4.5 Relative Difference-in-Difference
A second empirical approach is to use international data to create an additional dimension to the difference-in-difference estimation above, and ask whether more open
manufacturing sectors in the US tend to lose more jobs when the currency appreciates
relative to the same sectors in other large manufacturing countries. Figure 13 displays
the idea graphically. From 1979-1986 and from 1995-2002, the 3-digit ISIC sectors which
were more open tended to experience larger declines in employment in the US, but there
was no such relation in other major economies.17
Thus, we now estimate:
ln(
LU S,h,t
LG5,h,t
) − ln(
) = αt + β1 ((W ARU LC − 1) ∗ Openness)h,t−1 +
LU S,h,t−1 )
LG5,h,t−1
D
D
(S/L)
(4.2)
(S/L)
S,h,t
G5,h,t
S,h,t
G5,h,t
β2 (ln( DUUS,h,t−1
) − ln( DG5,h,t−1
)) + β3 (ln( (S/L)UUS,h,t−1
) − (ln( (S/L)G5,h,t−1
)) + αh + νt + ht ,
∀h = 1, ..., 29, t = 1978, ...1995, 1998, ..., 2003,
G5 = (Canada, F rance, Germany, Italy, U K).
Where the dependent variable is now the log change in sectoral US employment minus
the average log change in employment in Canada, France, Germany, Italy and the UK.
The data are 3-digit ISIC Rev. 2 data from UNIDOs, which does not report data for
the US for 1996. The first column in Table 5 runs the difference-in-difference regression
using just US data as in previous tables. This demonstrates that the main results are not
due to any specific factors related to the construction of the SIC dataset used previously,
and that the results are robust to lumping sectors into broader categories. In the second
column, I estimate the relative difference-in-difference regression in equation 4.2. Here,
the magnitude of the results even increases, although the estimate also becomes less
17
In the additional appendix, I also show that there is no correlation between openness and employment
for years when the dollar was weak.
32
Table 4: Additional Controls: Prices
(1)
ln∆ L
-0.00303
(0.00399)
(2)
ln∆ L
-0.00322
(0.00398)
(3)
ln∆ L
-0.00300
(0.00397)
(4)
ln∆ L
-0.00172
(0.00469)
L.ln(WARULC)*Rel. Openness
-0.0849***
(0.0136)
-0.0845***
(0.0135)
-0.0842***
(0.0136)
-0.0917***
(0.0143)
ln∆ VA-per-Production Worker
-0.207***
(0.0246)
-0.206***
(0.0247)
-0.205***
(0.0246)
ln ∆ Demand
0.440***
(0.0461)
0.439***
(0.0460)
0.438***
(0.0460)
0.566***
(0.0653)
L.(K/L)
0.0483**
(0.0230)
0.0496**
(0.0223)
0.0380*
(0.0227)
0.0293
(0.0213)
L.(K/L)*L.Real Interest Rate
-0.707**
(0.282)
-0.405
(0.300)
-0.704**
(0.327)
-0.531**
(0.249)
L.ln ∆ Hourly Wages
0.0216
(0.0180)
0.0201
(0.0180)
0.0229
(0.0178)
-0.00311
(0.0180)
L.ln ∆ Price of Shipments
0.0273***
(0.0105)
0.0306***
(0.0103)
0.0235***
(0.00898)
L.ln ∆ Price of Materials
0.0106
(0.0117)
0.117***
(0.0318)
0.0518*
(0.0284)
L.ln ∆ PM*(M/S)
-0.159***
(0.0484)
-0.0610
(0.0413)
L.ln ∆ PE*(E/S)
-0.395**
(0.156)
-0.240
(0.164)
L.ln ∆ PI*(I/S)
-0.649
(0.401)
-0.755**
(0.376)
14864
0.476
-0.587***
(0.0538)
14864
0.615
L.Relative Openness
ln ∆ Shipments-per-Worker
Observation
Overall R-squared
14864
0.473
14864
0.473
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. All regressions are
weighted by initial sectoral value-added, and include 4-digit SIC industry and year fixed effects
over the period 1973-2009. The dependent variable is the log change in sectoral manufacturing
employment. Lagged changes in the prices of investment and energy have been omitted to save space.
33
Table 5: Difference-in-Difference Relative to Other Major Economies
(1)
ln ∆ L
0.0622**
(0.0288)
(2)
ln ∆L (Relative)
0.0918**
(0.0456)
L.Openness*(WARULC-1)
-0.486***
(0.0589)
-0.633***
(0.102)
ln ∆(Y /L)
-0.899***
(0.0406)
ln ∆ Demand
0.894***
(0.0411)
L.Openness
ln ∆(Y /L) (Relative)
-0.519***
(0.0641)
ln ∆ Demand (Relative)
0.583***
(0.0579)
606
Observations
606
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. Both quantile
regressions include year and 3-digit ISIC industry fixed effects for 29 ISIC Rev. 2 sectors over
the period 1977-1995, 1998-2003. The dependent variable in column 1 is the log change in
sectoral manufacturing employment. In column 2, the dependent variable is the log change
in manufacturing employment relative to the log change in employment in the same sector
in other major economies.
34
.5
.2
0
0
−.2
−.5
−.4
−.6
−1
−.8
0
.05
.1
Openness, L7
.15
Log Change Employment, 1979−1986
.2
0
Fitted values
.1
.2
.3
Openness, L7
.4
Log Change Employment, 1979−1986
Fitted values
(b) Other Major Economies, 1979-86
−1
−1.5
−1
−.5
−.5
0
0
.5
.5
(a) US, 1979-86
.5
0
.1
.2
Openness, L7
Log Change Employment, 1995−2002
.3
.4
0
Fitted values
.2
.4
Openness, L7
Log Change Employment, 1986−1993
(c) US, 1995-2002
.6
.8
Fitted values
(d) Other Major Economies, 1995-2002
Figure 13: Employment Growth vs. Lagged Openness, 3 digit ISIC, 1979-1986
precise. The standard errors are clustered at the industry level and year and industry
dummies are included in a quantile regression minimizing the sum of absolute deviations.
Median quantile regressions are less sensitive to the inclusion of outliers and non-normal
errors. However, the magnitude of the interaction variable in this case actually increases
when estimating with OLS.
4.6 Defense Spending, the Budget, and the Crowding Out of the Tradables
Sector
In Proposition 2.1 in the Appendix section 7.3, I show that government spending can
lead to dynamic crowding out of more tradables sectors. Although the mechanism is not
spelled in out in this simple model, a larger government deficit can affect the tradables
sector in at least three ways. First, even in a closed economy setting, higher government
35
spending could induce more resources to be allocated to service sectors. Secondly, larger
government deficits can lead to higher real interest rates, which appreciate the currency.
Thirdly, a larger supply of US Treasuries may induce foreign purchases, given that globally there is a limited supply of very safe, positive-yielding assets, whose value appreciates
during recessions and financial crisis. Additionally, higher interest rates could have a
direct affect on more-tradable sectors if these sectors are also more sensitive to interest
rate movements.
In this section, I estimate reduced-form regressions using changes in defense spending
and the budget deficit ex-automatic stabilizers to predict differential changes in employment in more tradable sectors. It can be seen in Figure 14(a) that defense spending
as a share of GDP increased dramatically after the 1980 election, and increased again
after the US election in 2000. In Figure 14(b) it can be seen that changes in the US
budget deficit appear to be highly correlated with changes in weighted-average relative
1970
1980
1990
Defense Spending, Share of GDP
2000
−2
−1
0
1
2
Change in Budget (ex−Automatic Stabilizers)
1.5
1.4
WARULC
1.2
1.3
1.1
1
−8
.02
Defense Spending, Share of GDP
.04
.06
.08
−6
−4
−2
0
Budget Deficit (ex−Automatic Stabilizers)
2
.1
unit labor costs (although the correlation with WARP is slightly stronger).
2010
1970
Budget Deficit
1980
WARULC
(a) Defense Spending at the Budget Deficit
1990
2000
2010
Change in Federal Budget
(b) Structural Budget Deficit vs. WARULC
Figure 14: Defense Spending, the Structural Budget Deficit, and RULCs
In Table 6 Column (1), I regress lagged relative openness interacted with log changes
in defense spending over GDP (divided by ten to normalize the coefficient). Once again,
I get a negative, statistically significant coefficient, which implies that in 1985, when
defense spending as a share of GDP increased by 10%, a sector with a relative openness
of twice the average would have experienced a decline in employment by 2 percent
relative to a closed sector, and a sector with four times the average would have lost four
percent. This effect is not driven by GDP as the denomenator, as if we deflate defense
spending with total manufacturing shipments instead, as in column (2), the results only
36
Table 6: Government Spending and Crowding Out
(1)
ln∆ L
-0.0110***
(0.00396)
(2)
ln∆ L
-0.00964***
(0.00367)
(3)
ln∆ L
-0.0124***
(0.00427)
(4)
ln∆ L
-0.00795***
(0.00201)
ln∆ VA-per-Production Worker
-0.209***
(0.0248)
-0.210***
(0.0245)
-0.208***
(0.0247)
-0.301***
(0.0158)
ln ∆ Demand
0.451***
(0.0473)
0.452***
(0.0462)
0.451***
(0.0464)
0.598***
(0.0159)
L.(K/L)
-0.00718
(0.0224)
-0.00893
(0.0225)
-0.0161
(0.0233)
-0.00954
(0.0118)
L.(K/L)*L.Real Interest Rate
-0.543*
(0.319)
-0.584*
(0.314)
-0.454
(0.318)
-0.0343
(0.331)
L.ln ∆ Price of Shipments
0.0332***
(0.0109)
0.0323***
(0.0107)
0.0355***
(0.0108)
0.0210**
(0.00899)
L.ln ∆ Price of Materials
0.126***
(0.0336)
0.125***
(0.0326)
0.121***
(0.0329)
0.0765***
(0.0284)
L.ln ∆ PI*(I/S)
-0.941**
(0.473)
-0.904*
(0.470)
-1.063**
(0.437)
-0.499
(0.461)
L.ln ∆ PE*(E/S)
-0.415**
(0.170)
-0.422**
(0.170)
-0.408**
(0.164)
-0.288*
(0.152)
L.Rel.Open* ln∆ (Defense/GDP)
-1.179***
(0.147)
0.634***
(0.230)
14864
0.462
0.369***
(0.0879)
14864
L.Relative Openness
L.Rel.Open* ln∆ (Defense/Shipments)
-1.170***
(0.174)
Rel.Open*∆ Struct. Budget Balance
Observation
Overall R-squared
14864
0.467
14864
0.469
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. The first three regressions
are weighted by initial sectoral value-added, and include 4-digit SIC industry and year fixed effects over
the period 1973-2009. The fourth column minimizes the sum of absolute deviations, and includes
clustered errors and year and industry dummies. The dependent variable is the log change in sectoral
manufacturing employment. Lagged log changes in the price of investment and materials have been
omitted to save space.
37
get stronger. In column (3), we use the interaction of relative openness with changes in
the budget deficit ex-automatic stabilizers. The result is signficant at 90% with clustered
errors and year and sic dummies. The borderline significance of this result appears to be
driven by outliers, however, as if we estimate by minimizing the sum of absolute values
in column (4), the result becomes even stronger.
4.7 Additional Robustness
4.7.1 Specification
Table 7 shows that the baseline results from Tables 3 and 4 are robust to changes in
specification. In column (1), I minimize the sum of absolute values of residuals instead
of the sum of least squares, while keeping the same controls which were significant from
Table 3, clustering the errors, and including sectoral fixed effects. In column (2), I report
the results using random effects, in which the key results are little-changed from the FE
version in column (3).
