Trade, Employment, and Wages: What Does the Evidence Show?

^Pr^ep^ar^{ed f}^o^r^M^e^m^b^ers^an^d^Com^mⁱ^t^t^ees^of^C^ong^r^e^ss

There is a disconnect between economists and popular opinion on the merits of free trade.
Economic theory concludes that free trade is mutually beneficial because it allows countries to
focus on producing the goods for which they have a comparative advantage. By contrast, many
people believe that free trade is harmful because it destroys American jobs or lowers wages. This
report looks past theory at the empirical evidence to see which side of the argument is supported
by the data.
Economists do not deny that some American workers will lose their jobs because of import
competition. Greater imports will cause gross job loss, but theory predicts the job loss will be
offset by gross job gains among exporters, recipients of foreign capital inflows (in the case of a
trade deficit), and the economy as a whole because of general increases in efficiency caused by
comparative advantage. The result, economists believe, would be no net change in total
employment.
This report uses regression analysis of quarterly national data from 1948 to 2005 (and a sub-
sample from 1980 to 2005) to answer three questions. First, do higher imports systematically
correspond to periods of high unemployment or low employment growth? Second, do higher
trade deficits correspond to high unemployment or low employment growth? Third, do higher
imports or trade deficits correspond to weak growth in average worker compensation?
The findings of this report generally support economists’ views on trade and employment for
several reasons. First, most of the findings on the link between imports and unemployment, or
trade deficits and unemployment, are statistically insignificant. That is, by the accepted standards
of research, it cannot be ruled out that imports and the trade deficit had no effect on
unemployment at all. Second, the regressions generally had low R-squared values, which means
that imports and trade deficits explain very little of the variation in unemployment. Third, when
the relationship between the variables was statistically significant, it was too small to be
economically meaningful and, in some cases, moved in the opposite direction from the
relationship assumed by some free trade opponents. For example, a one percentage point increase
in imports as a share of GDP reduced the unemployment rate by 0.10 percentage points, all else
equal.
The relationship between trade and average worker compensation tended to be statistically
significant, but these regressions also had low R-squared values and were too small to be
economically meaningful. The effect of imports on worker compensation was the opposite of that
predicted by some free trade opponents. Trade deficits, on the other hand, were associated with
low growth in worker compensation, but the effect was small (a one percentage point increase in
the trade deficit as a share of GDP reduced worker compensation by 0.79%-1.17%, all else
equal.) It should be noted that average worker compensation does not address the question of how
trade affects wage inequality among workers of different income levels. It does, however, address
the question of how national income is divided between workers and owners of capital. This
report will not be updated.

Introduc tion ..................................................................................................................................... 1
Evidence on Trade and Jobs............................................................................................................2
Evidence on Trade Deficits and Jobs...............................................................................................6
Evidence on Trade, Trade Deficits, and Workers’ Compensation...................................................9
Conclusion ..................................................................................................................................... 13
Figure 1. Imports and Unemployment, 1948-2005.........................................................................3
Figure 2. Trade Deficits and Unemployment, 1948-2005...............................................................7
Figure 3. Worker Compensation and Trade, 1948-2005.................................................................11
Table 1. Imports and Jobs................................................................................................................4
Table 2. Employment and Trade Deficits........................................................................................9
Table 3. Trade, Trade Deficits, and Worker Compensation...........................................................12
Appendix. A Primer on Econometric Analysis..............................................................................15
Author Contact Information..........................................................................................................15

Economists are nearly unanimous that free trade raises economic well-being overall. For
example, 93% of economists surveyed in the early 1990s generally agreed with the proposition or ¹
agreed with provisos that “tariffs and import quotas usually reduce economic welfare.” The
general public tends to have a different view of trade. In polls taken in June 2005 and around the
time the economists’ survey was published, 48% of the public believed that trade was a threat to ²
the economy, while 44% believed it was an opportunity for economic growth.
The economic case for free trade starts with the principle of comparative advantage, which can be
summed up with the adage “focus on what you do best.” The insight behind comparative
advantage is that even if country A is better than country B at producing everything (called an
“absolute advantage”), there will still exist some goods that country B is less bad at producing,
and if it produces and trades those goods with country A, both countries can consume more than
if each tried to produce all types of goods by itself. This principle applies not just to trade
between countries, but every market transaction that occurs between individuals. In one famous
example, it explains why Michael Jordan does not mow his own lawn. Although his athletic
ability allows him to mow faster than anyone he can hire, he is better off paying someone else to ³
do it and spending the time he saved on basketball or endorsements.
This logic is considered so airtight by economists that little empirical research has been done
lately to try to confirm it. But theory makes little impression on a worker who is told that he or
she has lost his or her job to overseas competition. This report looks at empirical evidence of
what has happened historically to trade and employment. In evaluating this evidence, it is
important to distinguish between the gross effects of trade on employment and the net effects.
Economists would not dispute that trade causes some workers in the U.S. economy to lose their
jobs. But they contend that other jobs are created by trade and the efficiency gains that result from ⁴
trade, so that trade has no net effect on employment. This report analyzes empirical evidence and
finds support for economists’ contention that trade has no net quantitative effect on employment.
(This report focuses on the nation’s overall welfare, and does not address the issue of trade’s
effect on gross employment flows. That is, it does not analyze whether workers in certain regions, ⁵
income cohorts, or sectors of the economy benefit or suffer from trade more than others.)
Congress may find this evidence useful in considering any future free trade agreements that result ⁶
from ongoing bilateral and multilateral negotiations.

