Unit Values in International Trade and Product Quality

Is the unit value of traded goods representative of quality? To answer this question, we analyze unit value with respect to exporter countryâ€™s capacity to export, which is determined by its production cost, tariff, and distance. The change in a countryâ€™s export unit value is decomposed into the components associated with pure term-of-trade effect, quality effect, distance effect, and production cost effect. Our empirical results confirm that tariff, distance, and wages all significantly affect the unit values. Furthermore, by comparing CIF and FOB unit values, we show that quality is an important contributor on driving up the unit values: exporters increase unit price to distant trading partners through quality upgrading. This "Washington apple effect" is much larger than the pure distance effect or production cost increase.


Introduction
Is the unit value of traded goods representative of quality?A vast literature associates cross-country variation in export unit-value with variation in product quality.Brooks (2006) uses unit value differences to infer the quality gap for Colombian firms.Many policy researches also derive countries' quality competitiveness from crosscountry comparisons of export unit value (Aiginger, 1988;Verma, 2002;andIanchovichina et al., 2003 IADB/World Bank 2003).
Although countries with higher exporter price are also likely to produce higher quality, a moment reflection suggests that many factors are ignored by this assumption.International variation in export prices can be influenced by both demand and supply side factors.Based on sectoral data for bilateral trade among 60 countries, Hallak (2006) finds that richer nations tend to import relatively more from partners that produce higher quality products.Harrigan (2010) shows that imports from distant trading partners have much higher unit values, and are much more likely to arrive by airplane.Khandelwal (2010) uses a discrete choice random coefficient model to show that conditional on price, imports with higher market shares are assigned higher quality.Hummels and Klenow (2005) argue that product varieties and quality differences are necessary to explain the observed differences in unit values.They find that countries with twice the per-capita income export varieties with 9 to 23 percent higher quality.Schott (2004) demonstrates that U.S. import unit values are positively associated with exporter per capita GDP, exporter relative endowments, and exporter production techniques across time and

Econometric model and hypothesis
Consider the variation in product prices across countries.If we consider import tariffs and iceberg transportation cost in shipping, then the world price of product k produced by country j and imported to country i is: Where jk P is the price of product k manufactured and sold in exporter country j; ijk t is the ad-valorem tariff on product k when country i import from partner country j; jk w denotes the wage rate of product k in country j; ij d is the distance between country pair i and j, and k  >0 is the scaling factor for transportation cost between i and j when transferring product k.The second equality comes from the assumption of monopolistic competition, so the product price jk P is a constant markup over marginal cost.Thus we decompose the prices into factors that capture tariffs, production cost, and transportation cost.
Taking log on equation (1) and add some fixed effects; we can get the gravity-like regression equation as follows: CIF: Where the variables are: ijk UV denotes the unit value of imported product k by importer i from exporter j ijk t is the ad-valorem tariff levied by importer i on product k from exporter j jk w is the wage rates of product k in exporting country j, reflecting the production cost i D is the fixed effect for importing country j D is the fixed effect for exporting country k D is the fixed effect for product k, which controls for within product variation.So we can check unit value difference within narrow product categories.ijk  represents the myriad other influences on the bilateral imports, assumed to be orthogonal So the specification is a regression of bilateral sectoral export unit value on importer country dummies, exporter country dummies, factors that capture transport cost (bilateral distance), trade cost (tariff), and production cost (exporter wage).We explore the regressions by separating CIF and FOB prices.FOB denotes the producer reported free-on-board price of the products and does not include transportation cost, whereas CIF stands for cost-insurance-freight value of the traded products, which includes transport charges and tariff duties.Because CIF unit values capture the variations of transportation cost and tariff rates, we consider any difference in the FOB price from a given exporting firm must be due to quality.Our major hypotheses are: 1) 1   0 and 1   0: We consider 1  and 1  reflect the pure terms-of-trade (TOT) effect.Based on the trade theory, levying tariff would lower the world price if importer is a large country, and has no effect on the world price if the importer is of small country case, with little import share in the world market.So we expect the tariffs to have negative signs to the extent that trade costs are passed on to consumers.

2)
2  >0: When trading with faraway countries, exporters will choose products with higher unit FOB price by adding product feature or quality upgrade.Since FOB price is exclusive of transportation cost, 2  captures the pure quality upgrade or the "Washington apple effect"2 , meaning that unit values increase within narrow product categories systematically with distance. 3) 2  denotes the elasticities of a country's CIF unit value with respect to country pair distance.Compare with FOB price, CIF price is more sensitive to distance.Given equal distance, a small increase of distance leads CIF price to increase more than FOB.The difference between CIF and FOB price exactly reflects the effect of distance on unit values.Longer distance drives up the transport cost and final product price.