4.7.2 Adding and Subtracting Fringe Sectors
One of the key changes in the classification of manufacturing industries from the Standard Industrial Classification (SIC) system to the North American Industrial Classification system (NAICs) was that much of the publishing sector, which includes book and
newspaper printing, was reclassified as non-manufacturing. Much of the value-added in
these sectors likely revolves around content creation rather than manufacturing per se,
and so all of the results reported thus far have not included publishing. To show that this
is not influencing the results, in column (1) of Table 8, I repeat the regression in equation
4.1 while including the publishing sector, showing that the results are little changed. In
column (2), I remove defense-related industries, which includes all sectors with at least
10% of shipments purchased by the department of defense in 1992 according to BEA
data. For industries with no BEA defense shipments data, I added in those industries,
such as tanks and tank components, which are clearly defense-related. Again, the results
are little-changed. In the third column I omit the industries that include “not elsewhere
classified” in the title. The logic is that since these sectors tend to be an amalgam of
loosely related subsectors, it is prudent to show that these sectors are not driving the
results.
38
Table 7: Robustness: Specification
Quantile
0.000939
(0.00243)
RE
0.00359**
(0.00170)
FE
-0.00413
(0.00406)
L.ln(WARULC)*Rel. Openness
-0.0660***
(0.00849)
-0.0721***
(0.00916)
-0.0738***
(0.0138)
ln∆ VA-per-Production Worker
-0.303***
(0.0138)
-0.236***
(0.0192)
-0.209***
(0.0245)
ln ∆ Demand
0.599***
(0.0148)
0.481***
(0.0279)
0.457***
(0.0453)
L.(K/L)
-0.00391
(0.0124)
0.0229***
(0.00886)
-0.0158
(0.0226)
L.(K/L)*L.Real Interest Rate
0.0416
(0.288)
-0.278
(0.320)
-0.516*
(0.292)
L. ln ∆ Hourly Wages
0.0484***
(0.00945)
0.0662***
(0.0175)
0.0540***
(0.0199)
L.ln ∆ Price of Shipments
0.0131
(0.00938)
0.0245**
(0.0110)
0.0279***
(0.0104)
L.ln ∆ Price of Materials
0.0784***
(0.0271)
0.113***
(0.0248)
0.131***
(0.0299)
L.ln ∆ Price of Energy
0.0276***
(0.00862)
0.0286**
(0.0124)
0.0597***
(0.0169)
L.ln ∆ PM*(M/S)
-0.142***
(0.0413)
-0.165***
(0.0373)
-0.186***
(0.0449)
L.ln ∆ PI*(I/S)
-0.690*
(0.367)
0.341
(0.341)
-1.186***
(0.409)
L.ln ∆ PE*(E/S)
-0.342*
(0.177)
14864
-0.637***
(0.156)
14864
0.479
-0.418***
(0.160)
14864
0.465
L.Relative Openness
Observation
Overall R-squared
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. All
regressions weighted by initial sectoral value-added, and include 4-digit SIC industry
and year fixed effects over the period 1973-2009. The dependent variable is the
log change in sectoral manufacturing employment. Results for the lagged log changes
in the price of investment is suppressed.
39
Table 8: Robustness: Add and Subtract Fringe Industries
(1)
(+) Publishing
-0.00396
(0.00396)
(2)
(-) Defense
-0.00433
(0.00406)
(3)
(-) Misc. SICs
-0.00451
(0.00459)
L.ln(WARULC)*Rel. Openness
-0.0764***
(0.0135)
-0.0758***
(0.0138)
-0.0786***
(0.0151)
ln∆ VA-per-Production Worker
-0.210***
(0.0248)
-0.204***
(0.0252)
-0.193***
(0.0262)
ln ∆ Demand
0.451***
(0.0461)
0.448***
(0.0476)
0.430***
(0.0503)
L.(K/L)
0.00901
(0.0227)
0.00804
(0.0233)
0.00234
(0.0234)
L.(K/L)*Real Interest Rate
-0.362
(0.318)
-0.341
(0.322)
-0.258
(0.406)
L.ln ∆ Price of Investment
0.0569
(0.0523)
0.0430
(0.0554)
0.0301
(0.0599)
L.ln ∆ Price of Energy
0.0136
(0.0159)
0.0154
(0.0166)
0.0199
(0.0184)
L.ln ∆ PI*(I/S)
-0.735*
(0.391)
-0.743*
(0.391)
-0.620
(0.431)
L.ln ∆ PE*(E/S)
-0.335**
(0.153)
448
15164
0.511
0.254
0.474
-0.399***
(0.141)
426
14537
0.510
0.401
0.476
-0.393***
(0.150)
428
12717
0.498
0.344
0.467
L.Relative Openness
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
Standard errors clustered on 4-digit SIC industries in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01.
All regressions weighted by initial sectoral value-added, and include 4-digit SIC industry and year
fixed effects over the period 1973-2009. The dependent variable is the log change in sectoral
manufacturing employment. Lagged shipments and materials prices, and materials-interaction
terms suppressed for space.
40
4.7.3 Comparing Various Measures of Relative Prices
There are many measures of “the” real exchange rate. Figure 15 compares several stateof-the-art measures of relative prices which use PWT v8.0 data and methodology to
more commonly used measures with well-known flaws. It can be seen that if we index
the IMF’s RULC index to the WARULC series computed using the 1975 benchmark
year of the PWT and then extrapolate forward, the IMF’s index now implies that US
ULC’s are nearly 40% lower than trading partners, and that the dollar was actually
undervalued during the 2000-2003 period when manufacturing collapsed. Also featured
are an updated version using PWT v8.0 of Fahle et al.’s (2008) weighted average relative
prices (WARP), and Balassa-Samuelson adjusted weighted average relative prices, described in Campbell (2013b). The Federal Reserve’s CPI-based Broad Trade-Weighted
Real Exchange Index is also featured, and also implies a steady dollar depreciation over
the period. The three “weighted average relative” indices all yield broadly similar results, although there are certainly differences in the details and in the implied degree of
overvaluation. One of the major differences, the more negative slope of WARULC, is
due to the declining share of labor income in manufacturing in the US relative to many
other developed countries, which is a broad-based phenomenon in manufacturing not
.6
.8
1
Index
1.2
1.4
1.6
caused by outsized changes in a small number of sectors.
1960
1975
1990
WARULC
Balassa−Samuelson Adjusted WARP
IMF RULC Index
2005
WARP
Fed’s RER Index
Figure 15: Comparing Various Exchange Rate Measures
As argued in Campbell (2013b), unit labor cost-based relative price measures are not
necessarily a priori better measures of competitiveness than Balassa-Samuelson adjusted
41
weighted average relative price (BSWARP) indices. This is because manufacturing requires many more inputs, including nontraded inputs, than just labor, as labor costs
were just 16% of shipment revenue as of 2007. Thus broader measures of prices may be
just as appropriate to gauge competitiveness as ULC indices.
In Table 9 I show that the results hold for all three of the “weighted average relative”
exchange rate indices. The first column illustrates the results using weighted average
relative prices for the manufacturing sector as a whole. In column (2), I used the lagged
log of the WARP index, and in column (3) I use the log of the Balassa-Samuelson
adjusted WARP index. In each case, the results are little-changed. The results are also
robust to using other transformations of these indices, such as subtracting one instead.
If the indices are left as-is, then the interaction term becomes almost collinear with
relative openness.
Additionally, in column (4), I use sector-specific trade weights, where the difference is
only that the trade weights are simply imports plus exports at the sectoral level. Sectoral
real exchange rates may a priori seem like a vast improvement over using real exchange
rates for the manufacturing sector as a whole, and, indeed, the “between” R-squared
nearly doubles, while the overall R-squared also increases, providing further evidence
that relative prices affect manufacturing employment. However, the magnitude is much
smaller in part because the variance of the sector-specific exchange rate is much higher.
Estimating with this index implies more jobs lost in periods when the overal WARULC
index is low, but also implies fewer jobs lost when WARULC is elevated.
While using either the overall WARULC or the sectoral version yields broadly similar
results, there are some complications with the sectoral version of WARULC which leave
reason for preferring the overall WARULC, and using both to calculate the impact on
jobs. The wider dispersion of sectoral WARULC values, with values ranging from .52 to
6.35 (indicating US ULCs are over six times higher than trading partners) implies movements in the dollar will lead to proportionally smaller changes in these high-WARULC
indexes relative to the difference between the high-value WARULC sectors and the lowvalue WARULC sectors. Thus the sectoral WARULC index will tend to predict a more
constant rate of job losses, while the overall WARULC index will not predict any jobs lost
when the overall WARULC index indicates unit labor costs are the same at home and
abroad. The larger sectoral WARULC values will give rise to a subtle multicollinearity
issue, as the interaction term between lagged relative openness and sectoral WARULC
will not vary much less year-by-year compared to the lower-value sectors. Additionally,
changes in the WARULC index arise from changes in trade shares and from changes in
the Divisia-based RULC index, which appear to have directionally similar, but not the
42
Table 9: Comparing Various Measures of Relative Prices
(1)
ln∆ L
-0.00413
(0.00406)
(2)
ln∆ L
-0.00919*
(0.00542)
(3)
ln∆ L
-0.0173***
(0.00601)
(4)
ln∆ L
-0.00349
(0.00412)
ln∆ VA-per-Production Worker
-0.209***
(0.0245)
-0.209***
(0.0245)
-0.209***
(0.0244)
-0.210***
(0.0245)
ln ∆ Demand
0.457***
(0.0453)
0.455***
(0.0464)
0.455***
(0.0462)
0.452***
(0.0470)
L.(K/L)
-0.0158
(0.0226)
0.00194
(0.0186)
0.00429
(0.0184)
-0.0103
(0.0218)
L.(K/L)*L.Real Interest Rate
-0.516*
(0.292)
-0.506*
(0.303)
-0.516*
(0.304)
-0.456
(0.296)
L.ln ∆ Price of Shipments
0.0279***
(0.0104)
0.0248**
(0.0105)
0.0243**
(0.0105)
0.0321***
(0.0115)
L.ln ∆ PI*(I/S)
-1.186***
(0.409)
-1.109***
(0.385)
-1.096***
(0.385)
-0.839**
(0.422)
L.ln(WARULC)*Rel. Openness
-0.0738***
(0.0138)
L.Relative Openness
L.Rel.Openness*ln(WARP)
-0.0711***
(0.0102)
L.Rel.Openness*ln(BSWARP)
-0.0783***
(0.0107)
L.ln(Sectoral WARULC)
0.0127
(0.00831)
L.Rel.Openness*ln(Sectoral WARULC)
-0.0331***
(0.00773)
437
14861
0.501
0.472
0.481
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
437
14864
0.503
0.226
0.465
437
14864
0.503
0.245
0.466
437
14864
0.504
0.247
0.467
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. All regressions are
weighted by initial sectoral value-added, and include 4-digit SIC industry and year fixed effects
over the period 1973-2009. The dependent variable is the log change in sectoral manufacturing
employment. Lagged log changes in the price of investment has been omitted to save space.
43
same, elasticities (Campbell 2013b). The added dispersion in the sectoral WARULC
will magnify any differences here as well, as well as any uncertainty over ULC data for
specific countries, such as China in the 1980s and 1990s.
There is no one “correct” measure of the real exchange rate, as each of the measures
discussed have their own strengths and weaknesses, but all of the indices share similar
broad patterns. Even the divisia-based indices created by the FRB and the IMF imply
that the dollar had two periods of appreciations, in the the 1980s and in the late 1990s and
early 2000s. Additionally, while it is very plausible that large movements in real exchange
rates could impact employment, it is less plausible that small year-to-year adjustments
could have an impact. Thus another strategy is to separate the data into periods when
all real exchange rate indices show a large appreciation and the periods when the real
exchange rate depreciated. The results are listed in Table 10 using the entire postBretton Woods period into four periods: 1972-1979, 1979-1986, 1986-1996, and 19962005. Instead of using any particular real exchange rate, I now interract openness with
a dummy for the periods of large relative price appreciations. The magnitude of the
coefficient in column (4) suggests that during the high RER periods, moving from average
to twice the average level of openness implied losing an additional 10% of manufacturing
employment relative to periods when the dollar fell.