¹^Rⁱ^c^h^a^r^{d A}^l^s^t^on,^J^.^R^.^K^e^a^r^{l, a}ⁿ^d^Mic^h^a^e^l^Va^ug^hn,^“^I^s^T^h^e^r^e^a^C^oⁿ^s^eⁿ^s^u^s^A^m^ong^Ec^onom^is^ts^{in the}¹⁹⁹^0s^?^”^Am^e^r^ic^an
^Ec^onom^ic^Re^vⁱ^e^w^{, Ma}^y^1992,^p.²⁰³^.
²^Denⁿⁱ^{s Jaco}^b^e^,^“Mo^r^{e Am}^eri^can^{s S}^{ee T}^h^reat^,^No^t^Op^po^rt^uⁿⁱ^t^y^,ⁱⁿ^F^o^rei^gⁿ^T^r^ad^e,^{” G}^a^l^l^u^p^P^o^l^l^{News S}^e^rvi^ce,^A^u^g.²^,
²⁰⁰^{5. B}^e^tw^e^eⁿ¹⁹⁹²^aⁿ^d²⁰⁰⁵^{, G}^a^{llup a}^s^k^e^{d th}^is^q^u^e^s^tion^eⁱ^g^h^{t ti}^m^e^s^.^D^u^rⁱ^ng^tha^t^tⁱ^m^e^{, the}^s^h^a^r^e^o^f^r^e^s^ponde^nts^w^ho
^b^e^lⁱ^e^ved^t^h^at^t^r^ad^{e w}^a^{s a t}^h^reat^vari^ed^b^e^t^w^een³⁵^%^an^{d 48}^%.
³^{For a}^f^u^{ll e}^x^p^o^sitioⁿ^{on t}^h^e^t^h^e^o^r^y^o^f^tra^d^e^a^{nd c}^o^m^p^a^r^a^tiv^e^a^d^v^a^nt^a^g^e^,^se^e^{CRS Re}^{port RL}³²⁰^59,^T^r^ade^{, Tr}^ade
^Bar^rⁱ^e^r^s^,^an^{d Tr}^ade^D^e^fic^its^:^I^m^p^lic^atio^ns^f^o^r^U^.^S.^Ec^o^nom^ic^W^e^lf^ar^e^{, by}^Craⁱ^g^K^.^Elw^e^ll.
⁴^T^h^e^c^onc^e^p^{t of}^g^r^os^s^aⁿ^d^ne^{t j}^o^b^los^s^is^dis^c^us^s^e^{d m}^o^r^e^f^u^lly^{in C}^R^S^R^e^por^t^R^L³²¹⁹⁴^,^Jo^{b L}^o^{ss: Cau}^s^{es a}ⁿ^d^P^o^lⁱ^cy
^I^m^plic^ati^ons^,^by^Ma^r^c^L^a^bonte^.
⁵^F^o^{r sect}^o^r^al^evi^d^en^ce,^{see CRS}^C^u^rren^t^L^e^gi^sl^atⁱ^v^{e Issu}^e^U^.^{S. I}ⁿ^d^u^s^tr^ie^s^and^Tr^a^d^e^,^ht^tp:/^/be^t^a^.^c^r^s.g^o^v^/^c^li/
^c^li.a^s^px^?^P^RDS_CL^I_IT^EM_ID=⁵⁴⁹^{. Se}^e^a^l^{so A}ⁿ^a^Re^v^e^ng^a^,^“^E^x^porting^Jo^bs?^T^h^e^Im^pa^c^t^of^I^m^{port Com}^p^e^tition^oⁿ
^Em^plo^y^m^e^{nt a}^{nd W}^a^g^e^s^{in U}^.^S.^Ma^nuf^a^c^t^ur^ing^,^”^Q^uar^te^r^l^y^J^o^urⁿ^a^{l o}^f^Ec^o^nom^ic^s^,^vo^l.¹⁰^7,^no^.¹^,^F^e^b^.¹⁹⁹²^,^p^.²⁵⁵^.
⁶^{For t}^h^e^c^u^rreⁿ^t^s^t^a^t^us^of^m^u^ltila^tera^{l ne}^g^o^tia^tio^ns^{, s}^e^e^{CRS Re}^p^o^rt^RL³²⁰⁶^0,^W^o^r^l^{d Tr}^a^d^e^O^r^g^aniz^a^tion^N^e^got^iati^ons^:
^T^h^{e Doh}^a^Devel^o^p^m^ent^A^g^en^da^,^by^I^a^{n F. Fe}^r^g^us^s^on.^Se^e^a^l^s^o^C^R^{S R}^e^por^t^R^L³¹⁹³²^,^T^r^ad^{e A}^g^reemen^t^s^{: Imp}^a^c^t^oⁿ
⁽^c^onti^nue^d.^..)

Whether it is desirable to undertake an activity that creates winners and losers is a value
judgement that cannot be answered by economic theory. (However, it is useful to note that all
market activity, including competition among domestic firms and technological innovation,
creates winners and losers.) Most economists argue that, in principle, nobody need be made
worse off from trade since the gains from trade allow the winners to compensate the losers and
still be better off themselves. Of course, effective compensation is difficult to design in reality.
The Trade Adjustment Assistance Program is one real world example of how such compensation ⁷
can be structured. Economic theory cannot address the question of how much assistance those
workers who are harmed by trade deserve to receive; that is a social issue.

The argument that imports cause net loss in employment in the economy overall can be evaluated
at the simplest level by looking at the data in Figure 1. Imports have been growing steadily as a
share of GDP over the past 50 plus years, but unemployment has not. Imports and unemployment
were both very low in the 1950s (the data in the bottom left-hand corner of the chart). Trade
opponents might blame imports for the rise in unemployment in the 1970s and 1980s. But since
the 1990s, unemployment has steadily fallen while imports have continued to rise (the data in the
upper part of the chart.) If imports are to blame for job loss, why didn’t the relationship hold over
time?

^(...c^on^tin^ue^d)
^the^U^.^S.^Ec^o^nom^y^,^b^y^Ja^{mes K.}^J^acksoⁿ^.
⁷^Se^e^C^R^S^R^e^por^{t R}^S²²⁷¹^8,^Tr^ad^e^Adjus^t^m^e^nt^As^sⁱ^s^t^aⁿ^c^e^for^W^o^r^k^e^r^s⁽^T^AA)^{and A}^lte^r^nat^iv^e^Tr^ade^Adjus^t^m^eⁿ^t
^Assistaⁿ^{ce fo}^{r Ol}^d^e^{r Wo}^{rkers (}^A^T^AA)^{, by}^J^ohn^J^.^T^o^p^o^le^s^k^i.