4)
3  >0 and 3  >0: We expect the producer wage rates to have positive effect on the export unit values.This is the pure price effect due to markups on the production cost.

Data in regression
From the regression specification, we need data on bilateral trade unit value, factors that capture trade cost, transport cost, and wage rate across the countries, all at the commodity level.
Trade data: Our primary data set is the Standard International Trade Classification (SITC) Revision 2 4-digit level "World Trade Flows, 1962-2000" (NBER-UN henceforth) compiled by Feenstra et. al (2005).The world trade dataset reports both quantity and value for each traded products.We compute the unit value or "price", by dividing trade value by quantity.Whenever possible, quantities for a given SITC code are converted into common units firsthand.If it is hard to convert the same product into common units, we treat each combination of STIC code and unit of quantity as a separate product.Availability of unit value information occupies 80% in 1996.For the period of 1984-2000, this dataset only covers imports and exports for 72 countries3 .However, they are the relatively large countries in the world and accounts for 98% of world exports.We compute the bilateral FOB unit values of traded goods using reports from the export country.By focusing on the exporters' reports, we ensure that these unit values do not include any shipping costs.The bilateral CIF unit values are similarly obtained from importers reported trade data, which are transport cost and tariff-inclusive.
Distance: Bilateral distance across 225 countries are obtained from the Centre D'Etudes Prospectives Et D'Informations Internationales (CEPII)4 .Distance is measured as the great circle distance between the capital cities of those two countries.
Tariff Rates: The primary source of ad-valorem tariff associated with most-favored-nation (MFN) status comes from the Trains-Haveman tariff data at UNCTAD.We convert the 6-digit HS level tariff rates to match the SITC 4digit trade data.
Wage: To capture production cost varying by countries, industries, and years, we need consistent and complete wage data at SITC 4-digit level.We use three wage series in our estimation.The first one is the "industry wages around the world" (IWW) that we constructed based on the annual manufacturing sectoral wage data from United Nation's International Labor Organization (ILO)5 .ILO collects detailed sectoral wages for countries around the world.However, the extent of variation in the ILO data complicates the direct use of cross-country comparison.Its wage information comes from 12 sources such as Administrative reports, Labor-related establishment survey, or even Insurance records.There are also 6 different worker coverage such as employees, skilled, unskilled, and wage earners.Some countries report wages for wage rates whereas others report earnings.The time span also varies from day, hour, to month or week.Wage earners gender is mixed with men, women, or both men and women.The data is also mingled with different industry classification such as ISIC revision 2 or ISIC revision 3. Furthermore, wages are all denominated in domestic currency whereas some country currencies experienced currency denomination change.So we painstakingly follow the method proposed by Freeman and Oostendorp (2000) to calibrate the diverse statistics into a normalized monthly wage rates for male wage earners. 6The final dataset is organized by 733 SITC Rev. 2 4-digit industries in 115 countries, over the period of 1969 to 2004.Notes: IWW is constructed by authors based on the ILO yearly manufacturing sectoral wage data; UNIDO wage is the average monthly dollar wages and salaries coming from INDSTAT3 database 2005 edition; OWW is the occupational world wage rates constructed by Freeman and Oostendorp (2000) based on the "October Inquiry" Survey of ILO wages.
To check the accuracy of our own IWW wage data, we also use two other widely used wage series in this crosscountry study.The first source is the "Occupational wages around the world" (OWW) database constructed by Freeman and Oostendorp (2000).This dataset transforms the "October Inquiry" Survey from the ILO into a consistent data file for, which covers pays in 161 occupations over 151 countries from 1983 to 1998.The second standard wage dataset is the UNIDO wage coming from the INDSTAT3 database 2005 edition.It reports wages at the 3-digit level of ISIC Rev. 2 classification, covering the period 1963-2003 for 180 countries.Table 1 summarizes data coverage for the three wage databases.Compare with the two standard wage series, our own IWW dataset has the widest coverage at SITC 4-digit industry level.It standardizes the most far-ranging collection of wages into a consistent series of pay across industries, countries, and time.The IWW database will contribute significantly to international economic studies.Notes: IWW is constructed by authors based on the ILO yearly manufacturing sectoral wage data.OWW is the occupational world wage rates constructed by Freeman and Oostendorp (2000) based on the "October Inquiry" Survey of ILO wages.
To check the substitutability of the three wage datasets, we further regress the IWW wage rates on UNIDO and OWW wages respectively.
Where, i , j , t stand for industry, country, and year respectively.OLS estimates of equations ( 3) and ( 4) are reported in Tables 3 and 4. We can find a strong unit correlation across the three wage series.
2 R is also as high as 0.91 to 0.94, suggesting perfect fit and substitutability.