4.7.4 Import Penetration vs. Export Share
While using an interaction of openness with the degree of overvaluation is the most
compelling research design, in fact there is some heterogeneity depending on whether
the openness stems from import exposure or export penetration, especially after 1990
(compare the first two columns in Table 11). One explanation for these disparate results
is that, over time, the correlation between import penetration and export share has
increased, from .095 in 1972 to .555 in 2009. Thus there is a multicollinearity problem in
the later years when both variables are included in the same regression. Over the entire
period (column three), there is not a statistically different impact on export exposure
and import penetration, even though the former is only significant at 10%.
There are good reasons why one might expect import penetration to do a better
job predicting job losses than the export share of trade. The first is that the exportweighted average relative unit labor cost (eWARULC) index has diverged from the
import-weighted average relative unit labor cost (iWARULC) index since 1990 (Figure 19). The second reason is that since trade was growing rapidly overall, when the
real exchange rate appreciated, exports were generally flat rather than falling, as was
the case with domestic shipments, which might entail less of a differential impact on
44
Table 10: Benchmark Years Only
(1)
ln∆ L
0.00657
(0.00450)
(2)
ln∆ L
0.00376
(0.0147)
(3)
ln∆ L
0.00882
(0.0177)
(4)
ln∆ L
0.00845
(0.0178)
-0.0496∗∗∗
(0.0136)
-0.0883∗∗∗
(0.0208)
-0.0988∗∗∗
(0.0215)
-0.0982∗∗∗
(0.0216)
1979 Dummy
0.392∗∗∗
(0.0251)
0.0601∗
(0.0292)
0.116∗∗∗
(0.0302)
0.107∗∗∗
(0.0306)
1986 Dummy
0.157∗∗∗
(0.0221)
0.0916∗∗∗
(0.0162)
0.101∗∗∗
(0.0156)
0.0957∗∗∗
(0.0167)
1996 Dummy
0.289∗∗∗
(0.0251)
0.137∗∗∗
(0.0258)
0.0912∗∗
(0.0282)
0.0899∗∗
(0.0291)
0.574∗∗∗
(0.0286)
0.719∗∗∗
(0.0395)
0.693∗∗∗
(0.0428)
ln∆ Shipments-per-Worker
-0.694∗∗∗
(0.0529)
-0.666∗∗∗
(0.0573)
Campa-Goldberg Low Markup*High RER
-0.0265
(0.0168)
-0.0240
(0.0184)
L.K/L
0.270∗∗∗
(0.0784)
0.277∗∗∗
(0.0824)
L.Relative Openness
Lagged Rel. Openness*High RER Year
ln∆ Demand
Change in Tariffs
0.299
(0.339)
Ch. Ins. & Freight Costs
-0.136∗
(0.0659)
-0.321∗∗∗
(0.0230)
437
1687
0.310
0.0524
0.220
Constant
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
-0.414∗∗∗
(0.0184)
437
1635
0.605
0.603
0.587
-0.133∗∗∗
(0.0304)
437
1635
0.728
0.692
0.705
-0.131∗∗∗
(0.0333)
422
1433
0.710
0.650
0.681
All Regression include fixed effects and errors clustered on 448 4-digit SIC industries (in parenthesis),
using data from the benchmark years 1972, 1979, 1986, 1996, and 2005. The dependent variable
is the log change in sectoral manufacturing employment from the previous period.
45
Table 11: Import Penetration vs. Export Share
(1)
1973-1989
-0.00199
(0.0103)
(2)
1990-2009
0.0129
(0.0102)
(3)
1973-2009
0.00268
(0.00462)
-0.0418**
(0.0208)
-0.0810***
(0.0237)
-0.0596***
(0.0156)
L.Rel. Export Share
-0.00569***
(0.00217)
-0.0204**
(0.00984)
-0.00198
(0.00166)
L.Rel. Export Pen.*ln(eWARULC)
-0.0786***
(0.0276)
0.0206
(0.0301)
-0.0436*
(0.0254)
ln∆ VA-per-Production Worker
-0.216***
(0.0371)
-0.202***
(0.0249)
-0.206***
(0.0245)
ln ∆ Demand
0.456***
(0.0672)
0.434***
(0.0400)
0.439***
(0.0456)
L.K/L
0.254***
(0.0531)
0.0955*
(0.0546)
0.0403*
(0.0224)
L.(K/L)*Real Interest Rate
-1.406***
(0.314)
0.282
(0.924)
-0.543*
(0.311)
L.ln ∆ Price of Shipments
0.0148
(0.0105)
0.0297**
(0.0146)
0.0328***
(0.00888)
L.ln ∆ Price of Investment
0.246**
(0.115)
0.0592
(0.159)
0.116
(0.0819)
-1.306***
(0.460)
432
7259
0.567
0.0845
0.440
1.394
(1.155)
394
7605
0.479
0.0159
0.416
-0.869**
(0.378)
437
14864
0.516
0.258
0.476
L.Rel. Import Penetration
L.Rel. Import Pen.*ln(iWARULC)
L.ln ∆ PI*(I/S)
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. All regressions
weighted by initial sectoral value-added, and include 4-digit SIC industry and year fixed effects.
The dependent variables is the log change in manufacturing employment.
46
1.6
1.5
1.4
1.4
.25
1980
1985
Import Penetration
WARULC
1990
1995
2000
2005
.8
1
1
1.1
1.2
1.2
1.3
WARULC
.2
.15
.1
.05
1975
2010
1970
Export Share of Shipments
(a) Import Penetration and Export Share Growth
1980
1990
Import−Weighted
2000
2010
Export−Weighted
(b) Import vs. Export-Weighted Average RULC
Figure 16: Growth of Trade, Divergence of RULC Indices
employment as compared with years when exports were growing. Additionally, as mentioned in the theoretical section, firms in sectors which are import-competing but do
not export are more likely to have productivity closer to the cutoff productivity for remaining in business, while firms that export in sectors which import little are likely to
be further away from the cutoff productivity, and thus less likely to go bankrupt merely
from a decline in the growth rate of exporting. Thus there are theoretical reasons to
suspect that import competition would have a larger impact on the extensive margin of
unemployment than export competition.
4.8 Economic Impact
What share of the employment collapse in manufacturing in the early 2000s can be attributed to changes in relative prices? One method is the simple accounting approach
as in Table 19, which implies that potentially the entire decline in manufacturing employment can be traced to trade given the rise in the manufacturing trade deficit. Of
course, there is no need for the US trade deficit in manufacturing ever to be balanced,
given the smaller US surplus in services, the large endowment of natural resources, and
the popularity of US financial assets. Even so, one could imagine an alternative scenario
where relative prices did not appreciate and the manufacturing trade deficit merely doubled over this period, in which it case it would also imply that about 95% of jobs lost
were due to trade.
A second method, however, is to use our regressions to estimate the impact of the
differential job losses in more tradable sectors when the relative unit labor costs in the
47
US appreciate. So far, we have found that the sectoral WARULC and the WARULC
index, while both yielding broadly similar results, seemed to pick up separate sources
of job losses. For example, the WARULC index for the manufacturing sector as a
whole only implies that a significant number of jobs are lost in the years in which with
WARULC index appreciates. The sectoral WARULC index, by contrast, implies that
many jobs are lost even in years in which the dollar is generally weak in the sectors in
which the the WARULC index is particularly strong, such as those sectors in the early
1990s which had a large share of trade with China. The sectoral WARULC index, by
construction, contains much more overall variation, but within each sector, year-to-year
movements in the dollar are a much smaller share of the overall variation. Thus, the
sectoral WARULC index alone will imply fewer jobs lost in high RER years.
The solution, then, is to use both indices in the same regression, as in Table 12. In
column (1), suppressing all of the variables significant from Table 4 for space, I show that
both the overall and sectoral versions of WARULC are significant. However, the problem
with fixed effect estimation using sectoral WARULC is that it would be expected that
sectors with higher relative prices for all periods might experience a continuous decline
in employment, which could be picked up in the fixed effect. Estimating with random
effects instead in column (2), the impact of sectoral WARULC becomes relatively larger
than the overall WARULC. Estimating just with OLS in column (3), the relative impact
of sectoral WARULC becomes even larger.
Lastly, in the last section I discussed reasons why it might actually be preferable to use
import penetration rather than openness. Thus, in column (4) I show the results when I
control for both sectoral and overall WARULC. In this case, the t-score on the sectoral
import-weighted average relative unit labor costs (siWARULC) is higher than the overall
(iWARULC). This translates into a total impact on jobs by year in Figure 18, with the
implied jobs lost by the sectoral WARULC in red, and the total impact in blue.18 This
implies that 2.4 million jobs were lost over the period 1997-2005, vs. a realized decline of
three million workers. By contrast, Autor, Dorn and Hanson’s (forthcoming) estimates
of the impact of increased trade with China imply that one-quarter of the jobs lost over
this period were due to trade with China.
Taking these estimates at face value, we can partition the changes in employment
each year due to demand, technology, and the impact of changes in relative prices,
in Figure 18. Over the entire period, it appears that technology is a more consistent
source of job decline, while demand growth been a consistent source of job gain. More
strikingly, however, the implied job gains generated by demand growth in the 2000s
18
To get the total impact, I used
P
i
= (L1.iW ARU LC −1)∗β0 ∗L1.Li +(L1.siW ARU LC −1)∗β1 ∗L1.Li
48
Table 12: Comparing Various Measures of Relative Prices
(1)
FE
0.000331
(0.00421)
(2)
RE
0.00824***
(0.00185)
(3)
OLS
0.00863***
(0.00227)
(4)
FE
ln∆ VA-per-Production Worker
-0.210***
(0.0244)
-0.237***
(0.0193)
-0.217***
(0.0251)
-0.212***
(0.0253)
ln ∆ Demand
0.456***
(0.0453)
0.481***
(0.0275)
0.462***
(0.0442)
0.454***
(0.0492)
L.ln(WARULC)*Rel. Openness
-0.0620***
(0.0145)
-0.0577***
(0.00974)
-0.0502***
(0.0124)
L.ln(Sectoral WARULC)
-0.00326
(0.00717)
-0.0187***
(0.00487)
-0.00685
(0.00607)
L.Rel.Openness*ln(Sectoral WARULC)
-0.0165**
(0.00772)
-0.0197***
(0.00418)
-0.0306***
(0.00496)
L.Relative Openness
L.Import Penetration
0.00766
(0.0216)
L.Import Pen.*(iWARULC-1)
-0.236***
(0.0521)
L.Sectoral iWARULC - 1
0.0139***
(0.00479)
L.Import Pen.*(Sectoral iWARULC - 1)
-0.0527***
(0.0133)
436
14710
0.507
0.465
0.485
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
437
14861
0.505
0.401
0.482
437
14861
0.481
0.582
0.495
14861
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. All regressions are weighted
by initial sectoral value-added and include year fixed effects over the period 1973-2009. Columns (1) and
(4) include 4-digit SIC industry fixed effects. Column (2) uses random effects, and column (3) is OLS.
The dependent variable is the log change in sectoral manufacturing employment. All other significant
regressors from previous regressions have been omitted to save space.
49
1.5
1.4
0
1.2
1.3
iWARULC
Jobs (1000s)
−200
−100
1.1
−300
1
−400
1975
1980
1985
1990
Total Impact
iWARULC
1995
2000
2005
2010
Impact from Sectoral WARULC
Figure 17: Economic Impact
was relatively weak compared to economic expansions before 2000. This may simply
be a function of sectoral shifts and slightly slower US population growth after 2000. A
second possibility may be that as imports displaced domestic manufacturing production,
demand for downstream industries, in both manufacturing and in the rest of the economy,
might have suffered. Every dollar of output of Apparel manufacturing requires 30 cents of
output from textile mills.19 Every dollar of industrial machinery requires 6.9 cents worth
of the output from iron and steel mills. Overall, every dollar of aggregate manufacturing
output generally requires about $0.60 worth of additional manufacturing output from
other manufacturing industries. Adjusting the estimate of 2.4 million jobs lost by this
multiplier would imply that relative prices caused a loss of employment well in excess of
three million.
