Preliminary DRAFT; please do not quote without contacting the authors
Effects of Terms of Trade Gains and Tariff Changes
on the Measurement of U.S. Productivity Growth*
Robert C. Feenstra
University of California-Davis and NBER
Marshall B. Reinsdorf
U.S. Bureau of Economic Analysis
Matthew J. Slaughter
Tuck School of Business at Dartmouth and NBER
Since 1995, growth in productivity in the United States has accelerated dramatically, due in large
part to the information technology sector. In this paper we argue that part of the apparent speed-
up in productivity growth actually represents gains in the terms of trade and tariff reductions,
especially for high-tech products. Unmeasured gains in the terms of trade and declines in tariffs
cause real output growth and productivity growth to be overstated. Building on the GDP function
approach of Diewert and Morrison, we develop methods for measuring these effects. The growth
rates of our alternative price indexes for U.S. imports are as much as 2% per year lower than the
growth rate of price indexes calculated using official methods. The unmeasured terms-of-trade
gains can account for close to 0.2 percentage points per year, or about 20% of the apparent
increase in productivity growth for the U.S. economy. Biases in domestic price indexes, which
are beyond the scope of the research in this paper, could offset some of the effects of the biases
in the export and import indexes on the measurement of output and productivity growth.
* We are grateful for Mike Harper’s assistance with the analysis of the productivity measurement implications. We
draw heavily upon Alterman, Diewert and Feenstra (1999), and the authors are indebted to Bill Alterman and Erwin
Diewert for that earlier study which we apply here to U.S. productivity growth. For financial support Feenstra and
Slaughter thank the National Science Foundation. Finally, the views expressed in this paper are those of the authors,
not those of the Bureau of Economic Analysis.
Since 1995, growth in aggregate labor productivity in the United States appears to have
accelerated markedly. The U.S. Bureau of Labor Statistics (BLS) reports that from 1973 to 1995,
output per worker hour in the nonfarm business sector grew on average at just 1.40 percent per year.
From 1995 through 2007 this rate accelerated to an average of 2.55 percent per year. 1 This speed-up
in U.S. productivity growth would, if sustained, carry dramatic implications for the U.S. economy.
At the previous generation’s average annual growth rate of 1.40 percent, average U.S. living
standards were taking 50 years to double. Should the more-recent average annual growth rate of
2.55 percent persist, then average U.S. living standards would take just 28 years to double – or a
What are the explanations for this improvement in productivity growth? Among others, the
declining prices of information technology (IT) products, which accelerated in the late 1990s, are
often credited with key direct and indirect roles in this productivity speedup. Jorgenson (2001, p. 2)
argues that: “The accelerated information technology price decline signals faster productivity
growth in IT-producing industries. In fact, these industries have been the source of most of
aggregate productivity growth throughout the 1990s.” 2 In this paper we advance a related, but new
hypothesis: that international trade, and in particular the increased globalization of the IT sector,
accounts for an important part of the speed-up in productivity growth.3
On many measures, the global engagement of U.S. IT firms deepened after 1995—precisely
1 These calculations are based on BLS data series #PRS85006092, as reported at www.bls.gov. Similar trends are
evident in the BLS measures of multifactor productivity (MFP) for the private business sector, which we graph in
Figure 1. Since 2002, however, U.S. productivity growth appears to have decelerated again.
2 Similarly, Oliner and Sichel (2000, p. 17) state that, “In accord with the ‘dual’ framework described above, we
have interpreted the sharp decline in semiconductor prices after 1995 as signaling a pickup in that sector’s TFP
growth.” On the indirect role of IT in the productivity speedup, Jorgenson (2001, p. 22) finds that, “In response to
these [IT] price changes, firms, households, and governments have accumulated computers, software, and
communications equipment much more rapidly than other forms of capital.”
3 See also Mann (2003), who expresses a similar viewpoint.
the period of accelerated IT price declines that have been interpreted as total factor productivity
(TFP). An important factor in this price decline is that IT has been the only industry to have a
multilateral trade liberalization under the World Trade Organization. As we discuss in section 2, the
Information Technology Agreement (ITA) was ratified in 1996 by dozens of countries accounting
for nearly 95 percent of world IT trade, and the eliminated all world tariffs on hundreds of IT
products in four stages from early 1997 through 2000. This timing suggests that the ITA may have
played an important role in the post-1995 trends in IT prices.