Figure 1. Imports and Unemployment, 1948-2005
^Source:^Burea^{u of La}^{bor Statistics and Bur}^{eau of}^{Economic Analy}^sis
^Note:^{Imports are meas}^{ured as}^{a share of}^{GDP. Each data}^{point r}^eprese^{nts a}^quart^{er b}^etween¹⁹⁴⁸^and²⁰^05.
Simple graphing cannot identify the relationship between the two variables with any quantitative
precision. To do that, this report uses the standard tool of empirical research in economics,
regression analysis (see the appendix for more details). Regression analysis relies on finding a
large enough sample and looking at variations in employment levels and trade levels across each
unit of the sample, while holding other causes of variation constant, to estimate how much ⁸
employment will change when trade levels change. The units making up the sample will be the
U.S. economy as a whole over time, measured quarterly. If the argument made by some trade
opponents were correct, it would be found that in quarters when imports were high,
unemployment would be high. If the relationship were strong enough, it would be said to
statistically significant, which means that by commonly accepted standards of research, the
relationship is strong enough to rule out that it was caused by random chance. Alternatively, if
economists were correct that trade has no net effect on overall employment, then a statistically
insignificant relationship between the two variables would be expected. Economic time series
variables often have a rising trend over time. Therefore, the effect of the trend in pertinent

⁸^Ec^on^om^is^ts^w^{ould c}^o^ns^ide^r^tr^a^d^e^a^{nd e}^m^ploy^m^e^{nt to be}^tw^{o e}ⁿ^d^o^g^e^nous^v^a^rⁱ^a^b^le^sⁱⁿ^a^la^r^g^e^r^sy^s^t^em^of^e^qua^tio^ns
^de^s^c^rⁱ^bing^e^c^oⁿ^o^mⁱ^c^a^c^tiv^it^y^{on the}^w^hole^.^T^h^e^r^e^f^or^e^,^one^c^oul^dⁿ^o^{t c}^o^ns^ide^r^e^m^ploy^m^e^{nt to}^be^sⁱ^m^p^ly^c^a^u^s^e^{d by}^tr^a^d^e
^sinc^e^the^y^a^r^e^jointly^de^te^r^m^ine^d^{. Solv}ⁱⁿ^g^t^h^is^s^y^s^t^em^o^f^e^qua^tions^w^{ould be}ⁱⁿ^c^o^m^p^r^e^he^ns^ible^t^o^la^y^r^e^a^d^e^r^s^a^{nd d}^o^e^s
^not^dire^c^tly^a^ddre^{ss the}^c^onc^e^r^ns^of^tr^a^d^e^op^po^ne^{nts, s}^o^the^m^e^thod^w^{ill not}^be^e^m^ploy^e^d^he^re^{. T}^e^c^hnic^a^lly^{, the}^re^sults
ⁱⁿ^t^hⁱ^s^repo^rt^can^b^e^coⁿ^sⁱ^d^ered^t^o^sh^o^w^w^h^et^h^e^{r p}^e^ri^o^d^s^o^f^hⁱ^ghⁱ^m^p^o^r^t^s^o^r^t^r^ad^e^d^e^fⁱ^ci^t^s^{are s}^y^st^e^m^a^tⁱ^cal^l^y^a^sso^ci^at^ed
^w^{ith pe}^rⁱ^ods^of^hig^h^une^m^p^loy^m^{ent or}^low^w^o^r^k^e^r^c^o^m^p^eⁿ^s^a^tion.

variables used in this report has been eliminated by “differencing” the data (looking at changes in ⁹
values, rather than the values themselves) so that spurious correlation does not occur.
Using a sample of all available data, spanning from 1948:2 to 2005:4, Table 1 shows the effect of
trade, measured as imports as a share of GDP, on the unemployment rate. In a simple bivariate
regression between unemployment and imports using the ordinary least squares method, shown in
the first row of the table, the relationship is not significantly different from zero and is in the
opposite direction of what some trade opponents claim. Taken literally, these results (called an
estimate of a beta coefficient) suggest that when imports rise by 1% of GDP, unemployment falls
by 0.05 percentage points. The results are statistically insignificant, however, despite a relatively
large sample size, which means the variables are so weakly related that zero effect cannot be
ruled out. The R-squared measure for this regression is zero, which means that the independent ¹⁰
variables (in this case, just imports) explain none of the movements in unemployment.
For these estimates to be valid, the relationship between trade and employment must not be
fundamentally altered over time. Some might argue that looking at a more recent period would
give a better idea of the effect today since the economy has changed over time. For example,
economic variables such as employment and imports are less volatile now than they were several
decades ago, and the exchange rate changed from a fixed regime to a floating regime in 1971.
Therefore, all of the regressions in this report are also estimated over a sub-sample of 1980-2005.
Over that sub-sample, the estimated effect of imports on unemployment is larger (the beta
coefficient is now 0.69), but still statistically insignificant.
Table 1. Imports and Jobs
^{(Independent variables:}^impor^t^s/GD^{P, re}^{cession when noted)}
^Dependent^Time^Beta^Statis^tical^R-^{Model Type}^Dummy^{Variable for}
^Variable^Period^Estimate^Significance^Squared^Recession?
¹⁹⁴^8:2-^{-0.05 No}^0.00^unem^ployment^{least square}^s^no
²⁰⁰^5:4^rate
¹⁹⁸^0:1-^{-0.69 No}^0.01
²⁰⁰^5:4
¹⁹⁴^8:2-^0.30^{At 1% level}^0.04^employment-^{least square}^s^no
²⁰⁰^5:4^{population ratio}
¹⁹⁸^0:1-^0.42^{At 1% level}^0.12
²⁰⁰^5:4
¹⁹⁴^8:2-^-0.10^{At 5% level}^0.96^unem^ployment^{autoregression}^yes
²⁰⁰^5:4^rate^{(w/ 2 lags)}
¹⁹⁸^0:1-^{-0.27 No}^0.98
²⁰⁰^5:4

