2. Data treatment and methodology

Starting from the standard, theory-consistent gravity model of Anderson and van Wincoop (Reference Anderson and van Wincoop2004), Novy (Reference Novy2013) develops a comprehensive measure of bilateral trade costs.Footnote ⁴ Equation (1) presents that measure in ad valorem equivalent terms. It is the geometric average of bilateral trade costs for exports from country i to country j and from country j to country i , expressed relative to domestic trade costs in each country (${\textstyle{{t_{ij}} \over {t_{ii}}}} $ and ${\textstyle{{t_{\,ji}} \over {t_{\,jj}}}} $ respectively). To calculate it, all that is required is data on domestic usage relative to exports in both countries (${\textstyle{{x_{ii}} \over {x_{ij}}}} $ and ${\textstyle{{x_{\,jj}} \over {x_{\,ji}}}} $ respectively). The parameter σ is the elasticity of substitution among varieties in a sector, assuming the Anderson and Van Wincoop-based derivation of Novy's measure of trade costs. The parameter γ is a measure of firm-level heterogeneity if the model is instead derived from Chaney (Reference Chaney2008).

(1)$$\tau _{ijkt} = \mathop {\left( {\displaystyle{{t_{ijkt}. t_{\,jikt}} \over {t_{iikt}. t_{\,jjkt}}}} \right)}\nolimits^{{\textstyle{1 \over 2}}} - 1 = \mathop {\left( {\displaystyle{{x_{iikt}. x_{\,jjkt}} \over {x_{ijkt}. x_{\,jikt}}}} \right)}\nolimits^{{\textstyle{1 \over {2\left( {\sigma - 1} \right)}}}} - 1 = \mathop {\left( {\displaystyle{{x_{iikt}. x_{\,jjkt}} \over {x_{ijkt}. x_{\,jikt}}}} \right)}\nolimits^{{\textstyle{1 \over {2\gamma}}}} - 1$$

Intuitively, Novy's measure captures the fact that if a country's trade costs vis-à-vis the rest of the world fall, then a part of its production that was previously consumed domestically will instead be shipped overseas. Trade costs are thus closely related to the extent to which a country trades with itself rather than other countries, and data on this kind of relative openness can be used to make inferences about the level of trade costs and their variation over time.

This approach has three main advantages over the readily available alternatives. First, it represents a comprehensive measure of the full range of trade costs, namely the costs of providing goods or services across borders relative to the costs of providing them within countries. In goods markets, it captures international shipping – as in work using CIF/FOB ratios – but also a much wider variety of cost factors (see Anderson and van Wincoop, Reference Anderson and van Wincoop2004 for a full review). In services, it takes account of factors such as: geographical, cultural, and linguistic distance; regulatory barriers to trade; regulatory heterogeneity; access to finance for accessing new markets; differences in business and investment climates; and other behind-the-border measures that have asymmetric impacts on domestic and foreign suppliers. Even the effects of regulatory measures that are discriminatory in fact but not in law are included in this measure of trade costs.

The second advantage of Novy's measure is that its data requirements are minimal. As a result, it is feasible to obtain measures of trade costs in services sectors, where data availability is much more limited than for goods. We are thus able to obtain data on usage, exports, and trade costs across a wide variety of countries, sectors, and time periods.

Third, Novy's measure relies on a theory-based rearrangement of data, rather than econometric estimation. It thus does not suffer from the possibility of omitted variables bias, which potentially affects traditional gravity models of services markets.Footnote ⁵ Another potential problem with gravity model-based estimates of trade costs such as Walsh (Reference Walsh2006) is that they tend to confound trade restrictions and noise in the data, both of which are associated with trade flows that diverge from expected values based on model estimates.