To provide some suggestive evidence for this hypothesis, in Figure 1 we graph the U.S.
terms of trade (the ratio of the export price index to the import price index) since 1989, together
with multifactor productivity from BLS. U.S. nonfarm multifactor productivity growth rose from
0.53 percent per year during 1987-1995 to 1.41 percent per year during 1996-2006. The overall
terms of trade are heavily influenced by oil imports, so to avoid that influence we use the overall
export price divided by the non-petroleum import price, both from the BLS. This index of U.S.
terms of trade shows a declining trend up until 1995 in Figure 1. Since 1995 – at precisely the
time that productivity growth picked up – its behavior changed, with a string of solid gains in the
non-petroleum terms of trade index from 1995 through 2007. The average annual gain in the
BLS non-petroleum terms of trade from 1995 to 2007 is 1.0 percent, so the cumulative gain was
nearly as large as the deterioration in terms of trade from the petroleum price shocks in 1973-74
and 1979-80, which totaled around 15%.
The BLS uses a Laspeyres formula to construct price indexes for imports and exports
based on price quotes it collects from importing and exporting firms. We have this price data for
September 1993 through December 1999, so we are able to reconstruct the Laspeyres price
indexes of BLS for that time period. The ratio of our Laspeyres export price index to our
Laspeyres non-petroleum import price index is also shown in Figure 1. Our Laspeyres terms-of-
trade index does not exactly match the one constructed from published BLS indexes due to
missing data for some industries, but the difference is immaterial.
The finding that the U.S. terms of trade began to improve at precisely the time of the
productivity speedup, as shown in Figure 1, suggests that there could be some connection
between the two. Yet there are strong theoretical reasons to think that changes in the terms of
trade have no effect on productivity growth. Kehoe and Ruhl (2007) have recently argued that
changes in the terms of trade have no impact on productivity when tariffs are zero. When tariffs
are present but small, then the impact of terms of trade shocks on productivity is correspondingly
small. In section 3 we will extend the analysis of Kehoe and Ruhl (2007) from a one-sector to a
multi-sector model and also consider tariff reductions. We show that tariff reductions and
changes in the terms of trade have only a second-order impact on GDP and productivity.
If the terms of trade are mismeasured, however, the story is different. Unmeasured
changes in the terms of trade have a first-order impact on reported productivity growth. In
particular, if the reduction in import prices is understated, productivity growth will be
correspondingly overstated. There are three reasons to expect that the U.S. terms of trade are
mismeasured: (i) as already noted, the import and export prices indexes published by the BLS are
Laspeyres indexes, rather than a superlative formula; (ii) in the calculation of GDP, imports
exclude duties, and the BLS import indexes—which the Bureau of Economic Analysis (BEA)
uses to deflate imports—also measure import prices free of tariffs; (iii) the BLS import price
index does not account for increases in the variety of imports coming from new supplying
countries, as analyzed by Feenstra (1994) and Broda and Weinstein (2006). In section 4 we
construct price indexes that correct for these three features, and in section 5 we analyze the
impact of the ITA on the prices and variety of high-technology products. We find that high-tech
products are most affected by these sources of mismeasurement.
In terms of Figure 1, our central argument is that the improvement in the terms of trade
was even higher than displayed there. To preview our main results, several alternative terms-of-
trade indexes based on the calculations in this paper are shown in Figure 2. We repeat the BLS
and our computed Laspeyres terms of trade indexes from Figure 1, and also show: (i) an exact
Törnqvist index for the terms of trade; (ii) the Törnqvist index that also incorporates tariffs into
the imports prices; (ii) the Törnqvist index that incorporates tariffs and also import variety. The
first two of these indexes are set equal to the Laspeyres index in September 1993, the beginning
of our sample period, whereas the variety adjustment (which is annual) begins in 1990. It is
noteworthy that most of the variety adjustment occurs in the period since 1995, however, just
like our other adjustments. The cumulative impact of these three adjustments to the terms of
trade means that the rise in the Törnqvist index, incorporating tariffs and variety, to December
1999 is nearly equal to the cumulative rise in the BLS index to December 2007 (compare Figures
1 and 2). While the BLS index rises 1.0 percent per year over 1995-2007, the Törnqvist index
incorporating tariffs and import variety rises twice as fast, at 2.1 percent per year over 1995-
1999. Evidently, the terms-of-trade gain for the United States since 1995 has been much higher
than suggested by official price indexes.