4.9 Alternative Explanatory Variables
How do exchange rate movements impact output, other inputs, and prices? In Table 13,
I compare the impact on labor overall with the impact on production workers in column
(2) with the impact on non-production workers in column (3). The measured impact is
a bit larger for production workers, but the difference is not statistically signficant. In
column (4) it can be seen that the decline in the ratio of production workers to total
workers is not significant.
19
Data comes from the BEA’s Total Requirements Input-Output table.
50
Table 13: Impact on Production and Non-Production Workers
(1)
ln∆ L
-0.00293
(0.00396)
(2)
ln∆ PW
-0.00258
(0.00433)
(3)
ln∆ non-PW
-0.00123
(0.00378)
(4)
ln∆ PW Share
0.000353
(0.00129)
L.ln(WARULC)*Rel. Openness
-0.0811***
(0.0137)
-0.0858***
(0.0145)
-0.0764***
(0.0126)
-0.00466
(0.00300)
ln∆ VA-per-Production Worker
-0.206***
(0.0249)
-0.285***
(0.0289)
-0.0630***
(0.0215)
-0.0788***
(0.00965)
ln ∆ Demand
0.440***
(0.0462)
0.493***
(0.0492)
0.313***
(0.0373)
0.0528***
(0.00497)
L.K/L
0.00847
(0.0228)
0.0450***
(0.0146)
-0.0573
(0.0352)
0.0365***
(0.0133)
L.(K/L)*Real Interest Rate
-0.389
(0.318)
0.197
(0.300)
-1.508**
(0.600)
0.586***
(0.174)
L.ln ∆ Price of Shipments
0.0332***
(0.00918)
0.0284***
(0.00816)
0.0505***
(0.0143)
-0.00475
(0.00468)
L.ln ∆ Price of Investment
0.282***
(0.0769)
0.315***
(0.0705)
0.211**
(0.100)
0.0326
(0.0288)
L.ln ∆ Price of Energy
0.0350**
(0.0163)
0.0399**
(0.0167)
0.0167
(0.0215)
0.00492
(0.00664)
L.ln ∆ PI*(I/S)
-0.697*
(0.417)
-0.784**
(0.389)
-0.183
(0.714)
-0.0865
(0.187)
L.ln ∆ PE*(E/S)
-0.349**
(0.167)
437
14864
0.510
0.238
0.470
-0.362**
(0.165)
437
14864
0.558
0.264
0.511
-0.285
(0.207)
437
14859
0.189
0.189
0.153
-0.0132
(0.0649)
437
14864
0.132
0.0746
0.107
L.Relative Openness
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. All regressions weighted
by initial sectoral value-added, and include 4-digit SIC industry and year fixed effects over the period
1973-2009. The dependent variables are log changes in total labor, production workers, non-production
workers, and the share of production workers.
51
Table 14: Impact on Hours, Investment, VA, and Shipments
(1)
ln∆ PW Hours
-0.000644
(0.00123)
(2)
ln∆ I/S
0.00572
(0.00805)
(3)
ln∆ VA
-0.00123
(0.00449)
(4)
ln∆ Ship
0.000269
(0.00516)
L.ln(WARULC)*Rel. Openness
0.00664**
(0.00305)
-0.0463*
(0.0256)
-0.0867***
(0.0156)
-0.0942***
(0.0164)
ln∆ VA-per-Production Worker
0.0482***
(0.00692)
-0.198***
(0.0561)
0.653***
(0.0392)
0.112***
(0.0417)
ln ∆ Demand
0.0200***
(0.00521)
-0.000188
(0.0437)
0.443***
(0.0474)
0.596***
(0.0599)
L.K/L
-0.00341
(0.00500)
-0.121
(0.129)
-0.0249
(0.0343)
-0.0677*
(0.0354)
0.259
(0.382)
-2.597
(2.777)
-4.161***
(1.048)
-4.754***
(1.042)
L.ln ∆ Price of Shipments
0.000287
(0.00539)
0.224***
(0.0540)
0.00866
(0.0312)
0.00531
(0.0241)
L.ln ∆ Price of Investment
0.0979***
(0.0219)
1.024***
(0.197)
0.143
(0.105)
-0.0295
(0.0977)
L.ln ∆ Price of Energy
-0.00146
(0.00725)
-0.111
(0.0775)
0.0141
(0.0279)
0.0104
(0.0241)
L.ln ∆ PI*(I/S)
0.00964
(0.238)
-32.72***
(3.574)
-0.0534
(1.055)
-0.0725
(0.714)
L.ln ∆ PE*(E/S)
-0.0651
(0.0814)
437
14864
0.132
0.0768
0.0784
2.710***
(0.737)
437
14864
0.112
0.0137
0.0687
-0.826***
(0.312)
437
14864
0.673
0.154
0.592
-0.637**
(0.281)
437
14864
0.640
0.218
0.573
L.Relative Openness
L.(K/L)*Real Interest Rate
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. All regressions weighted
by initial sectoral value-added, and include 4-digit SIC industry and year fixed effects over the period
1973-2009. The dependent variables are log changes in production worker hours, investment/shipments,
value-added, and shipments.
52
Table 15: Impact on Wages
(1)
ln∆ PW
0.00305***
(0.00104)
(2)
ln∆ non-PW
0.00283**
(0.00123)
(3)
ln∆ Ratio
-0.000207
(0.00125)
(4)
ln∆ PW hourly
0.00369***
(0.00101)
L.ln(WARULC)*Rel. Openness
-0.00889**
(0.00376)
-0.00112
(0.00476)
0.00781
(0.00594)
-0.0155***
(0.00383)
ln∆ VA-per-Production Worker
0.105***
(0.0106)
0.00845
(0.00940)
-0.0961***
(0.0114)
0.0567***
(0.00727)
ln ∆ Demand
0.0296***
(0.00900)
0.0508***
(0.00910)
0.0211*
(0.0123)
0.00956
(0.00591)
L.K/L
-0.0666***
(0.0170)
-0.0220**
(0.00925)
0.0446***
(0.0137)
-0.0632***
(0.0189)
L.(K/L)*Real Interest Rate
-0.752***
(0.260)
-0.468**
(0.236)
0.281
(0.323)
-1.011*
(0.612)
L.ln ∆ Price of Shipments
0.00966
(0.00765)
0.0148*
(0.00822)
0.00512
(0.0119)
0.00937
(0.00602)
L.ln ∆ Price of Investment
0.194***
(0.0507)
0.00454
(0.0551)
-0.190***
(0.0586)
0.0962**
(0.0467)
L.ln ∆ Price of Energy
0.0170
(0.0104)
0.00365
(0.0172)
-0.0134
(0.0166)
0.0185
(0.0116)
L.ln ∆ PI*(I/S)
-0.277
(0.391)
-0.257
(0.364)
0.0178
(0.359)
-0.286
(0.387)
L.ln ∆ PE*(E/S)
0.155**
(0.0711)
437
14864
0.346
0.217
0.224
0.122
(0.0940)
437
14856
0.119
0.152
0.0511
-0.0333
(0.0907)
437
14856
0.0545
0.000992
0.0253
0.220*
(0.121)
437
14864
0.305
0.239
0.205
L.Relative Openness
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. All regressions weighted
by initial sectoral value-added, and include 4-digit SIC industry and year fixed effects over the period
1973-2009. The dependent variables are production worker pay and non-production worker pay per person,
non-production pay divided by production worker pay (per person), and hourly production worker pay.
53
Table 16: Impact on Inventory, Output Prices, and TFP
(1)
ln∆ Inventory
0.00480
(0.00424)
(2)
ln∆ Prices
-0.0000144
(0.00214)
(3)
ln ∆ TFP5
0.00182
(0.00206)
(4)
ln ∆ TFP4
0.00363
(0.00302)
L.ln(WARULC)*Rel. Openness
-0.0122
(0.0140)
-0.00449
(0.00483)
-0.0284***
(0.00464)
-0.0363***
(0.00677)
ln∆ VA-per-Production Worker
-0.759***
(0.0339)
-0.0613***
(0.0170)
0.265***
(0.0189)
ln ∆ Demand
-0.0381**
(0.0190)
-0.0435***
(0.0126)
0.196***
(0.0246)
0.290***
(0.0315)
L.K/L
-0.157***
(0.0374)
0.0695*
(0.0374)
0.0252
(0.0263)
0.0500**
(0.0237)
L.(K/L)*Real Interest Rate
0.505
(1.003)
-4.369***
(0.889)
-0.0749
(0.222)
1.071***
(0.369)
L.ln ∆ Price of Investment
0.144
(0.144)
-0.278***
(0.0964)
0.0211
(0.0592)
0.0757
(0.0688)
L.ln ∆ Price of Energy
0.0683
(0.0491)
-0.0295
(0.0205)
0.00782
(0.0119)
-0.00417
(0.0154)
L.ln ∆ PI*(I/S)
-2.202
(1.488)
1.283
(0.949)
-1.598***
(0.437)
-1.392***
(0.518)
0.981***
(0.325)
426
11442
0.398
0.113
0.285
-0.492**
(0.232)
426
11442
0.370
0.179
0.270
0.317***
(0.103)
426
11442
0.601
0.628
0.574
0.270**
(0.137)
426
11442
0.399
0.361
0.354
L.Relative Openness
L.ln ∆ PE*(E/S)
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
Standard errors clustered on 4-digit SIC industries in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01.
All regressions weighted by initial sectoral value-added, and include 4-digit SIC industry and year
fixed effects over the period 1973-2009. The dependent variable is the log change in sectoral
manufacturing employment. The coefficients on Tariffs and changes in insurance and freight
rates are not significant and are omitted for space.
54
Table 17: Impact on Job Creation, Job Destruction, and ULCs
(1)
Job Creation
-0.0217
(0.274)
(2)
Job Destruction
0.480*
(0.263)
(3)
ln∆U LC
0.000760
(0.00186)
L.ln(WARULC)*Rel. Openness
-1.506***
(0.343)
4.885***
(1.211)
0.000867
(0.00589)
ln∆ VA-per-Production Worker
-5.289***
(0.724)
6.678***
(1.449)
-0.772***
(0.0284)
-0.772***
(0.0284)
ln ∆ Demand
9.158***
(1.400)
-15.34***
(3.108)
0.0220
(0.0134)
0.0222*
(0.0134)
L.(K/L)
5.120**
(2.089)
8.672**
(4.199)
-0.0170
(0.0440)
-0.0178
(0.0452)
L.(K/L)*Real Interest Rate
-42.14**
(16.82)
56.21
(71.92)
3.454***
(1.205)
3.457***
(1.209)
L.ln ∆ Price of Materials
-7.159***
(2.739)
-9.627**
(3.746)
-0.235***
(0.0453)
-0.233***
(0.0452)
L.ln ∆ Price of Investment
-9.172
(6.608)
-20.55**
(7.948)
0.264***
(0.100)
0.262***
(0.100)
L.ln ∆ Price of Energy
-0.681
(0.831)
-2.480*
(1.287)
0.0230
(0.0182)
0.0235
(0.0182)
11.93***
(4.601)
15.17***
(5.496)
0.414***
(0.0782)
0.411***
(0.0788)
L.ln ∆ PI*(I/S)
40.87
(27.15)
148.3***
(52.87)
-0.257
(1.100)
-0.276
(1.104)
L.ln ∆ PE*(E/S)
-6.714
(9.350)
10.09
(14.93)
0.608***
(0.218)
0.605***
(0.218)
L.Relative Openness
L.ln ∆ PM*(M/S)
(4)
ln∆U LC
L.Import Penetration
0.00574
(0.0104)
L.Import Pen.*ln(WARULC)
0.0137
(0.0317)
437
14864
0.653
0.116
0.601
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
448
10985
0.274
0.0133
0.123
55
437
10729
0.331
0.00521
0.141
437
14864
0.653
0.110
0.600
Standard errors clustered on 4-digit SIC industries in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01.
All regressions weighted by initial sectoral value-added, and include 4-digit SIC industry and year
fixed effects over the period 1973-1998. The dependent variables are the percentages of job creation,
job destruction, and the log change in ULCs (wages over value-added).