This methodology for computing time-varying trade costs therefore requires data on both domestic output and bilateral trade at the sector level. We use a 29 sector classification based on ISIC Rev.3, which is the industry classification used in national accounts data such as input-output tables. Below, we provide a brief overview of the way data were collected and handled in order to calculate τ _ijkt across countries, sectors, and years. In the end, we construct Novy (Reference Novy2013) measures of trade costs in services covering up to 61 countries and 12 services sectors. As a point of comparison, we also compile data on 17 goods sectors. Data at this level of detail are only available for a subset of countries, however, and so we also use an aggregate version of the data. This approach allows us to expand the analysis to other countries for which disaggregated output data are unreliable and/or relatively unavailable. In the next section, where we present results, we are careful to use a balanced sample, in the sense of country–partner–sector combinations that are observed consistently over the full time period for which the analysis is presented. This limitation is important in order to avoid entry and exit of countries or sectors from the sample, with corresponding changes to measured trade costs that reflect that dynamic rather than the underlying situation.

In all calculations, the elasticity of substitution σ is set equal to 8, which is the same rule of thumb used by Novy (Reference Novy2013). It is very close to the typical country's average elasticity of substitution reported by Broda et al. (Reference Broda, Greenfield and Weinstein2006) of 6.8. Setting σ = 8 is equivalent to setting γ = 7, which is also reasonable in light of previous work on firm level productivity distributions.Footnote ⁶ Although ad valorem equivalents are quite sensitive to the value chosen for σ or γ, using indices relative to a base year reduces that problem to economically insignificant levels. For example, using σ = 10 results in substantially different estimated worldwide ad valorem equivalents for services: 106% in 1995 and 107% in 2007, compared with 153% and 155% respectively with σ = 8. However, the change in those numbers over time is nearly identical: 1.3% versus 1.4%. The following subsection presents results using both methods.

3. Trade costs in services, 1995–2007

Full results for 1995 and 2007 are reported in the Appendix. In terms of interpretation, first, Table 1 column 1 compares our results with those reported in Novy (Reference Novy2013) for USA trade costs in goods vis-à-vis its major trading partners. In line with the methodology set out above, we and Novy are both measuring international trade costs (between countries) relative to domestic trade costs (within countries). Our results are generally very close to his, although slightly higher in some cases. This finding provides reassurance that our new data, which cover a much broader range of countries and sectors than previous estimates, are reliable with regard to existing work. It is worth stressing that the relatively high numbers reported in Table 1 reflect the full range of cost factors that affect international transactions, not just protection (i.e. tariffs in the case of goods markets).Footnote ⁷ Based on a comprehensive review of the literature, Anderson and van Wincoop (Reference Anderson and van Wincoop2004) estimate that international trade costs of the type we are measuring amount to around 74% in ad valorem equivalent terms for goods markets. Our number is thus slightly lower than theirs, even though both figures may seem high compared to those used to discussing rates of protection in advanced countries, which are generally in single digits. It is important to highlight, however, that not all trade costs can be reduced by policymakers in the way that tariffs and non-tariff barriers can. There will always be some level of trade costs due to, for example, geographical distance.

Table 1.

Comparison of trade costs in goods and services, USA vis-à-vis major partners, 2000 (percent ad valorem equivalent)

The standout result from Table 1 is that the level of trade costs is much higher in services sectors than in goods – about double, on average, for the US and its major trading partners. The same result emerges from Table 2, which compares trade costs of major trading economies vis-à-vis the rest of the world. The absolute levels of trade costs in services are very high: over 100% in all cases, and over 200% for India. These figures should be interpreted with caution, since they rely on an assumption as to the value of the elasticity of substitution, and on an assumption that it is the same in goods and services sectors. But even if the numbers are subject to uncertainty, the relative pattern of trade costs is clear: they are much higher in services sectors than in goods sectors. Again, we stress that not all components of the trade costs figures we have calculated are amenable to reduction by policymakers in the way that tariff-like barriers are. There will always be some additional costs of doing business abroad relative to selling services domestically due to geographical distance, cultural and linguistic barriers, and the characteristics of the services being produced.

Table 2.

Comparison of trade costs in goods and services, major trading economies vis-à-vis the rest of the world, latest year (percent ad valorem equivalent)

At first glance, the results in Table 1 might appear surprising: even between two markets that are close geographically and culturally, and that are perceived to be relatively open – the USA and Canada – there are trade costs in services of around 100%, compared with only 30% in goods. It is important to remember, however, that this number is not a measure of protection, rathert encompasses all types of trade costs.