⁹^F^o^{r exam}^p^l^e,^t^h^{e avera}^g^{e h}^eⁱ^g^h^t^o^f^Ameri^can^{s an}^d^t^h^{e U.}^S^.^st^o^c^{k market}^ar^{e t}^w^o^vari^ab^l^e^{s t}^h^at^h^a^v^e^t^r^en^d^e^d^u^p^w^ard^s
^o^v^{er t}ⁱ^m^e.^Aⁿ^aⁱ^v^{e co}^rrel^a^tⁱ^oⁿ^w^o^u^l^d^cl^ai^m^t^h^at^h^eⁱ^g^h^t^cau^{ses st}^o^c^{k p}^rⁱ^ces.^Bu^tⁱ^t^w^o^u^l^d^b^e^exp^ect^ed^t^h^atⁱ^f^ch^an^{ges i}ⁿ
^h^eⁱ^g^h^t^an^{d ch}^an^{ges i}ⁿ^t^h^{e st}^o^c^{k m}^a^rket^w^e^{re co}^m^p^ared^,^co^rrel^a^tⁱ^oⁿ^w^o^u^l^d^b^r^{eak d}^o^wⁿ^.
¹⁰^T^h^e^R^-^s^qua^r^e^{d m}^e^a^s^ur^e^c^aⁿ^be^a^s^low^a^s^z^e^r^{o a}^{nd a}^s^hig^h^a^s^one^{. I}^f^the^R^-^s^qua^r^e^{d w}^a^s^one^{, it w}^oul^d^m^e^aⁿ^tha^t^t^h^e
^inde^pe^nde^nt^v^a^rⁱ^a^b^le^s^c^oul^{d e}^x^plaⁱ^{n a}^ll^of^the^m^o^v^e^m^e^{nt in t}^h^e^de^p^e^nde^{nt v}^a^rⁱ^a^b^le^.

^Dependent^Time^Beta^Statis^tical^R-^{Model Type}^Dummy^{Variable for}
^Variable^Period^Estimate^Significance^Squared^Recession?
¹⁹⁴^8:2-^{0.24 No}^0.17^unem^ployment^{distributed lag}^yes
²⁰⁰^5:4^rate^model
¹⁹⁸^0:1-^{0.04 No}^0.22^{(w/ 2 lags)}
²⁰⁰^5:4
^Source:^{CRS calculations}
^Notes:^{Employment-population}^{ratio and import-GDP ratio have}^{been differ}^{enced. Beta estimate f}^{or distributed}
^{lag model is the sum of the two l}^{agged beta coefficients.}
To test the robustness of the results, it is useful to look at different measurements of labor markets
and see if similar results are found. When employment (as a fraction of the total adult population,
to control for population growth) is substituted for the unemployment rate, the effect is highly ¹¹
statistically significant—but causes employment to rise, not fall. Now, an increase in imports
equal to 1% of GDP causes a 0.30 percentage point rise in the employment-population ratio using
the entire sample, and this result is statistically significant at the 1% level. The R-squared for this
regression is also extremely low (0.04), so the regression can only explain 4% of the change in
employment over time. Similar results are found over the more recent sub-period.
The regressions presented so far are simplistic, and a few modifications can make them more
realistic, and hopefully more accurate. First, the regressions presented above assumed that
nothing affects unemployment except imports. In reality, there are many other variables that one
would expect to affect unemployment. While it is beyond the scope of this report to consider
them all, a very simple one can be added—a dummy variable for whether or not the economy is
in a recession. If it is accepted that higher imports do not tend to cause recessions, then it can be
controlled for the predictable effect of a recession on unemployment to analyze whether the
remaining variation in the unemployment rate can be explained by changes in imports. This small
change can make a large difference. For example, when a recession variable is added, the effect
of imports on the employment-population ratio is no longer statistically significant and the effect
is only one-third as large (for brevity, these results are not presented in Table 1).
Second, regression analysis is based on the assumption that the units in the sample are
independent of each other—one unit is not systematically influenced by another. For many time
series, this is not the case. Two different methods are used next to correct for this fact. Because
labor markets adjust slowly, the unemployment rate this quarter can help predict what the
unemployment rate will be in the next quarter. This can be controlled for using the first method,
called autoregression. Empirically, if it can be controlled for the effects that the unemployment
rate had over the last two quarters on unemployment in the current quarter (empirical tests
suggests that the statistically significant effects come within two quarters), it can then be
evaluated whether the remaining variation in unemployment is caused by imports. As seen in
Table 1, when this is done, the effect of imports on unemployment doubles and becomes ¹²
significant at the 5% level. (The relationship becomes statistically insignificant when confined

¹¹^Sigⁿ^if^ic^aⁿ^c^e^a^t^the^1%^le^v^e^{l m}^e^ans^tha^tⁱⁿ⁹⁹^ou^{t of}¹^{00 c}^a^s^e^s^the^e^f^f^e^c^t^ofⁱ^m^por^ts^on^e^m^ploy^m^e^{nt w}^{ould n}^o^t^be
^zero^.
¹²^A^lthoug^{h t}^h^e^r^e^s^u^lts^a^r^eⁿ^o^t^pr^e^s^eⁿ^te^{d f}^o^r^the^s^a^k^e^of^br^e^v^ity^{, the}^e^f^f^e^c^t^ofⁱ^m^por^ts^is^s^t^a^tis^tic^a^lly^insⁱ^gⁿ^if^ic^aⁿ^{t if}^the
^em^ploy^m^e^nt-^p^o^pula^tⁱ^on^r^a^{tio is}^u^s^e^d^a^s^a^de^pe^nde^{nt v}^a^rⁱ^a^b^le^ins^t^e^a^{d of}^the^une^m^p^loy^m^eⁿ^{t r}^a^te^.^A^l^s^o^{, the}^r^e^s^u^lts
^p^r^esen^t^e^dⁱⁿ^t^h^e^t^a^b^l^{e i}ⁿ^cl^u^d^e^t^h^e^recessi^oⁿ^d^u^mmy^v^a^ri^ab^l^e^an^d^t^w^o^l^a^gs,^b^u^t^t^h^{e resul}^t^s^{are no}^t^qu^alⁱ^t^a^tⁱ^v^el^y^dⁱ^ff^eren^t
⁽^c^onti^nue^d.^..)

to the 1980-2005 period, however.) Most likely, the regressions are showing that higher imports
are associated with lower unemployment because causation runs in the opposite direction—when
the economy is booming and unemployment is low, import demand rises. The R-squared for this
regression is very high, 0.96, which illustrates that past unemployment rates are a much better
determinant of current unemployment than trade.
Also imagine that the effects of imports on unemployment do not all occur immediately. Perhaps,
because labor markets adjust slowly, the effects of a change in imports on unemployment take a
few quarters before they are fully felt. Using the second method, a distributed lag model, this
hypothesis can be investigated. Measuring the effect of imports on unemployment over the next
two quarters leads to higher imports causing unemployment to rise, but the results are not ¹³
statistically significant. The R-squared is low (0.16).