1000
500
Jobs (1000s)
0
−500
−1000
1975
1980
1985
1990
1995
Demand
Relative Prices
2000
2005
2010
Technology
Figure 18: Sources of Job Growth and Losses
In Table 14, I show that production hours per worker do increase modestly but significantly, and that investment as a share of shipments, value-added, and shipments all
fared worse in sectors with higher levels of openness when relative unit labor costs were
higher. In this specification, investment over shipments is borderline significant, and
alternative specifications are only sometimes significant.
In Table 15, I show that total pay per person for production workers falls slightly, but
that non-production workers were not significantly affected, nor was the ratio. I also find
a slight negative impact on the hourly wages of production workers, a new result for this
literature. This impact is likely mitigated by compositional changes during periods of
job destruction – low productivity workers, who likely also have low wages, are probably
more likely to be laid off first. Even so, the relatively small impact on wages seems to
suggest that trade is unlikely to be a major cause of the rise in inequality as compared
to institutional changes (see Levy and Temin, 2007).
In Table 16, I report the impacts for changes in inventory, prices, and two kinds
of TFP, one assuming four factors and one assuming five. There does not appear to
be any significant relationship between the overvaluation and openness interaction and
inventory or price changes. Interestingly, there is a general negative relationship between
openness and prices, but this price competition does not appear to become more intense
when the real exchange rate increases. There is a significantly negative impact on TFP,
which is another new result for this literature.
56
Lastly, in Figure 17 I show the impact of exchange rate movements on job creation, job
destruction, and unit labor costs. When unit labor costs in the US rise relative to trading
partners, there is suppressed job creation, but the bigger impact is on job destruction.
Since job creation varies much less than job destruction overall, this assymmetry is an
important “fingerprint” of hysteresis. Nearly four good years of job creation are needed
for every bad year of destruction. Lastly, while recent research (Elsby et. al., 2013)
have found that offshoring is partly responsible for the decline in the overall US share
of income, in column’s (3) and (4) it does not appear that there is a differential impact
of exchange rate movements on the unit labor costs of more open sectors. Additionally,
in Figure 19(a), it can be seen that there has been a steady decline in average unit
labor costs in manufacturing since 1958, regardless of whether we plot a simple weighted
average (weighting by sectoral employment), or by doing a weighted average of changes
in sectoral ULC indices (divisia). This tells us that the decline in ULCs was broad-based,
and not due to compositional changes in the size of different sectors. In Figure 19(b),
separating the sectors with more than 10% of import penetration in 1975 from those
with less, it can be seen that while there are different patterns, overall the trend is very
similar. If anything, the years of fast import growth seemed to raise ULCs of the sectors
.2
.2
.3
.3
.4
.4
.5
.5
which had more import competition in 1975.
1960
1970
1980
1990
Weighted−Average ULCs
2000
2010
1960
Average ULCs (Divisia)
1970
1980
Import Competing
(a) Divisia vs. WAULC
1990
2000
2010
Not Import Competing
(b) ULCs by Import Share of Trade, 1975
Figure 19: Unit Labor Costs
4.10 Impact on Trade
For there to be an impact of real exchange rates on manufacturing, a necessary condition
is that there is an impact on trade. In this section, first I look at the example of Japan,
57
a relevant case-study, and then I model imports using a panel vector error correction
model.
4.10.1 The Case of Japan
As US trade with China has increased rapidly since 1980, regaining the bilateral trade as
a share of US GDP the two countries enjoyed in the 1820s in the early 2000s (Figure 20),
Chinese growth has become the center of focus for those wishing to explain the decline
of US manufacturing. Similarly, in the 1980s many Americans blamed manufacturing
job losses on Japan’s rise, as imports from Japan displayed an astonishing increase from
0
Bilateral Trade over US GDP
.02
.04
.06
.08
the 1950s to the mid-1980s, only slightly more gradual than China’s later rise.
1820 1830 1840 1850 1860 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
China
Japan
Germany
France
Canada
UK
Figure 20: The US Pattern of Trade, 1820-2010
Japan is also a good case study because the Yen was heavily managed and Japan
had capital controls until 1980, with the dollar rising thereafter for reasons already
explained, and additional liberalization measure coming in 1984. Figure 21 displays
the recent relative price history between the US and Japan. It can be seen that the
Yen appreciated substantially vs. the dollar after the 1985 Plaza Accord. As a result,
US hourly manufacturing workers went from enjoying wages twice than that of their
Japanese counterparts in 1985 to having wages that were close to parity three years
later. US unit labor costs relative to Japan fell 47% and the real exchange rate using
PPP from the Penn World Tables, v8.0, implies an appreciation of Japanese relative
prices of 37%. Thus the case of Japan yields a relatively clean quasi-natural experiment
for the impact of currency undervaluation and large exchange rate adjustments. It may
58
1970
1975
1980
1985
1990
1995
RULC
Japan−US PWT RER
2000
2005
40
30
20
10
0
0
1
2
3
4
0
100
200
300
400
500
National Currency per U.S. Dollar, period average
also yield policy lessons for what to do about China.
2010
1970
Relative Hourly Man. Wages
Nominal Exchange Rate
1975
1980
1985
1990
US Wages
(a) US-Japan Exchange Rates
1995
2000
2005
2010
Japanese Wages in Dollars
(b) Manufacturing Wages (2011 Dollars)
Figure 21: Sticky Wages, ULCs, and Imports
The result of this real appreciation was that as wages in the Japanese manufacturing
sector suddenly increased substantially relative to their American counterparts, the meteoric Japanese export growth from 1975-1986 suddenly ground to a halt (Figure 22(a)).
This is also not purely related to factors going on in Japan, such as government encouragement of increasing market share in export markets, since different trends hold up in
other markets. In the UK case, Japanese exports grew very quickly in the early 1980s,
when the Yen was weak relative to the pound, but Japanese import penetration to the
UK market did not grow at all from 1983-1985, when Japanese unit labor costs were
1980
1985
1990
1995
Japanese Import Penetration
2000
2005
1.4
.025
1.2
.8
1
UK−Japan RULC
.6
.005
.6
.015
1975
Japanese Import Penetration
.01
.015
.02
.8
Import Penetration
.02
.025
1
1.2
Bilateral RULC
1.4
.03
higher than UK unit labor costs.
2010
1980
US−Japan RULC
1985
1990
1995
Japanese Import Penetration
(a) US
2000
(b) UK
Figure 22: RULCs and Japanese Import Competition, US vs. UK
59
2005
UK−Japan RULC
Hence, on balance Krugman (1986) appears to have been correct to guess that the
Yen’s appreciation in that year meant that “the Japan problem was over” in the sense
that Japanese manufacturers would not continue to increase their market share in the
US. He was also correct that Japan’s experience in the 1980s was not special – when the
dollar was strong, other countries saw market share increases in the US, and on average
this was not reversed after the dollar weakened, although the growth did not continue.
4.10.2 Modeling Imports Using A Panel Vector Error Correction Model
In this section, I test the impact of relative unit labor costs on US imports by sector
and destination country using a simple error correction formulation, similar to Fahle et
al. (2008) and Chinn (2004).
∆lnMh,c,t = α +
4
X
βj · ∆lnMh,c,t−j + ·ψ∆lnDh,t +
j=1
7
X
µk ∆ · ln RU LCc,t−k
(4.3)
k=1
j
∗ − · lnRU LC
+θ(lnXt−1 − ϕ · lnYt−1 − η · lnYt−1
t−1 ) + ut
t = 1979,...,2009 ; c = 41 Major US Trading Partners, h = 448 4-digit SIC
Manufacturing Sectors
Where Mh,c,t are US imports from country c in sector h at time t, and Dh,t is demand
in sector h at time t, and RU LCc,t−k are the relative unit labor costs between the US and
country c at time t-k. As there is potential for simultaneity bias, I helmert-transform
the data and posit a panel vector-autoregression model explaining imports, demand,
and unit labor costs and apply Love and Ziccino’s (2006) panel VAR estimator from
Holtz-Eakin, Newey, and Rosen (1988).
In column (1), when I just run OLS with country-industry fixed effects, I find that the
lagged level of RULC impact I find that movements in RULCs impact trade up to seven
lags. Column (2) is the first stage of the ECM estimator, and column (3) estimates the
second stage. However, with fixed effects and a lagged dependent variable, there is a
problem with Nickel bias, and so the panel VAR estimator in equation (4) should be
preferred.
In Figure 23 I present graphs of the impulse response functions and the 5% error
bounds generated by Monte Carlo simulation. The middle column in the top row shows
the impulse response of log changes in imports to movements in relative unit labor costs,
with the impact dying off after about six years.
60
Table 18: Imports and Relative Prices: Error Correction Model Results
(3)
Second Stage
0.739∗∗∗
(0.0249)
(4)
GMM
L.lchimports
-0.230∗∗∗
(0.00525)
-0.216∗∗∗
(0.0051)
L.lchRULC
0.318∗∗∗
(0.0163)
0.280∗∗∗
(0.0174)
L2.lchRULC
0.255∗∗∗
(0.0167)
0.279∗∗∗
(0.0173)
L3.lchRULC
0.296∗∗∗
(0.0189)
0.291∗∗∗
(0.0197)
L4.lchRULC
0.111∗∗∗
(0.0174)
0.152∗∗∗
(0.0186)
L5.lchRULC
0.150∗∗∗
(0.0162)
0.221∗∗∗
(0.0170)
L6.lchRULC
0.0831∗∗∗
(0.0174)
0.093∗∗∗
(0.0186)
L7.lchRULC
0.0546∗∗
(0.0171)
0.109∗∗∗
(0.0167)
L.ln∆ Sectoral Demand
0.264∗∗∗
(0.0170)
0.264∗∗∗
(0.0170)
ln∆ Sectoral Demand
L.Bilateral RULC
(1)
FE
0.721∗∗∗
(0.0228)
(2)
First Stage
0.0648∗∗∗
(0.00485)
0.412∗∗∗
(0.0297)
0.903∗∗∗
(0.0219)
Log Sectoral Demand
-0.0336∗∗∗
(0.000810)
L.resid3
Country-Industry Combinations
Observations
R-squared
8714
234015
0.0170
241454
177272
177272
Errors clustered at the Country-SIC industry level in parenthesis. The Dependent variable in the
first and third columns is the log change in US imports from country i in sector j, over the period
1973-2009. The dependent variable in the second column is log imports. The lagged residuals from the
second column are then used as a regressor in column 3. Column 4 is a Panel VAR using GMM.
61
Impulse−responses for 7 lag VAR of lchimports lchRULC lchdemand
Sample : (pif5)majorec
==lchimports
1 & tobs == 37
lchimports
(p 5) lchRULC
lchRULC
(p 95) lchimports
(p 5) lchdemand
(p 95) lchdemand
(p 95) lchRULC
0.7241
0.0325
−0.1558
−0.0043
6
0
0.0000
6
0
s
(p 5) lchimports
(p 95) lchimports
s
response of lchimports to lchRULC shock
lchimports
(p 5) lchRULC
(p 95) lchRULC
0.0038
response of lchimports to lchdemand shock
lchRULC
(p 5) lchdemand
(p 95) lchdemand
0.1090
−0.0003
6
0.0000
0
6
s
0
6
s
s
response of lchRULC to lchRULC shock
lchimports
(p 5) lchRULC
(p 95) lchRULC
0.0172
response of lchRULC to lchdemand shock
lchRULC
(p 5) lchdemand
(p 95) lchdemand
0.0000
0.0000
6
0.0000
0
6
s
response of lchdemand to lchimports shock
lchdemand
0.1391
−0.0156
0
lchdemand
0.0176
−0.0194
0
response of lchRULC to lchimports shock
(p 5) lchimports
(p 95) lchimports
6
0
s
response of lchimports to lchimports shock
lchdemand
0.0541
0
s
response of lchdemand to lchRULC shock
6
s
response of lchdemand to lchdemand shock
Errors are 5% on each side generated by Monte−Carlo with 200 reps
Figure 23: Impulse-Response Functions from Panel VECM
5 Future Research
There are two projects which are a natural continuation of this line of research. The first
is to show that in a New Keynesian model similar to Krugman and Eggertson (2011),
capital inflows can lead to both exchange rate appreciation and debt-accumulation. In a
situation in which monetary authorities either cannot or will not act to stabilize demand,
capital inflows have a direct negative affect on aggregate demand and make deleveraging
more difficult.