The same reasoning applies to the estimates in Table 2. It is to be expected that the measures we present will be substantially higher than, for example, the ad valorem equivalents reported by Walsh (Reference Walsh2006) for services markets. Taking that author's results as a measure of protection, it is by no means unrealistic that additional trade costs – i.e. those not related to direct trade protection – should add an additional 70% or so to the overall level of trade costs for the USA vis-à-vis the rest of the world. The comparison with goods markets is again instructive. The World Bank's Overall Trade Restrictiveness Index suggests a protection rate in the USA of a little over 6% for goods, but total trade costs are more on the order of 60%. So in ad valorem terms, total trade costs in services are perhaps double the rate of protection, whereas in goods the multiple is more like ten times.

As noted above, our calculation of the level of trade costs in ad valorem equivalent terms is sensitive to our assumed value for σ, the intra-sectoral elasticity of substitution, or γ, an indicator of firm-level heterogeneity. By expressing results in terms of a base year–i.e., as index numbers–the sensitivity problem practically disappears. Index number results are very close for values of γ in the range five to nine, around our central value of seven (as in Novy, Reference Novy2013).

The index number approach is useful for examining performance over time (Figure 1 through Figure 5). We see that on a world aggregate basis (Figure 1), trade costs in goods have declined by more than 15% over the last ten years. The same is by no means true for services: trade costs have barely moved over the last decade, and if anything, may have slightly increased. The same pattern emerges for most of the major trading economies in the next set of figures. It is interesting to note that the US services index declines by a few points over the period 1995–1998, but the EU index does not – even though this period should coincide with realization of the full impact of the 1992 program on the internal market in services. Of course, the EU internal market in services does not mean that trade costs are centrally set in Brussels. Member States still have extensive powers to impose domestic regulations covering a wide variety of circumstances. We should therefore not necessarily expect to see a strong reduction in international relative to domestic trade costs over our sample period, because both measures have been changing in particular Member States over time. Moreover, the nature of our trade costs measure limits the strength of the interpretation we can put on these kinds of changes. Because it is the ratio of international to domestic trade costs, the failure to record a fall in the ratio could equally well be due to proportional declines in both international and domestic trade costs over time. Nonetheless our interpretation in terms of a lack of movement in trade costs is consistent with the fact that major markets have not engaged in substantial new liberalization of services in recent years, in particular due to the stalled Doha Round negotiations in this area.

Figure 1.

World aggregate trade costs indices for goods and services, 1995–2007 (1995 = 100)

The obvious exception to the pattern described above and in Figure 1 through Figure 4 is China (Figure 5): trade costs have declined substantially in services markets as well as in goods. It is important to highlight two caveats to this finding. First, Table 2 shows that China still has high levels of trade costs relative to other major economies. So there is still considerable room for further reductions. Nonetheless, the proportional reductions since 2000 are impressive.

Figure 2.

US trade cost indices for goods and services, 1995–2007 (1995 = 100)

Figure 3.

EU trade cost indices for goods and services, 1995–2007 (1995 = 100)

The second caveat is that the trade costs that we are measuring represent an average of trade costs facing foreign producers exporting to China, and trade costs facing Chinese firms exporting to other countries. Thus, we cannot conclude that the very large trade cost reductions observed for China are solely due to policy reforms undertaken at home. They are also linked to trade cost reductions by China's trading partners following accession to the WTO, which provided China with permanent MFN status.

Figure 4.

Japan trade cost indices for goods and services, 1995–2007 (1995 = 100)

Figure 5.

China trade cost indices for goods and services, 1995–2007 (1995 = 100)

Nonetheless, our findings are in line with recent work suggesting that the terms of China's WTO Accession Agreement had real ‘bite’ in services, in the sense that they required significant liberalization of applied policies, at least to the extent that they relate to foreign market access (Mattoo, Reference Mattoo, Deepak Bhattasali and Martin2004). It is important to note, however, that since our measure is the ratio of international to domestic trade costs, what we are picking up in this change is primarily a reduction in discrimination vis-à-vis foreign service providers, rather than domestic regulatory reform that is arguably important in the context of services sector liberalization more generally. In any case, the Chinese example is in stark contrast to most countries’ GATS schedules submitted at the end of the Uruguay Round: they established binding policy ceilings, but contained little in the way of genuine liberalization. Whatever liberalization has occurred, since the Uruguay Round, has been undertaken unilaterally or preferentially, and it is therefore not surprising that for other countries, changes in trade costs on a multilateral basis have been limited.