Result & discussion
We start testing our hypotheses with the broadest possible sample available.The summary statistics of the regression data for CIF price and FOB price are presented in Table 5a and 5b respectively.Since OWW wage covers non-manufacturing sectors, we supplement the missing IWW wage with OWW wage (IWW_OWW henceforth) so as to extend to agriculture industries as well.The range of each variable is quite big, suggestive of substantial heterogeneity across products.Interestingly, the mean of CIF unit values is even smaller than FOB prices.Freeman and Oostendorp (2000) based on the "October Inquiry" Survey of ILO wages.
Table 6 reports the results of estimating equation (2), using the broadest data available and includes all countries that have non-missing trade, wage, transportation, or tariff data.Tables 6a and 6b present the results of CIF and FOB unit values, respectively.Each column uses a different wage: IWW wage, IWW_OWW wage, UNIDO wage, OWW using uniform weighting, and OWW using lexicographic weighting.Robust standard errors are reported in the parentheses.Notes: UNIDO wage is the average monthly dollar wages and salaries coming from INDSTAT3 database 2005 edition; IWW is the industrial world wage rates constructed by us based on the ILO yearly manufacturing sectoral wage data; OWW is the occupational world wage rates constructed by Freeman and Oostendorp (2000) based on the "October Inquiry" Survey of ILO wages; IWW_OWW stands for the IWW supplemented by the OWW uniform weighting wage rates.
As the first row of Table 6a shows, the effect of tariff rate on CIF unit values is large, robust and significantly negative, supporting the large importing country TOT improvement hypothesis.On average a 10 percent increase in tariff rates lowers the exporters' price by 28-40 percent.Distance is positive and significantly correlated to unit value.The sign of wage rate coefficient is mixed: it is positively correlated with the unit values for IWW, IWW_OWW, and UNIDO wage, but negative for the two OWW wages, though not significantly differ from zero.Table 6b reports the results using importer reported FOB price.Overall the positive correlation between distance and unit value still remains highly significant.The estimated wage effect also improves to be significantly positive all the wage series, with the wage elasticity ranging from 0.04 to 0.23.Tariff rates, however, have almost zero impact on the FOB unit values, supporting the small country scenario.The effect of distance on both CIF and FOB unit values are large, robust, and statistically significant.The estimated effects of distance are invariably larger for CIF unit values, supporting our hypothesis that the difference between CIF and FOB price reflects the pure distance effect.Overall, goods have higher unit values when they travel a greater distance.Given a 10 percent increase in bilateral trading partners' distance, exporters will increase FOB prices by 11-15 percent.This is exactly the "Washington Apple" effect because exporters try to update improve product quality and create more value-added to distant destinations.Importers reported CIF unit value will increase by 17-20 percent, indicating that 5-7 percent of prices increase is the pure distance effect.Hence quality effect is about twice important than the distance effect.This finding is consistent with Harrigan (2010), who finds that more distant exporters will choose to sell products with higher unit values, controlling for other country specific factors which might affect unit values.
Many possible reasons can explain why tariff rates are not significant for the FOB unit value.First, comparing with importer reports, it is easier for exporters make reporting errors when going through the customs office (Feenstra et al. 2005).Second, measurement errors may also be created on purposely for transfer pricing or tax evasion purposes.As shown by Fisman and Wei (2004), facing high import tariffs, exporters may on purposely underreport the unit value, under-report the taxable quantities, or mislabel the higher-taxed products as lower-taxed products.Thirdly, the difference is caused by the limitation of the NBER-UN world trade flow dataset, which only collects the import and export reports from 72 countries.Table A1 lists all the country pairs included in the two regressions.For importer reported CIF prices, there are 50 large countries importing from 182 countries in the regression sample.So the importers fit large country cases.Whereas for exporter reported FOB prices, there are 58 countries reporting exports to 103 countries, which contains many smaller importers compare with CIF prices.Hence the exporter reported FOB prices is more suitable for the small importing country case, whereas importer report is more suitable for the big importer country case.
To sort this problem out and make the coefficients comparable across country pair and sectors, we restrict the bilateral trade flows to the same 50 importers and 58 exporter countries in either CIF or FOB datasets.Table 7 reports the regressions of equation ( 2) again, based on the narrower sample.
The results with CIF prices are listed on the left panel, and FOB prices on the right.For each regression, we test three wage series: IWW, IWW_OWW, and UNIDO.A striking feature of these results is that the estimated tariff rate coefficients improves to be significantly negative for both CIF and FOB prices, and much larger compare with the unrestricted sample.This significant net terms-of-trade gain strongly supports the classical theory of large country welfare gain with small tariff protection case.Wage rate still enters positively, though only significant for IWW_OWW wage series.The coefficient estimates of distance remain robust and significantly positive: increasing export distance by 10 percent will result in 11-13 percent increase in product quality, and a further 2-3 percent increase in transportation cost.So after controlling for other country specific factors which might affect unit values, about 80 percent of the observed variations in export unit values can be attributed to quality, whereas the pure distance effect is much weaker.The predominance of quality to unit values is also found by Feenstra and Romalis (2014), where more distant exporters will choose to sell products with higher unit values,.Notes: UNIDO wage is the average monthly dollar wages and salaries coming from INDSTAT3 database 2005 edition; IWW is the industrial world wage rates constructed by us based on the ILO yearly manufacturing sectoral wage data; OWW is the occupational world wage rates constructed by Freeman and Oostendorp (2000) based on the "October Inquiry" Survey of ILO wages; IWW_OWW stands for the IWW supplemented by the OWW uniform weighting wage rates.
Then to see how these relationships vary across industries, we attempt the regression again, breaking down by SITC 1-digit (SITC1).We omit products belonging to the ninth SITC1 industry "Not Elsewhere Classified".So all together we report regression for eight SITC 1-digit industries in Table 8.Across industries, a negative relationship between tariff and unit value is evident in six out of eight industries, indicating the TOT effect.Wage effect is positive for five out of eight industries, though only statistically significant for SITC1=4 "Animal and Vegetable Oils".Distance enters significantly positive for all the industries.The magnitude of quality improvement is highest for SITC1=5 "Chemicals and related products", where a 10 percent increase in distance will improve quality by 19 to 24 percent.The smallest quality improvement takes place in SITC1=3 "Mineral Fuel and Lubricant".Glimpsing through different estimates of distance between CIF and FOB prices, we can see that the pure distance elasticity is rather small overall, ranging from 2 to 7 percent across all the industries.In summary, our results demonstrate that unit values of internationally traded products are positively associated with distance across countries and industries.TOT gains due to tariff protection also prove to be large and significant, supporting large country case.More interestingly, the evidence of "Washington apple effect" is large and robust to a number of sensitivity analyses: faraway countries increase their export price by quality upgrading, and this effect is far more important in raising the unit values than distance and production costs.