Other opponents of trade concede that balanced trade can be beneficial, but argue that trade is
harmful to jobs when it results in a trade deficit. They reason that because Americans are buying
imports, jobs are being destroyed, but because foreigners are not buying American products in
exchange, no American jobs are being created. Economic theory suggests that trade deficits create
jobs in the interest-sensitive sectors of the economy because a trade deficit is matched by foreign
capital inflows that reduce domestic interest rates. Lower interest rates stimulate spending on U.S.
capital investment, residential investment, and interest-sensitive consumer durables. Again, the
net effects of the jobs created and lost from a trade deficit would be expected to roughly balance,
assuming the change in the trade deficit was smooth. Economic theory also suggests that a trade
deficit is more likely to occur when overall spending in the economy is strong, since this is when
the demand for borrowing would be greater. If the trade deficit is caused by strong aggregate ¹⁴
spending (demand), by definition, it would be associated with high employment.
Turning to the data in Figure 2, it can be seen at first glance that the largest trade deficits during
the past 5½ decades have been associated with lower-than-average unemployment rates. Most of
these large trade deficits were recent. Most of the quarters in which trade deficits exceeded 2.5%
of GDP have occurred since 1999; during that period the unemployment rate has varied from
3.9% (the lowest rate in 30 years) to 6.1%. The highest unemployment rates over the sample were
mostly associated with trade deficits much smaller than the recent ones, and also included a few
periods with small trade surpluses.

^(...c^on^tin^ue^d)
^if^the^re^c^e^{ssion v}^a^ria^b^le^{is om}^itte^{d, a}^li^ne^a^r^treⁿ^d^{is c}^o^ntrol^l^e^d^f^o^r,^{or f}^{our la}^g^s^a^r^e^use^dⁱⁿ^ste^a^d^of^two.
¹³^T^h^e^re^la^tionshi^{p is t}^h^e^o^p^p^o^site^,^{but st}^{ill sta}^tistic^a^lly^insigⁿ^if^ic^aⁿ^{t, if}^the^e^m^ploy^m^e^nt-po^pula^tⁱ^oⁿ^ra^ti^{o is use}^d^a^s^a
^d^e^p^eⁿ^d^eⁿ^t^v^a^riab^{le in}^stead^o^f^th^e^uⁿ^e^m^p^lo^y^m^en^{t rate. It is}^alsoⁱⁿ^si^gⁿ^if^ican^{t if}^th^{e reces}^sioⁿ^v^a^riab^le^{is o}^m^itted^.
^Di^ff^er^en^t^l^a^{g l}^eⁿ^g^t^h^{s w}^e^{re u}^s^ed^,^an^d^t^h^{e rel}^a^tⁱ^oⁿ^s^hⁱ^p^w^a^{s i}ⁿ^si^gnⁱ^fⁱ^c^an^tⁱⁿ^each^case.
¹⁴^For^m^o^r^e^inf^o^r^m^a^{tion, s}^e^e^C^R^S^R^e^por^{t R}^L³²⁰⁵^9,^Tr^ade^,^Tr^a^d^e^B^a^r^rⁱ^e^r^s^,^an^{d Tr}^a^d^e^D^e^fic^its^:^I^m^plⁱ^c^atio^ns^for^U^.^S.
^Ec^onom^ic^W^e^lfar^e^,^by^Craⁱ^g^K^.^Elw^e^ll.

Figure 2. Trade Deficits and Unemployment, 1948-2005
^Source:^Burea^{u of La}^{bor Statistics and Bur}^{eau of}^{Economic Analy}^sis
^Note:^{Trade Defic}^{its are measur}^{ed as a share}^{of GDP. Each}^data^{point rep}^resents^{a q}^uarter^betwe^{en 1}⁹⁴⁸^and
²⁰⁰^5.

Table 2 presents regression results making the same adjustments to the data described in the
previous section, and performing the same regressions using the trade deficit as an independent
variable instead of imports.
In the simple least-squares, bivariate regression between the unemployment rate and the trade
deficit, the results were not statistically significant and the opposite direction from what some
trade opponents claim (a one percentage point increase in the trade deficit-GDP ratio led to a 0.42
percentage point decline in the unemployment rate) over the full sample. The R-squared was
zero: the trade deficit could not explain any of the variation in the unemployment rate. When the
employment-population ratio is substituted for the unemployment rate, the results are weakly
significant and small, a one percentage point increase in the trade deficit-GDP ratio led to a 0.17
percentage point decline in the employment-population ratio. Similar results are found for
unemployment and the employment-population ratio when the more recent subsample is used.

Table 2. Employment and Trade Deficits
^{(Independent variables: tr}^ade^deficit-^G^D^P^ratio^,^recess^{ion when noted)}
^Dependent^Time^Beta^Significance^R-^{Model Type}^Dummy^{Variable for}
^Variable^Period^Estimate^Squared^Recession?
¹⁹⁴^8:2-^{-0.42 No}^0.00^unem^ployment^{least square}^s^no
²⁰⁰^5:4^rate
¹⁹⁸^0:1-^{-0.28 No}^0.00
²⁰⁰^5:4
¹⁹⁴^8:2-^-0.17^{At 10%}^level^0.02^employment-^{least square}^s^no
²⁰⁰^5:4^{population ratio}
¹⁹⁸^0:1-^-0.20^{At 10%}^level^0.03
²⁰⁰^5:4
¹⁹⁴^8:2-^0.10^{At 10%}^level^0.96^unem^ployment^{autoregression}^yes
²⁰⁰^5:4^rate^{(w/ 2 lags)}
¹⁹⁸^0:1-^{0.64 No}^0.87
²⁰⁰^5:4
¹⁹⁴^8:2-^{-0.25 No}^0.17^unem^ployment^{distributed lag}^yes
²⁰⁰^5:4^rate^{model (w/ 2 lags)}
¹⁹⁸^0:1-^{-0.39 No}^0.24
²⁰⁰^5:4
^Source:^{CRS calculations}
^Notes:^{Employment-population}^{ratio and trade deficit-GDP ratio}^{have be}^{en differe}^{nced. Beta estim}^{ate for}
^{distributed lag model is the sum}^{of the two lagged betas.}
In an autoregression with a two-period lag that also included a dummy variable for whether the
economy was in a recession, the trade deficit was found to increase the unemployment rate, but
this result was very small and only weakly significant. The effect becomes statistically
insignificant over the more recent subsample. In a distributed lag model, the result was very small
and statistically insignificant, and the independent variables explained only 17% of the variation ¹⁵
in the unemployment rate over the full sample.