The second is to examine the local labor market effects of exchange rates movements
for the 1980s, similar to Autor, Dorn, and Hanson’s (forthcoming) work on Chinese
import competition since 1990. Figure 24 is a chloropleth map of log changes in income by country from 1979-1986. It is easy to see that the midwest, specialized in
import-competing manufacturing and agriculture, which is also sensitive to exchange
rate movements, suffered relative to other areas of the country.
For example, Indiana is the smallest of the core manufacturing states, and has the
least diversified economy. In 1979, Indiana’s GDP per capita was close to that of New
York. However, the import-competing sectors were hard hit when the dollar rose in the
mid-1980s, and employment, and per capita income in Indiana suffered.
62
Figure 24: Log Change in Income, 1979-1986
Indiana and Ohio, Total Employment vs. National
Indiana, Manufacturing Employment
Index, 1977 = 100
400,000
180
380,000
170
360,000
160
340,000
150
All Other
Manufacturing
Sectors
320,000
300,000
Total Employment
National
140
130
280,000
120
260,000
Total Employment
Ohio
(light blue)
110
240,000
100
Import-Competing
Sectors
220,000
Total Employment
Indiana (Dark Blue)
90
200,000
(a) Dollar Depreciation: 1972-1979
(b) Dollar Appreciation: 1979-1986
Figure 25: Rust Belt Hysteresis
63
1997
1996
1995
1994
1993
1992
1991
1990
1989
1988
1987
1986
1985
1984
1983
1982
1981
1980
1979
1978
1977
1997
1996
1995
1994
1993
1992
1991
1990
1989
1988
1987
1986
1985
1984
1983
1982
1981
1980
1979
1978
1977
80
6 Conclusion
In this paper, I documented that when nominal real exchange rates move, nominal wages
are sticky, leading to large implied changes in competitiveness. Dollar appreciations
appear to lead to increased imports and decreased exports, and adverse impacts on the
manufacturing sector. These adverse impacts are concentrated in sectors with higher
initial levels of openness, and were also large relative to more tradable sectors in other
large economies.
This paper also confirms an assymmetry between job creation and job destruction,
with both responding to real exchange rate appreciation, but job destruction being four
times as volatile. This is one “fingerprint” of hysteresis, as four solid years of job growth
are needed to counteract the losses from one bad year. The 1980s experience was a
clear case of a random shock to government spending and taxation having a hysteretic
effect on more open sectors. More open sectors in 1979 lost many more jobs over the
period 1979-1986, but did not regain any of the jobs lost in subsequent period with
US ULCs were approximately close to the ULCs of US trading partners. Testing for
hysteresis in the 2000s is a bit more difficult given the financial crisis and given that
the continued low unit labor costs in China has meant that pressure on a growing
number of manufacturing sectors has not yet subsided. However, the meager recovery of
manufacturing employment after 2009, and the continued large current account deficit
despite greatly improved relative prices suggests that hysteresis has not yet gone away.
The magnitude of the impacts suggest that the bulk of the reason for manufacturing’s
collapse after 2000 (or after 1997 on a cyclically-adjusted basis) was movements in relative prices, matching the US experience in the 1980s. Increased trade with China was
a large part of the increase in WARULCs for the US, but the dollar’s general strength
in the early 2000s was also a result of other factors, including fiscal policy choices of the
US government.
The decline in manfucturing employment and rise in the current account deficit, which
has continued to the present, almost certainly lowered the US savings rate over the
past 10 years and has made the task of deleveraging more difficult given the ability and
willingness of highly-indebted consumers to spend. As the US has now been in a liquidity
trap for five years, a time in which the Federal Reserve has been far more passive20 than
20
For example, from 1955 to 2008, there was never a two year period where the federal funds rate was
constant. In 2009 and 2010, by contrast, the federal funds rate was constant and the Fed’s balance
sheet did not change substantially. (QE did not begin in earnest until the very end of 2010.) Since
then, the Federal Reserve has made policy changes roughly once per year, even in the context of
large-scale downward revisions to its own growth forecasts.
64
it was when not restricted by the zero lower bound, it appears that the Federal Reserve is
either unable or unwilling to offset the contractionary effects of continued official capital
inflows.
Two key policy lessons can be drawn from the US experience in the 1980s. The
first is that there can be long-term negative impacts of large-scale fiscal deficits at full
employment. This lesson is not relevant for current policy, as the US is not currently at
full employment, and the current Federal Reserve Board does not apear likely to offset
the contractionary effects of fiscal contraction. The second lesson is that pressuring
Japan to liberalize its capital account fully and the Plaza Accord intervention appear to
provide a relevant example of what Presidential leadership might accomplish regarding
the Bretton Woods II system of managed exchange rates, capital controls, and large-scale
accumulation of official reserves.
7 Appendix
7.1 Appendix: Figures
65
2.5
12%
10%
2
8%
6%
1.5
4%
2%
1
0%
1954
0.5
1962
1970
1978
1986
1994
2002
2010
-2%
-4%
0
1954
1960
1966
1972
1978
1984
1990
1996
2002
2008
-6%
(a) Log Scale
(b) Annual, 5 year, and 15 year MAs (% ∆)
Figure 26: Manufacturing Productivity: Chained Quantity Index per Worker
The Decline and Fall of US Manufacturing
80
90
RER
100
110
Structural Log Manufacturing Employment
9.4
9.5
9.6
9.7
9.8
1948
1990
1995
2000
year
2005
Structural Log Manufacturing Employment
2010
2015
RER
Figure 27: Structural Manufacturing Employment
66
1.5
1.4
.2
1.2
1.3
iWARULC
0
1.1
−.2
1
−.4
1975
1980
1985
1990
Coeff. on Import Penetration
2 s.e. Error Bound
iWARULC
1995
2000
2005
2010
2 s.e. Error Bound
Real Interest Rate
Figure 28: The RER and the Impact of Import Penetration by Year
67
Figure 29: Income per Capita, 1979
Figure 30: Import-Competing Manufacturing Employment (1,500 worker minimum),
Share of Total Employment, 1979
68
1
.5
0
−.5
−1
−1.5
0
.2
.4
.6
Import Penetration, 1979
Log Change in Employment, 1986−1996
.8
Slope not Significant
Figure 31: Hysteresis: No Rebound After Collapse
7.2 Appendix: Tables
Table 19: Manufacturing Employment Accounting
Productivity Deficit Δ from Actual Jobs Man. Cons.
Manufacturing Manufacturing (1 million
1995 over
Lost in Man.
Over
Year Consumption Trade Deficit
workers)
Productivity Since 1995 Productivity
1995
1340
159
68
0.00
0
19.56
1996
1361
152
70
-0.09
-0.01
19.41
1997
1432
155
73
-0.05
0.17
19.53
1998
1542
215
76
0.75
0.32
20.41
1999
1661
293
79
1.70
0.08
21.03
2000
1780
364
82
2.50
0.02
21.70
2001
1688
344
82
2.26
-0.80
20.64
2002
1760
404
89
2.76
-1.99
19.81
2003
1822
448
95
3.05
-2.74
19.24
2004
2023
540
104
3.68
-2.93
19.53
2005
2158
590
110
3.92
-3.02
19.58
69
Table 20: Impact of Import Penetration on Employment by Year (no controls)
(1)
1979
-0.327
(0.180)
(2)
1986
-0.684∗∗∗
(0.166)
(3)
1996
-0.240
(0.136)
(4)
2005
-0.757∗∗∗
(0.119)
0.0965∗∗∗
(0.0168)
440
0.00823
-0.118∗∗∗
(0.0173)
441
0.110
0.0360
(0.0224)
440
0.0156
-0.178∗∗∗
(0.0238)
388
0.161
Lagged Import Penetration
Constant
Observations
r2
Standard errors in parentheses
Each column represents a Quantile Regression from an individual year with the log change in Employment
from previous benchmark years as the dependent variable. E.g., 1972-1979, 1979-1986, etc.
∗
p < 0.05,
∗∗
p < 0.01,
∗∗∗
p < 0.001
Table 21: Explaining the 1980s, high RERs or RIRs?
(1)
ln∆ I, 1983
0.198∗
(0.1000)
(2)
ln∆ I, 1986
-0.207∗
(0.102)
(3)
ln∆ PW, 1983
-0.0799∗∗∗
(0.0162)
(4)
ln∆ PW, 1986
-0.151∗∗∗
(0.00992)
ln∆ Shipments per Worker
-0.862∗∗∗
(0.250)
-0.865∗∗
(0.301)
-0.924∗∗∗
(0.0264)
-0.809∗∗∗
(0.0291)
ln∆ Demand
1.078∗∗∗
(0.153)
1.158∗∗∗
(0.242)
0.861∗∗∗
(0.0162)
0.867∗∗∗
(0.0235)
Constant
0.00245
(0.0180)
441
0.201
-0.0813∗∗∗
(0.0230)
442
0.0575
-0.0128∗∗∗
(0.00279)
441
0.873
-0.0000966
(0.00223)
442
0.789
Lagged Import Penetration
Observations
r2
Standard errors in parentheses
Each Regression is for a single year.
∗
p < 0.05,
∗∗
p < 0.01,
∗∗∗
p < 0.001
70
Table 22: Import Penetration and Manufacturing Employment
(1)
ln∆ L
-0.0315*
(0.0175)
(2)
ln∆ L
0.00213
(0.0222)
(3)
ln∆ L
0.0621**
(0.0245)
(4)
ln∆ L
0.0605**
(0.0273)
-0.261***
(0.0722)
-0.351***
(0.0553)
-0.443***
(0.0603)
-0.436***
(0.0645)
ln ∆ Shipments-per-Worker
-0.594***
(0.0553)
-0.611***
(0.0644)
-0.612***
(0.0641)
ln ∆ Demand
0.569***
(0.0687)
0.565***
(0.0841)
0.565***
(0.0839)
Change in Tariffs
-0.000992
(0.00338)
-0.000960
(0.00332)
Change Ins. & Freight Costs
0.00227
(0.00429)
0.00224
(0.00426)
L.K/L
0.0431**
(0.0219)
0.0432**
(0.0215)
-0.379*
(0.213)
-0.448**
(0.215)
L.Import Penetration
L.iWARULC*Import Penetration
L.(K/L)*Real Interest Rate
L.Import Penetration*Real Interest Rate
-0.0834
(0.525)
L.ln ∆ Wages
-0.00799
(0.0171)
Campa-Goldberg Low Markup*L.iWARULC
0.0256
(0.0167)
Intermediate Import Share*L.iWARULC
0.0568
(0.133)
436
11675
0.625
0.617
0.614
Industries
Observations
Within R-squared
Between R-squared
Overall R-squared
448
15232
0.219
0.183
0.170
448
15164
0.639
0.460
0.630
436
11675
0.625
0.645
0.616
Clustered standard errors in parenthesis. *p < 0.1, ** p < 0.05, *** p < 0.01. All regressions weighted
by initial sectoral value-added, and include 4-digit SIC industry and year fixed effects over
the period 1973-2009. The dependent variable is the log change in sectoral manufacturing employment.