Another case that is deserving of attention is that of the EU. Since trade policies are set in Brussels, one might expect a certain level of uniformity of trade costs with respect to the rest of the world. However, this is not the case. First, different countries display highly divergent levels of trade costs in goods and services markets alike. Germany, for instance, has tariff equivalents of 65% and 127% for goods and services respectively for 2007. The same figures for another large Eurozone economy, France, are 91% and 168%, while for a small economy, Ireland, they are 114% and 165%. For a ‘new’ EU member, Poland, the tariff equivalents are 75% and 144%. This pattern suggests that there are a range of factors that influence trade costs, but which remain heterogeneous at the level of individual member states. It is beyond the scope of this paper to delve into the reasons for that heterogeneity in detail, but it seems likely that future work could uncover evidence of regulatory measures and other behind-the-border factors that exert a significant influence on trade costs, but which are not fully encompassed in the EU's common trade policy. It is also important to recall that we are measuring international relative to domestic trade costs, and EU rules allow the latter to vary substantially across countries, perhaps accounting for some of the differences we observe in the ratios.

In addition, we find that changes in services trade costs over time are also very different across EU countries (Figure 6–Figure 9). There are substantial reductions in trade costs between 1995 and 2007 in Germany, Ireland, and Poland, but a noticeable increase over the same period in France. This is in marked contrast with goods markets, where all four economies register substantial declines in trade costs. Although more in-depth analysis is needed before drawing any firm conclusions, this heterogeneity in time paths suggests that member states may have taken substantially different tracks in implementing the single market for services, and that significant imperfections remain (witness the aborted Bolkestein Directive).

Figure 6.

Germany trade cost indices for goods and services, 1995–2007 (1995 = 100)

Figure 7.

France trade cost indices for goods and services, 1995–2007 (1995 = 100)

Figure 8.

Ireland trade cost indices for goods and services, 1995–2007 (1995 = 100)

Figure 9.

Poland trade cost indices for goods and services, 1995–2007 (1995 = 100)

Finally, the case of Poland is interesting for what it can potentially say about the process of EU accession. In services markets, we see that trade costs only started to fall rapidly from about 2002 onwards. The changes for 2002–03 and 2003–04 were particularly large, on the order of 5% each year. The rate of decline in goods trade costs, however, is much more consistent over the full sample. This example tends to suggest that the degree of ‘bite’ in the single market for goods is substantially greater than that of the single market for services: the process of acceding to the EU is associated with a consistent decline in trade costs of over 25% for goods, but a stop-and-start fall of less than 20% for services.

The index number approach is also useful for examining the evolution of sectoral trade costs over time. As an example, Figure 10 presents a comparison between construction, transport, financial, and computer services. Trade costs in construction services have increased markedly over the last decade, by nearly 20%. Construction is also the most insulated sector, with ad valorem equivalent trade costs of around 200%. Of course, performance is likely to vary markedly across countries; the figure presents aggregate data for the world as a whole, and therefore includes relatively restricted as well as fairly open markets.

Figure 10.

Trade costs indices for construction services, transport services, financial services, and computer services, 1995–2007 (1995 = 100)

None of the other three sectors exhibits a substantial increase in trade costs over time. Trade costs in transport services have remained approximately constant over the last decade. By contrast, trade costs in financial and computer services have fallen by more than 10%. This finding is perfectly consistent with the rise of outsourcing in those sectors over roughly the same time period: as trade costs fall, probably due to improved information and communications technologies (ICTs), it becomes feasible for firms to have more of these kinds of tasks performed overseas. Sectors such as transport and construction, on the other hand, are largely immune to such developments because of the need for physical proximity between producer and consumer.