Some opponents of free trade concede that trade does not lower net employment, but argue that it
should be opposed on the grounds that it reduces workers’ compensation (wages and benefits).
Economic theory suggests that trade’s effects on compensation would depend on the unit labor
costs of U.S. workers compared to foreigners in each industry. Unit labor costs are a measure that
compares a worker’s productivity to his compensation. After opening up to trade, imports would
be expected to rise in industries where foreign unit labor costs are lower than in the United States,
and U.S. compensation in those industries is likely to be adversely affected. In industries with

¹⁵^T^h^e^re^la^tionshi^{p w}^a^{s a}^l^{so sta}^tistic^a^l^l^y^insigⁿ^if^ic^aⁿ^{t w}^h^eⁿ^othe^{r la}^g^le^ng^{ths a}^r^e^use^d^.

relatively low U.S. unit labor costs, production and compensation would rise. Although
compensation in the United States is high relative to the world average, unit labor-costs are
unlikely to be prohibitively high across the economy for two reasons. First, U.S. labor
productivity is also relatively high. Second, most U.S. trade is with other wealthy nations. But for
the remaining trade with low income nations, it would be expected that low income nations would
have low unit labor costs for labor-intensive industries and high unit labor costs for skill-intensive
or capital-intensive industries. This suggests that the compensation of high-skilled U.S. workers
would rise with trade and the compensation of low-skilled U.S. workers would fall (or rise more
slowly), so that unit labor costs across countries moved closer together. On average, the effect
would be theoretically ambiguous, and cannot be predicted without empirical evidence.
It is important to note that average compensation includes both high-wage workers and low-wage
workers. An increase in average compensation may mask divergent experiences for high-wage
and low-wage workers (for example, if one group’s wages rose while the other’s fell, the changes ¹⁶
would cancel each other out). But average compensation can address another argument: that
capital owners benefit from trade at the expense of workers because trade holds down wages so
that profits flow to owners instead of workers. At least one standard trade model (the Hecksher-
Olin model) used by economists, with its prediction that trade will make wage and capital income
equalize across countries, suggests that this is possible. But some economists argue that this ¹⁷
model makes unrealistic simplifications that would prevent this from happening in reality. For
example, income may already be equalized because of capital mobility.
As a first step, this report looks at what has happened to average real workers’ compensation
(wages and benefits, adjusted for inflation) as trade has increased. As can be seen in Figure 3,
compensation rose in most, but not all quarters. There is no clear pattern, but on balance more of
the quarters where compensation fell occurred when imports were relatively low.

¹⁶^T^hⁱ^s^evi^d^en^{ce i}^s^exa^mⁱⁿ^edⁱⁿ^CR^S^Rep^o^r^t⁹⁸^-44¹^,^{Is Gl}^ob^a^lⁱ^z^a^tⁱ^o^{n t}^h^{e F}^o^{rce B}^e^hⁱ^nd^R^ecen^t^P^o^o^r^U.^S^.^Wag^e
^Pe^r^f^or^m^anc^e^?^:^Aⁿ^A^naly^s^is^,^by^Craⁱ^g^{K. Elw}^e^{ll. Se}^e^a^l^{so W}^illia^m^Cline^,^“^T^ra^de^a^{nd I}ⁿ^c^o^m^e^Distributio^{n: T}^h^e^De^ba^te
^a^{nd N}^e^w^Ev^ideⁿ^c^e^,^”^Iⁿ^te^rⁿ^a^tio^na^l^Iⁿ^s^tit^ute^f^o^r^Ec^on^om^ic^s^,^polic^y^br^ie^f^99-^{7, Se}^p.¹⁹^99.
¹⁷^F^o^{r exam}^p^l^e,^{see Ja}^gdⁱ^s^h^Bh^agwatⁱ^an^d^M^a^rviⁿ^Ko^st^ers,^ed^s.^,^Tr^a^d^e^an^{d W}^a^ge^s^{, A}^m^e^r^ic^aⁿ^Ec^onom^ic^s^Ins^titute^Pre^s^s
⁽^W^a^s^hing^{ton, D}^.^C^.^{: 1}⁹⁹⁴⁾^,^c^h^{. 2}^.

Figure 3. Worker Compensation and Trade, 1948-2005
^Source:^Burea^{u of La}^{bor Statistics and Bur}^{eau of}^{Economic Analy}^sis
^Note:^{Imports are meas}^{ured as}^{a share of}^{GDP. Each data}^{point r}^eprese^{nts a}^quart^{er b}^etween¹⁹⁴⁸^and²⁰^05.
Table 3 presents the results of the same regressions performed in the last two sections using
percentage change in worker compensation per capita as a dependent variable. The least-squares,
bivariate regression produces results that are highly statistically significant, but relatively small
and the opposite of the effect predicted by some free trade opponents. A one percentage point ¹⁸
increase in imports as a share of GDP increases worker compensation by 1.19%. Only 6% of
the variation in compensation can be explained by imports. Both the autoregressive method and
the distributed lag model reduce the effect by three-fourths and make it statistically insignificant
(although the autoregression is weakly significant and the distributed lag is statistically significant ¹⁹
at the 5% level over the more recent subsample). Only the autoregression had an R-squared
value that was not negligible.