71
Table 23: Impact of Openness on Employment by Period
(1)
1972-9
0.107∗
(0.0507)
(2)
1979-86
-0.806∗∗∗
(0.108)
(3)
1986-96
0.155
(0.0983)
(4)
1996-2005
-0.567∗∗∗
(0.146)
ln∆ Demand
0.978∗∗∗
(0.0138)
0.991∗∗∗
(0.0251)
0.831∗∗∗
(0.0397)
0.893∗∗∗
(0.0590)
ln∆ Shipments-per-Worker
-0.984∗∗∗
(0.0190)
-1.028∗∗∗
(0.0330)
-0.849∗∗∗
(0.0467)
-0.871∗∗∗
(0.0855)
L.K/L
-0.0271
(0.0454)
0.0923∗
(0.0433)
-0.0295
(0.0526)
0.0197
(0.101)
L.Log VA per Worker
0.0134
(0.00815)
0.0179
(0.0108)
0.00437
(0.0151)
0.0261
(0.0203)
Change in Tariffs
0.124
(0.138)
-0.0202
(0.250)
0.0668
(0.124)
Ch. Ins. & Freight Costs
0.0154
(0.113)
-0.0871∗
(0.0355)
0.0110
(0.0841)
0.107
(0.0613)
372
0.818
0.0422
(0.0716)
331
0.455
0.0902
(0.101)
379
0.612
Lagged Openness
Constant
Observations
r2
0.0911
(0.0562)
440
0.802
Errors clustered at the 2-digit SIC level in parenthesis. Each column represents a
Quantile Regression from an individual year with the log change in Employment from
previous benchmark year. E.g., 1972-1979, 1979-1986, etc.
72
7.3 Mathematical Appendix
Proposition 1.1: Transition Dynamics with Homogenous Firms
If we impose symmetry among C countries, H industies, and all firms, so that all firms
have the same productivity (similar to the Krugman 1980 model) and all markets have
the same fixed cost of entry, and assume that we are starting from autarky in the steady
state, it is straightforward to derive a simple dynamic gravity equation for exports from
country i to j at year t following the move from autarky to free trade at time t=0.
Xij,t = δ
t
X
Yj,t−s Yi,t−s
s=0
Yw,t−s
(1 − δ)s
(7.1)
With fixed country sizes, eventually the economy will reach a new steady state:
lim Xijt =
t→∞
Yj Yi
Yw
(7.2)
Where we have used the fact that, by assumption, all countries are the same size. See
below for a proof. The above formulation is for specific parameter values, and while
I make no pretense of generality, the implications of dynamic gravity are confirmed in
more general, and more complicated, formulations.
Proof : Let Yj be the GDP of country j in autarky, with a saturation number of firms.
Since each firm is the same size, δYj of output is lost each period by firms naturally
going out of business. With resources available, new firms will enter, and since the entry
costs into any market are exactly the same, a new firm from country i will be indifferent
between entering into any one of C markets. Output will be unchanged, but in the first
period, exports from one country to another will be equal to:
Xij0 = δYj0 /C = δ
Yj0 Yi0
Yw0
(7.3)
Where the last equality holds due to our assumption of symmetry. In the next period,
again due to our assumption of random firm deaths, more resources are freed up and
then randomly distributed over the C markets, so trade in period 1 is equal to:
Xij1 = δ
Yj0 Yi0
Yj1 Yi1
+ (1 − δ)δ
Yw1
Yw0
(7.4)
The proposition follows from iterating out in this manner.
Note, the “saturation” number of firms comes due to the fact that since most firms
survive from one period to the next, if all firms were destroyed (by a war or a tsunami,
73
for example), then in the first period, the total number of firms would be given by:
Lj =
M
X
k=1
lk =
M
X
qk
(
k=1
ϕ
+ f ) = M (
qk
+ f )
ϕ
(7.5)
Here, qk will be pinned down by the zero profit conditions for entry. In the next period,
however, there will be M0 (1−δ) firms which no longer need to pay the entry cost. Hence,
the full employment condition now gives us:
Lj =
M
X
lk = M0 (1 − δ)(
k=1
qk
qk
) + M1 ∗ ( + f )
ϕ
ϕ
(7.6)
Clearly, the number of firms should increase in this period for a wide range of parameter
values, since the incumbent firms no longer need to pay down the fixed costs of entry.
In the “saturated” steady state, the total number of firms will not change each period,
giving us the condition that:
M ss = M ss (1 − δ) + M new ⇒ M ss δ = M new
(7.7)
Proposition 1.2: Dynamic Gravity, Heterogenous Firms
Melitz (2003) considers a movement from autarky to free trade, and in the process
derives the steady state equilibrium values. While the dynamics of the Melitz model
are complicated, with a few minor simplifying assumptions we can derive the transition
dynamics. Specifically, assume H=1 as in Melitz, e.g., that there is just one industry
(or, alternatively, that each firm is in its own industry), and that the only fixed costs
are the equity-financed costs of entry. Thus the home entry fixed costs are inclusive of
the costs of beginning production such as from building a factory or buying equipment,
and the foreign entry costs are the costs associated with entering a new market, and so
one must enter the domestic market in order to sell abroad. The per-period overhead
costs in Melitz, for simplicity, will be set to zero – these can be thought of as being
comprised by the variable costs. Otherwise, following Melitz (2003), I assume symmetry
among countries which move from autarky to free trade at year 0. The inclusion of fixed
costs that do not depreciate completely each period give rise to the following “dynamic”
gravity equation at time t, where lagged entry costs figure explicitly.
Xij,t =
t−1
1X
Yj,t−s Yi,t−s
(1 − δ)s
x̃ijt,t−s , t ∈ (0, k)
σ s=0
Yw,t−s fx,t−s
74
(7.8)
Where x̃ijt,t−s is the average sales per firm in period t of the new exporters from period
t-s, which is an average of exports from firms within the cutoff productivities at time t-s1 and t-s, ϕ∗x,t−s−1 and ϕ∗x,t−s (the integral of total exports at each level of productivity
divided by the measure of firms at each level of productivity, which cancels), and where
in the first period after moving from autarky to free trade there is no upper bound on
the integral.
x̃ij,t =
Z ϕ∗
x,t−s−1
ϕ∗x,t−s
xijt,t−s (ϕ)µ(ϕ)dϕ
(7.9)
The economy reaches the new steady state at time k>0, given by:
(1 − δ)k =
fe
fe + nfx (1 − G(ϕ∗x,ss ))
(7.10)
The Melitz (2003) model will only have a separating equilibrium where not all firms
export with relatively large values for the fixed costs of exporting compared to the overhead costs. While exporting is a relative rarity, it is hard to envision large entry costs
arising from trade costs traditionally rendered, unless these entry costs are considered
to be inclusive of costs associated with entering a market for the first time and successfully gaining market share from other firms – generally an expensive, difficult, and rare
undertaking. Relatively large values for the fixed costs of exporting to a specific market
– necessary to match the empirical fact that few firms export – would imply a relatively
long transition according to equation 7.10. Note that including fixed overhead costs
as in Melitz (2003) would carry the same key implication that current trade depends
on historical trade costs as the dynamic gravity equation presented above, although it
would change the pace of convergence. Chaney (2005) studies the dynamics of a Melitz
type model with different simplifying assumptions (for example, he allows firms to stay
in the market but not produce) and, not surprisingly, arrives at a different dynamic
relationship.
If we stipulate a Pareto distribution for the productivities following Melitz, Helpman,
and Yeaple (2004) and Chaney (2008), we can solve for x̃ij,t , and get an expression which
is a direct negative function of both lagged variable trade costs and lagged entry costs.
Proof of Proposition: 1.2 Dynamic Gravity, Heterogenous Firms
We can start from the steady-state labor market clearing condition in autarky:
Z
L=
Z
li di =
(
q(ω)
+ fe )µ(ω)dω
ϕ(ω)
(7.11)
Substituting in solutions for q(ω) from the consumer’s problem, using an aggregate
75
ss which ensures that firm births equal firm
steady-state stability condition Mass = δMa,e
deaths, and then using the CES price markup, we arrive at the mass of firms at each
level of productivity in autarky.
Z
L=
Lp(ω)−σ
Mass
µ(ω)dω
P
Z
=
Mass
R
+
ss
Ma,e
fe µ(ω)dω
−σ ϕσ−1
L( σ−1
σ )
µ(ω)dω + δMass fe
P
=⇒ Mass =
(Note: in the above P =
Z
ω∈Ω p(ω)
1−σ dω
L
σδfe
(7.12)
(7.13)
for simplicity.) With similar logic, we get
an expression for the steady state free trade mass of firms.
=⇒ Mfsst =
L
σδ(fe + fx n(1 − G(ϕ∗x,ss ))
(7.14)
As in Melitz (2003), if any firms export, then the mass of firms at each level of
productivity in the free trade steady state must be strictly less than the mass of firms
in autarky. Since our duration for each period is arbitrary, we can stipulate that this
will take more than one period without loss of generality. Since there are no per-period
overhead costs, no incumbent firm would ever want to leave the market, but firms do die
naturally at rate δ. In the first period after opening to free trade, the aggregate stability
condition will need not hold, so instead we use the labor market clearing condition once
again to determine how many firms enter foreign markets.
Z
L=
R
=⇒ L = R
Mf1t
Lp(ω)−σ
µ(ω)dω + Mf1t nfx (1 − G(ϕ∗x,1 ))
P ϕ(ω)
σ −σ
Mf1t L( σ−1
) ϕ(ω)σ−1 µ(ω)dω
σ 1−σ
Mf1t ( σ−1
)
ϕ(ω)σ−1 µ(ω)dω
=⇒ Mf1t =
+ Mf1t nfx (1 − G(ϕ∗x,1 ))
L
σnfx (1 − G(ϕ∗x,1 )
(7.15)
Where Mf1t (1 − G(ϕ∗x,1 )) gives the total number of firms which will enter n export
markets in the first year after opening to trade. The mass of firms at each level of
productivity will have shrunk at rate δ from the steady state mass in autarky. Total
76
exports to one specific market in the first period after opening up to trade are thus:
Xij,1 = Mf1t (1 − G(ϕ∗x,1 )x̃ij1,1 =
Yi,1 Yj,1
L
x̃ij1,1 =
x̃ij1,1
σnfx
σYw,1 fx
(7.16)
The last equality holds since, for each country, L = R = PQ = Y, and by taking
advantage of the Melitz (2003) assumption of symmetry, which implies that Yw = nYj
for n countries in the world. Average exports per firm to a specific market, x̃ijt,t is as
before: x̃ij1,1 =
R∞
ϕ∗x,1
xij1 (ϕ)µ(ϕ)dϕ. In the second period, a share of all firms, δ, die,
shrinking the mass of firms at each level of productivity, while an equivalent amount
of new investment in various export markets takes place. Equation (7.8) soon follows.
Note that the cutoff productivity level for exporting must fall each period until the
steady state is reached, which concomitantly implies that trade increases fastest just
after opening to trade and more slowly thereafter. While this is straightforward from
the labor market clearing condition, it can also be seen from the zero profit condition
for entry into the export market.
∗(σ−1)
ϕxt
Pt
≥
fx δσ σ (wτjk )σ−1
L(σ − 1)σ−1
(7.17)
The right hand side of this equation includes wages, which are set to unity in Melitz
(2003), variable trade costs τ , and other parameters, which we can hold constant. Since
P will depend positively on the number of firms which shrink each period, and negatively
on the cutoff productivity for exporting, if ϕ∗x were constant than P would fall. If the
cutoff productivity were rising, then P would be falling, which implies that the righthand side of equation (7.17) would not be constant, a contradiction. Intuitively, since
the mass of domestic firms in each country is shrinking each period, this should increase
the amount exporters can sell, which also lowers the barriers for entry.
Proposition 2.1: Dynamic Crowding Out From Government Spending in Autarky
Let Log be the initial level of government employment in autarky. Using the same
assumptions as in Proposition 1.1 (no overhead costs, equity-financed fixed costs, and
there is just one industry), we can solve for the steady-state mass of firms from the labor
market clearing condition.
L − Log
o
= Mss
σδfe
(7.18)
If the labor devoted to government falls, then the equilibrium mass of firms will rise.
However, each period the share δ of firms die out, which means that the net mass of new
77
entrants in the first year after the cut in government spending is:
o
M1e − Mss
δ=
L − L1g
L − Log
Log − L1g
−
=
σfe
σδfe
σfe
(7.19)
Where M1e is the mass of new entrants in the first period after government spending
falls. The total mass of new firms gained to reach the new steady state is:
1
o
Mss
− Mss
=
Log − L1g
L1g − Log
>
σδfe
σfe
(7.20)
If government spending stays at its new level, then we can iterate out to solve for the
mass of firms at t.