From our aggregate terms-of-trade indexes in Figure 2, however, we cannot infer how
unmeasured terms of trade gains impact reported U.S. productivity growth. The reason is that
BLS’s aggregate export and import indexes have no role in BEA’s measures of real output
growth, which drive the calculations of productivity growth. Rather, BEA constructs GDP
deflators from detailed industry export and import price indexes, generally the five-digit Enduse
indexes produced by BLS, using a chained Fisher formula and GDP weights.
To estimate the impact of mismeasured terms of trade on reported productivity growth,
we construct alternative price indexes at the 5-digit (or, if appropriate, 3-digit) Enduse level of
detail. We then aggregate these detailed indexes using a chained Fisher index formula with
weights that reflect their importance in GDP. The effects of the alternative detailed price indexes
on the measure of productivity growth are generally the same as the ones that we calculate for
real output growth.4 The alternative detailed indexes that we consider are: Laspeyres indexes that
mimic the BLS indexes; Törnqvist indexes; Törnqvist indexes including tariffs; and Törnqvist
indexes including tariffs and a correction for new and disappearing varieties. These alternative 5-
digit indexes are used to construct deflators for GDP and for the subset of GDP that excludes
government, the gross value added of private business.5
By comparing productivity growth calculated from our corrected indexes with that
obtained with the reconstructed BLS indexes, in section 6, we estimate the portion of reported
U.S. productivity growth 1990s that was actually due to unmeasured gains in terms of trade. Our
central estimates are that properly measured terms-of-trade gains can account for close to 0.2
percentage points of the post-1995 increase in productivity growth for the U.S. economy.
Comparing that amount to the increase in labor or multifactor productivity, the terms of trade
accounts for about 20% of the speedup in productivity growth. Section 7 concludes.
4 Except that, as we note below, inputs of capital services used to estimate TFP are sensitive to the price indexes for
imported capital goods.
5 In analyzing TFP for a sector, the conceptually correct measure of output is the sector’s gross sales outside the
sector, which equal its value added plus its purchases of intermediate inputs from outside the sector. The output
concept for the private business sector ought therefore to equal its value added plus imported intermediate inputs,
and imported intermediates ought to be included in inputs . However, including imported intermediates in inputs
instead of netting them out of the output measure was found to have little effect on TFP estimates, so BLS measures
the output of private business by its value added. See Gullikson and Harper (1999, p. 50 and fn 29).
2. Globalization of the Information Technology Industry
To gauge the role of international trade in the production of IT goods and services, a
sensible starting point is to present trade flows for some specific industries. Take, for example,
computers, peripherals and semiconductors (Enduse category 213) and telecommunication
equipment (Enduse 214). 6 These sectors include some of the most high-profile information and
communication technology (ICT) industries. Table 1 reports current-dollar trade flows in these two
sectors for three years spanning most of the 1990s —1992, 1996, and 2000. The bottom of Table 1
also reports the share of economy-wide exports and imports flows accounted for by these industries.
Over the 1990s exports in these sectors have been rising faster than the national total, such that their
share of that total rose from 10.7 % to 15.4 %. But a more striking feature is the even higher level of
imports in these sectors. Over the 1990s their national import share rose from 12 % to 16 %.
This means that these two central ICT sectors are substantial net importers whose trade
imbalance widened during the decade. Within these sectors, computers and semiconductors show
that smallest trade deficits, while computer accessories and telecommunication equipment have the
largest deficits. By 2000 the combined trade deficit in these ICT sectors was $57 billion, or fully 17
percent of the non-oil U.S. trade deficit that year.
Table 2 offers some additional evidence on the trade intensity of IT industries, defined as
trade flows as a share of output. For 1997, Table 2 shows exports, imports, and net exports, all as a
share of output for two IT industries – computers and peripheral equipment, and semiconductors
and electronic components. (These two IT industries differ from the Enduse classifications used in
Table 1.) The key message of Table 2 is that IT industries are much more trade intensive than the
overall U.S. economy. In these industries both exports and imports as a share of output range
6 The Enduse industry classification is used by the BEA for measuring GDP, so we also use it here. Trade data for
Enduse industries comes from Bureau of the Census, 1992-2000.
between 19 and 38%. These measures of trade intensity are higher than manufacturing industries in
general, for which exports and imports were just 14–21% of output. Taken together, Tables 1 and 2
indicate that many of the central IT industries in the United States are more trade-intensive than is
the rest of the economy, and are substantial net importers.