¹⁸^If^{a d}^u^mmy^vari^ab^l^e^f^o^{r recessi}^oⁿⁱ^s^add^e^d^,^t^h^{e ef}^fect^o^fⁱ^m^p^o^r^t^sⁱ^s^red^u^ced^b^y^h^a^l^f^an^d^b^eco^m^e^{s o}ⁿ^l^y^margiⁿ^al^l^y
^sigⁿ^if^ic^aⁿ^t.
¹⁹^Sⁱ^{x l}^a^{gs are u}^s^ed^f^o^{r t}^h^{e au}^t^o^regressi^ve^m^o^d^e^l^b^ecau^{se t}^h^{e f}ⁱ^rst^,^se^c^ond,^fⁱ^f^t^{h, a}^{nd s}ⁱ^x^t^h^la^g^a^r^e^hig^h^l^y^s^t^a^tis^tic^a^l^l^y
^si^gnⁱ^fⁱ^can^t^.^T^w^o^l^a^{gs are u}^s^ed^f^o^r^t^h^{e d}ⁱ^st^ri^b^u^t^e^d^l^a^{g m}^o^d^e^l^b^ecau^{se t}^h^{e recessi}^oⁿ^vari^ab^l^eⁱ^s^hⁱ^ghl^y^si^gnⁱ^fⁱ^can^tⁱⁿ^t^hⁱ^s
^s^p^e^cⁱ^fⁱ^ca^{tion. A}^num^be^r^of^dif^f^e^r^e^{nt la}^g^s^w^e^r^e^tr^ie^{d, a}^{nd t}^h^e^im^por^ts^v^a^rⁱ^a^b^le^w^a^s^sⁱ^gⁿ^if^icaⁿ^{t in}^no^ne^of^the^m^{. I}^f^the
^r^e^c^e^s^sⁱ^{on dum}^m^y^va^rⁱ^a^b^le^is^r^e^m^o^v^e^{d f}^r^om^the^a^u^tor^e^g^r^e^s^sⁱ^on,^the^im^{ports v}^a^ria^b^le^{is sig}ⁿ^if^ic^aⁿ^{t a}^t^the^5%^le^v^e^{l, b}^u^t
^ha^s^the^o^p^p^o^s^ite^e^f^f^e^c^t^f^r^o^m^w^h^a^t^s^o^m^e^tr^a^d^e^oppo^ne^nts^pr^e^d^ic^t.

Table 3. Trade, Trade Deficits, and Worker Compensation
^{(Dependent vari}^a^{ble: percent change in worker}^compensati^o^{n per capita)}
^Independent^Time^Beta^Statis^tical^R-^{Model Type}^Dummy^{Variable for}
^Variable^Period^Estimate^Significance^Squared^Recession?
¹⁹⁴^8:2-^1.19^{At 1% level}^0.06^imports-GDP^{least square}^s^no
²⁰⁰^5:4^ratio
¹⁹⁸^0:1-^1.58^{At 1% level}^0.15
²⁰⁰^5:4
¹⁹⁴^8:2-^-0.79^{At 1% level}^0.03^{trade deficit-}^{least square}^s^no
²⁰⁰^5:4^{GDP ratio}
¹⁹⁸^0:1-^-1.17^{At 1% level}^0.09
²⁰⁰^5:4
¹⁹⁴^8:2-^{0.33 No}^0.39^{Autoregression}⁽⁶^imports-GDP^yes
²⁰⁰^5:4^lags)^ratio
¹⁹⁸^0:1-^0.81^{At 10%}^level^0.31^{autoregression (2}
²⁰⁰^5:4^lags)
¹⁹⁴^8:2-^{-0.14 No}^0.39^{autoregression}⁽⁶^{trade deficit-}^yes
²⁰⁰^5:4^lags)^{GDP ratio}
¹⁹⁸^0:1-^-0.85^{At 1% level}^0.33^{autoregression (2}
²⁰⁰^5:4^lags)
¹⁹⁴^8:2-^{0.22 No}^0.06^imports-GDP^{distributed lag}^yes
²⁰⁰^5:4^{model (2 lags)}
¹⁹⁸^0:1-^0.98^{At 5% level}^0.10
²⁰⁰^5:4
¹⁹⁴^8:2-^-0.96^{At 1% level}^0.10^{trade deficit-}^{distributed lag}^yes
²⁰⁰^5:4^GDP^{model (2 lags)}
¹⁹⁸^0:1-^-0.95^{At 1% level}^0.11
²⁰⁰^5:4
^Source:^{CRS calculations based}^{on data from Burea}^{u of La}^{bor Statistics, Bureau of Economic Analy}^{sis, and}
^{National Bureau of}^{Economic Research}
^Notes:^{Data for import-GDP ratio and trade deficit-GDP ratio ha}^{s been}^differen^{ced. Beta estimate for}
^{distributed lag model is the sum}^{of the two lagged betas.}
When using the trade deficit instead of imports, the results agree with the predictions of some free
trade opponents, but they are also small. A one percentage point increase in the trade deficit as a ²⁰
share of GDP decreases worker compensation by 0.79%, an effect that was highly significant.
Only 3% of the variation in compensation can be explained by imports, however. The distributed
lag method had a slightly larger effect and was still highly significant. The autoregressive method
reduced the effect of trade deficits to nearly zero and made it statistically insignificant. Over the
more recent subsample, however, the autoregressive method yielded a result that was highly
significant and similar in size to the simple bivariate regression. Only the autoregression had an
R-squared value that was not negligible.

²⁰^If^{a d}^u^mmy^vari^ab^l^e^f^o^{r recessi}^oⁿⁱ^s^add^e^d^,^t^h^{e ef}^fect^b^eco^{mes st}^a^tⁱ^s^tⁱ^cal^l^yⁱⁿ^si^gnⁱ^fⁱ^c^an^t^.