Mt =
t
X
o
(1 − δ)t−j Mje + (1 − δ)t Mss
j=1
Mt =
t
X
(1 − δ)t−j
j=1
(L − L0g )
(L − L1g )
+ (1 − δ)t
σfe
σδfe
(7.21)
The new steady state will be reached in the limit. By contrast, if government spending
increases, the mass of firms will shrink at rate δ until the new steady state is reached.
References
Acemoglu, D., D. Autor, D. Dorn, G. H. Hanson and B. Price, “Import Competition and the Great US Employment Sag of the 2000s,” Nber conference working
paper, National Bureau of Economic Research, Inc, 2013.
Alessandria, G. and H. Choi, “Establishment Heterogeneity, Exporter Dynamics,
and the Effects of Trade Liberalization,” Technical Report, 2012.
Autor, D., D. Dorn and G. H. Hanson, “The China Syndrome: The Local Labor Market Effects of Import Competition in the US,” American Economic Review
(Forthcoming).
Baldwin, R., “Hyteresis in Import Prices: The Beachhead Effect,” American Economic
Review 78 (September 1988), 773–85.
———, “Hysteresis in Trade,” Empirical Economics 15 (1990), 127–42.
Baldwin, R. and P. Krugman, “Persistent Trade Effects of Large Exchange Rate
Shocks,” The Quarterly Journal of Economics 104 (November 1989), 635–54.
78
Bartlesman, E. and W. B. Gray, “The NBER Manufacturing Productivity
Database,” (1996).
Belke, A., M. Göcke and M. Günther, “Exchange Rate Bands of Inaction and Play
– Hysteresis in German Exports – Sectoral Evidence for Some OECD Destinations,”
Metroeconomica 64 (2013), 152–179.
Bergin, P., “Should We Worry About the Large U.S. Current Account Deficit?,”
FRBSF Economic Letter (2000).
Bergin, P. R. and C. Y. Lin, “The Dynamic Effects of Currency Union on Trade,”
Journal of International Economics (2012).
Bergin, P. R. and S. M. Sheffrin, “Interest Rates, Exchange Rates and Present
Value Models of the Current Account,” Economic Journal 110 (April 2000), 535–58.
Berman, N., P. Martin and T. Mayer, “How do Different Exporters React to
Exchange Rate Changes?,” The Quarterly Journal of Economics 127 (2012), 437–492.
Bernard, A. B., J. B. Jensen and P. K. Schott, “Survival of the Best Fit: Exposure to Low-Wage Countries and the (Uneven) Growth of U.S. Manufacturing Plants,”
Journal of International Economics 68 (January 2006), 219–237.
Branson, W. H. and J. P. Love, “The Real Exchange Rate and Employment in U.S.
Manufacturing: State and Regional Results,” NBER Working Papers 2435, National
Bureau of Economic Research, Inc, November 1987.
———, “The Real Exchange Rate, Employment, and Output in Manufacturing in the
U.S. and Japan,” NBER Working Papers 2491, National Bureau of Economic Research, Inc, February 1988.
Burstein, A. and M. J. Melitz, “Trade Liberalization and Firm Dynamics,” NBER
Working Papers 16960, National Bureau of Economic Research, Inc, April 2011.
Campa, J. M. and L. S. Goldberg, “Employment Versus Wage Adjustment and the
US Dollar,” Review of Economics and Statistics 83 (2001), 477–489.
Campbell, D. L., “History, Culture, and Trade: A Dynamic Gravity Approach,”
MPRA Paper 24014, University Library of Munich, Germany, July 2010.
———, “Estimating the Impact of Currency Unions on Trade: Solving the Glick and
Rose Puzzle,” The World Economy (2013a).
79
———, “Through the Looking Glass: A WARPed View of Real Exchange Rate History,”
Working paper, 2013b.
Chaney, T., “Distorted Gravity: The Intensive and Extensive Margins of International
Trade,” American Economic Review 98 (September 2008), 1707–21.
Dixit, A., “Entry and Exit Decisions under Uncertainty,” Journal of Political Economy
97 (June 1989a), 620–38.
———, “Hysteresis, Import Penetration, and Exchange Rate Pass-Through,” The Quarterly Journal of Economics 104 (May 1989b), 205–28.
———, “Analytical Approximations in Models of Hysteresis,” Review of Economic Studies 58 (January 1991), 141–51.
———, “Investment and Hysteresis,” Journal of Economic Perspectives 6 (Winter 1992),
107–32.
Dooley, M., D. Folkerts-Landau and P. Garber, “The Revived Bretton Woods
System: The Effects of Periphery Intervention and Reserve Management on Interest
Rates & Exchange Rates in Center Countries,” NBER Working Papers 10332, National
Bureau of Economic Research, Inc, March 2004.
———, “An essay on the revived Bretton Woods system,” Proceedings (2005).
———, “Direct Investment, Rising Real Wages and the Absorption of Excess Labor
in the Periphery,” in G7 Current Account Imbalances: Sustainability and AdjustmentNBER Chapters (National Bureau of Economic Research, Inc, 2007), 103–132.
———, “Bretton Woods II Still Defines The International Monetary System,” Pacific
Economic Review 14 (08 2009), 297–311.
Ebenstein, A., M. McMillan, Y. Zhao and C. Zhang, “Understanding the Role
of China in the “Decline” of US Manufacturing,” Working Paper (2012).
Eichengreen, B. and D. A. Irwin, “The Role of History in Bilateral Trade Flows,” in
The Regionalization of the World Economy (National Bureau of Economic Research,
Inc, 1998), 33–62.
Elsby, M. W. L., B. Hobijn and A. Sahin, “The Decline of the US Labor Share,”
Brookings Papers on Economic Activity (2013).
80
Fahle, S., J. Marquez and C. Thomas, “Measuring US International Relative Prices:
A WARP View of the World,” FRB International Finance Discussion Paper (2008).
Feenstra, R. C., “Globalization and Its Impact on Labour,” wiiw Working Papers 44,
The Vienna Institute for International Economic Studies, wiiw, July 2007.
Feenstra, R. C., R. Inklaar and M. Timmer, “The Next Generation of the Penn
World Table,” Working paper, April 2013.
Feenstra, R. C., H. Ma, J. P. Neary and D. P. Rao, “Who Shrunk China?
Puzzles in the Measurement of Real GDP,” NBER Working Papers 17729, National
Bureau of Economic Research, Inc, January 2012.
Feenstra, R. C. and S.-J. Wei, “Introduction to China’s Growing Role in World
Trade.,” in China’s Growing Role in World Trade. (National Bureau of Economic
Research, Inc, 2010), 1–31.
Forbes, K., “How Do Large Depreciations Affect Firm Performance?,” Working papers
4379-02, Massachusetts Institute of Technology (MIT), Sloan School of Management,
October 2002.
Ghironi, F. and M. J. Melitz, “International Trade and Macroeconomic Dynamics
with Heterogeneous Firms,” The Quarterly Journal of Economics 120 (August 2005),
865–915.
———, “Trade Flow Dynamics with Heterogeneous Firms,” American Economic Review,
Papers and Proceedings 97 (May 2007), 356–361.
Gourinchas, P.-O., “Exchange Rates and Jobs: What Do We Learn From Job
Flows?,” in NBER Macroeconomics Annual 1998, volume 13 (MIT Press, 1999), 153–
222.
Grossman, G. M., “Imports as a Cause of Injury: The Case of the U.S. Steel Industry,”
Journal of International Economics 20 (May 1986), 201–223.
Head, K., T. Mayer and J. Ries, “The Erosion of Colonial Trade Linkages after
Independence,” Journal of International Economics 81 (May 2010), 1–14.
Helpman, E., M. Melitz and Y. Rubinstein, “Estimating Trade Flows: Trading
Partners and Trading Volumes,” The Quarterly Journal of Economics 123 (05 2008),
441–487.
81
Helpman, E., M. J. Melitz and S. R. Yeaple, “Export Versus FDI with Heterogeneous Firms,” American Economic Review 94 (March 2004), 300–316.
Klein, M. W., S. Schuh and R. K. Triest, “Job Creation, Job Destruction, and
International Competition: A Literature Review,” Technical Report, 2002.
———, “Job Creation, Job Destruction, and the Real Exchange Rate,” Journal of International Economics 59 (March 2003), 239–265.
Krishnamurthy, A. and A. Vissing-Jorgensen, “The Effects of Quantitative Easing on Interest Rates: Channels and Implications for Policy,” Technical Report, National Bureau of Economic Research, 2011.
Krugman, P., “Scale Economies, Product Differentiation, and the Pattern of Trade,”
American Economic Review 70 (December 1980), 950–59.
———, “Is the Strong Dollar Sustainable?,” Proceedings (1985), 103–155.
———, “Is the Japan Problem Over?,” NBER Working Papers 1962, National Bureau
of Economic Research, Inc, June 1986.
———, “The Narrow Moving Band, the Dutch Disease, and the Competitive Consequences of Mrs. Thatcher: Notes on Trade in the Presence of Dynamic Scale
Economies,” Journal of Development Economics 27 (October 1987), 41–55.
———, “Deindustrialization, Reindustrialization, and the Real Exchange Rate,” NBER
Working Papers 2586, National Bureau of Economic Research, Inc, May 1988.
———, Geography and Trade, volume 1 of MIT Press Books (The MIT Press, 1992).
Krugman, P. R. and R. E. Baldwin, “The Persistence of the U.S. Trade Deficit,”
Brookings Papers on Economic Activity 18 (1987), 1–56.
Levy, F. and P. Temin, “Inequality and Institutions in 20th Century America,” NBER
Working Papers 13106, National Bureau of Economic Research, Inc, May 2007.
Liu, D. and C. Meissner, “Market Potential and Economic Performance in the Early
20th Century,” Working papers, University of California at Davis, Department of
Economics, 2012.
Love, I. and L. Zicchino, “Financial development and dynamic investment behavior:
Evidence from panel VAR,” The Quarterly Review of Economics and Finance 46
(2006), 190–210.
82
Meese, R. A. and K. Rogoff, “Empirical Exchange Rate Models of the Seventies:
Do They Fit Out of Sample?,” Journal of international economics 14 (1983), 3–24.
Meissner, C. M. and A. M. Taylor, “Losing our Marbles in the New Century? The
Great Rebalancing in Historical Perspective,” NBER Working Papers 12580, National
Bureau of Economic Research, Inc, October 2006.
Melitz, M. J., “The Impact of Trade on Intra-Industry Reallocations and Aggregate
Industry Productivity,” Econometrica 71 (November 2003), 1695–1725.
Moser, C., D. Urban and B. W. di Mauro, “International Competitiveness, Job
Creation and Job Destruction–An Establishment-Level Study of German Job Flows,”
Journal of International Economics 80 (March 2010), 302–317.
Nino, V. D., B. Eichengreen and M. Sbracia, “Real Exchange Rates, Trade, and
Growth: Italy 1861-2011,” Quaderni di storia economica (Economic History Working Papers) 10, Bank of Italy, Economic Research and International Relations Area,
October 2011.
Ottaviano, G. I., G. Peri and G. C. Wright, “Immigration, Offshoring and American Jobs,” NBER Working Papers 16439, National Bureau of Economic Research, Inc,
October 2010.
Pierce, J. R. and P. K. Schott, “The Surprisingly Swift Decline of U.S. Manufacturing Employment,” Working Paper 18655, National Bureau of Economic Research,
December 2012.
Pozzolo, A. F. and F. Nucci, “Exchange Rate, Employment and Hours: What FirmLevel Data Say,” Economics & Statistics Discussion Papers esdp08049, University of
Molise, Dept. SEGeS, October 2008.
Ugai, H., “Effects of the Quantitative Easing Policy: A Survey of Empirical Analyses,”
Monetary and Economic Studies-Bank of Japan 25 (2007), 1.
83