There are many factors that contribute to the increasing globalization of the IT industry,
including the creation and spread of global production networks. But in the second-half of the
1990’s, one event in the global economy was of particular importance. Under the auspices of the
World Trade Organization (WTO), an Information Technology Agreement (ITA) committed
signatory countries to eliminate all tariffs on a wide range of nearly 200 ICT products. These
products covered both finished and intermediate goods such as computers and networking and
peripheral equipment; circuit boards and other passive/active components; semiconductors and their
manufacturing equipment; software products and media; and telecommunications equipment.
The original Ministerial Declaration on Trade in Information Technology Products was
concluded in December 1996 at the first WTO Ministerial in Singapore. This declaration stipulated
that for the ITA to take effect, signatory countries would have to collectively represent at least 90%
of world trade in the covered products. The 29 original signatories accounted for only about 83% of
covered trade. But by April 1997 many more countries had signed on to push the share over 90%,
and the agreement entered into force in July 1997. Ultimately there were more than 50 ITA
signatories that accounted for more than 95% of world trade in the covered ITA products. All ITA
signatories agreed to reduce to zero their tariffs for all covered ITA products in four equal-rate
reductions starting in 1997 and ending no later than the start of 2000. 7 Some developing countries
were granted permission to extend rate cuts beyond 2000, but no later than 2005. Also, an ITA
7 The four tariff cuts for the U.S. occurred in July 1997, January 1998, January 1999, and for a small number of
commodities, January 2000.
Review Committee was established to monitor compliance. The overarching goal of the ITA was to
eliminate world tariffs in a wide range of IT products. Thanks to the number and commitment of
signatory countries, it has virtually achieved that goal.
The tariff reductions over 1997-1999 experienced by a number of U.S. ICT industries are
shown in Table 3. The ITA tariff cuts are defined at the 8-digit level of the Harmonized System
(HS) system, used to track import commodities. In the second column of Table 3, we indicate the
percentage of import value within each industry that are covered by ITA commodities. For
computers, peripherals and semiconductors, 100% of imports were included in the ITA tariff cuts.
In the smaller industry of blank tapes for audio and visual use, 90% of the imports were covered by
the ITA, and in the large sector of telecommunication equipment, 80% of the import value was
covered by the ITA. Table 3 also includes the information for several other industries where more
than 50% of import value was covered by the ITA, and industries such as industries such as
business machines and equipment, and measuring, testing, and control instruments, where less than
50% of the import value was impacted by the ITA agreement.8
In Table 3 we show the average tariffs at the beginning of 1997, before the ITA was
implemented, and at the end of 1999, when it was concluded. It is apparent that U.S. tariffs in these
industries even before the ITA agreement were low: average tariffs are between one and four
percent in all industries, and zero or nearly so in computer accessories and semiconductors. This
means that the ITA tariff cuts for the United States were correspondingly small. But remember that
the ITA was a multilateral agreement, so that tariff cuts in the U.S. could be matched by equal or
larger tariff cuts abroad. For firms sourcing their IT products from overseas locations, the tariffs cuts
within the ITA could therefore have a multiplied impact on lowering their import prices and costs,
as we will argue in section 5.
In addition to their potential impact on prices, multilateral tariff cuts under the ITA could
expand the range of supplying countries, providing differentiated varieties of IT products. Recent
literature has shown how to measure the product variety of imports – or the “extensive margin” of
trade (Feenstra, 1994; Hummels and Klenow, 2005; Broda and Weinstein, 2006). In the final
column of Table 3 we show the growth in import varieties for each 5-digit Enduse category,
reflecting the increasing number of supplying countries. The formula for import variety will be
discussed in the following sections, but we note here that the increases in import variety for the ITA
products was above that for all other U.S. imports, suggesting that the multilateral cuts under the
ITA were effective in increasing the range of supplying countries.
3. Measurement of Productivity Growth with International Trade
We have seen in the previous section that tariffs on imported ICT products fell during the
period covered by the ITA. In this section we describe how such declines in tariffs are
incorporated into the measurement of total factor productivity (TFP) at an economy-wide level.
In our exposition of the theory, we assume that productivity is measured for GDP as a whole
even though, in practice, the available information only permits the private business portion of
GDP to be included in the broad measure of productivity growth.
Our theoretical model extends the model of international trade and productivity of
Diewert and Morrison (1986), which treats imports as intermediate inputs into the economy’s
GDP function. Though such a uniform treatment for all imports might appear to ignore the
significant representation of finished goods in U.S. imports, Diewert and Morrison argue that
imports can all be treated as intermediate inputs in the GDP function because even finished good
imports generally require domestically produced distribution services to get to final consumers.
8 Omitted from Table 3 are industries where less than 10% of imports are covered by the ITA.