Most economists argue that free trade is mutually beneficial because it allows countries to
specialize in what they do best, thereby employing their resources in the most efficient way. They
argue that trade neither creates nor destroys jobs on net, because jobs destroyed by trade are
replaced by new jobs elsewhere in the economy, and the economy is capable of reaching full
employment regardless of what happens to imports or the trade deficit. Some opponents of free
trade argue that imports destroy jobs, or trade deficits destroy jobs, or trade lowers wages. This
report looks at quarterly national data from 1948 to 2005 to try to clarify some of these issues.
The first set of regressions performed in this report can be thought to answer the first of the three
questions posed in the Summary: Do increases in imports systematically correspond to periods of
high unemployment or negative/low employment growth? The empirical evidence disputes that
claim on three counts. First, most of the evidence is statistically insignificant, meaning that it
cannot be ruled out that any correlation found between the variables is simply random chance.
This means the evidence does not meet generally accepted standards of professional research.
Some of the evidence that is statistically significant becomes insignificant with small
modifications in methods or data. If a relationship between variables were fundamentally strong,
it would be expected to hold under many different statistical specifications. Second, the R-
squared measure for these regressions is very low (except when lagged values of the dependent
variable are included), which means that very little of the variability in unemployment or
employment can be explained by imports. Finally, even when the results are statistically
significant, they do not support the position that imports have a negative effect on employment.
In one case, a one percentage point increase in imports as a share of GDP was predicted to
increase the employment-population ratio by 0.30-0.42 percentage points, all else equal. In
another case, a one percentage point increase in imports as a share of GDP was predicted to
decrease the unemployment rate by 0.10 percentage points, all else equal.
The second set of regressions answers the second question: Do increases in the trade deficit
systematically correspond to periods of high unemployment or low employment growth? Here
too, the empirical evidence is weak. Most of the evidence is statistically insignificant and explains
little of the variation in unemployment or employment growth. In this case, the relationship is
consistent with the predictions of some free trade opponents, but is too small to have a significant
economic impact. In one case, a one percentage point increase in the trade deficit as a share of
GDP was predicted to reduce the employment-population ratio by 0.17-0.20 percentage points, all
else equal. In another case, a one percentage point increase in imports as a share of GDP was
predicted to decrease the unemployment rate by 0.10 percentage points, all else equal. All of these
results were only marginally statistically significant, and lost significance over different time
periods or using different methods.
The third set of regressions answers the third question: Do increases in imports or the trade deficit
systematically correspond to periods of weak growth in worker compensation? For these
regressions, the results were more consistently statistically significant, with all the results
significant over the period of 1980-2005. The results still had low R-squared measures, however.
Imports consistently had the effect that some free trade proponents predicted—a one percentage
point increase in the import-GDP ratio led to a 0.81%-1.58% increase in worker compensation, all
else equal. Trade deficits consistently had the effect predicted by some free trade opponents,
although the effect was small: a one percentage point increase in the trade deficit as a share of
GDP reduced worker compensation by 0.79-1.17%, all else equal. These results are based on

average worker compensation, and do not address the question of how trade affects income
inequality among workers at different income levels.
These results may seems surprising to non-economists, but they are generally consistent with
events in recent decades. Imports have been steadily rising as a share of GDP for decades, and yet
the economy has mostly maintained full employment over that time. The 1990s serve as an
example of a period of low unemployment, rapid employment growth, and steady compensation
gains in the midst of high imports and large trade deficits. Recent experience provides another
example: imports and the trade deficit reached record highs in 2005, and the unemployment rate
fell to 5.1%.
These empirical results should be considered suggestive rather than definitive, as more elaborate
methods would be required to tease out the true relationship between trade and labor markets. For
instance, trade probably disproportionately affects some regions or occupations or sectors of the
economy positively and others negatively. Nevertheless, the results generally confirm the view
that, on balance, trade’s adverse impact on labor markets overall are small or non-existent
because the negative effects on some American workers are roughly canceled out by positive
effects on other workers.

This section provides a brief introduction to the econometric techniques used in this report.
Equations estimated by linear regression take their form from some hypothesized relationship in
which the behavior of one or more variables (the independent variables) is held to influence some
other variable (the dependent variable). The application of regression analysis involves fitting a
straight line to a group of observations, usually a sample selected from the universe of those
variables, that are suggested by the hypothesized relationship. The straight line is fitted such that
the deviations of the actual observations from those suggested by the straight line are minimized.
The value of the slope of that line then gives the effect of the independent variable (or variables)
on the dependent variable.
While the calculated value of the slope of the line may be positive or negative (or even zero), it’s
true value may not be statistically significantly different from zero. Because this is so, it will be
necessary to briefly discuss what is meant by a calculated value being “statistically significant.”
To understand statistical significance, say that the calculated effect of the independent variable on
the dependent variable is 0.10. Thus, changes in the independent variable by one unit change the
dependent variable by 0.10. This assumes, of course, that the value 0.10 is really different from
zero. Recall, that it was calculated, not from the universe of the independent variable, but from a
sample taken from that universe. Thus, it is possible that our assumption that the true value of the
effect of this variable on the dependent variable is different from zero is wrong. It is, however,
possible to control for making this type of error—that is, for accepting as true a relationship that
is, in fact, not true. It is common to set the control factor at 1 to 5 chances in 100 of accepting the
hypothesis that the variable is different from zero when it is not. If the calculated value of the
independent variable lies within a range that limits the error to 1% to 5%, it is said to be
statistically significant (or statistically significantly different from zero).
The R-squared measures a goodness of fit. The R-squared is designed to measure the fraction of
the variation of the dependent variable that is explained by the variation of the independent
variable(s). The R-squared ranges in value between 0 and 1. The higher the R-squared is, the
larger is the proportion of the variation in the dependent variable that is explained by the variation
of the independent variable(s).
^{Marc L}^a^bon^t^e
^Specialis^{t i}ⁿ^Macroecon^o^mⁱ^c^P^o^licy
^m^l^abon^t^e^@^c^rs^.l^oc.g^o^v^{, 7-}⁰⁶⁴⁰

²¹^T^h^is^s^e^c^{tion w}^a^s^pr^e^p^a^r^e^d^w^{ith G}^a^{il Ma}^kⁱ^ne^{n, G}^o^v^e^rⁿ^m^e^{nt a}^nd^Fⁱ^na^nc^e^Dⁱ^vⁱ^sⁱ^on.