Conclusion
This paper has focused on the supply side factors of product price changes.We decompose the variation of product unit values into TOT gain, pure quality effect, pure distance effect, and production cost effect.We construct a comprehensive dataset that interacts commodity trade, distance, tariff rates and production labor costs across countries.Our hypotheses find strong support by the data.The statistical analysis finds three strong and robust empirical relationships explaining product price variations.The first is that raising tariff rates do lower export prices.Interestingly, this negative relation is especially significant when we restrict our data sample to large importers in the world, which provides strong support to the classical large importer tariff protection case.So for a large importing country, implementing a tariff protection may indeed raise national welfare.The second is that exporter labor costs are in general positively associated with export unit values, though not significant for all the cases.The third result is that exports to faraway countries will have significantly higher unit values than goods shipped to nearby countries.But most interestingly, the "Washington Apple effect" dominates the pure distance effect, i.e. the price increases are mainly driven by quality upgrading instead of the increase in transportation cost.As a conclusion, on the supply side, quality effect is the largest contributor to unit value increase, compared with distance effect and production cost.So how should an exporter increase export prices?Our empirical findings suggest that a firm should target small importers, faraway destinations, and most important of all, improve quality!Much can be done for further research.There are many factors that can affect unit values and we only control for some of them.We can further control for the supply side factors such as common language, border, and trade agreements.Importer-demand condition can be added to further decompose the unit values, such as income level of import market, import price indices, importer trade balance, and macro indicators of comparative advantage.We can also utilize our full sample from 1964 to 2003 and attempt panel regressions.More interesting results on international studies will be generated based on this comprehensive dataset.

Table 1 :
Data coverage for three wage database

Table 2 :
Summary statistics of wages at SITC Rev.2 4-digit level

Table 2
further compares summary statistics of the three wage datasets based on the same SITC Rev. 2 nomenclature.The range of OWW and IWW wages are quite similar, but the average monthly UNIDO has a much larger variance, ranging from 6 cents to $20,050.

Table 3 :
IWW monthly $wage regress on UNIDO monthly $wage, Equation (3) : UNIDO wage is the average monthly dollar wages and salaries coming from INDSTAT3 database 2005 edition; IWW is the industrial world wage rates constructed by us based on the ILO yearly manufacturing sectoral wage data. Notes

Table 5a :
Summary Statistics of Importer CIF Price Dataset UNIDO wage is the average monthly dollar wages and salaries coming from INDSTAT3 database 2005 edition; IWW is the industrial world wage rates constructed by us based on the ILO yearly manufacturing sectoral wage data; OWW is the occupational world wage rates constructed by

Table 6a :
Regression equation 2a (Based on Importers reported CIF Price)

Table 6b :
Regression Equation 2b (Based on Exporters reported FOB Price)

Table 7 :
Regressions 2a and 2b (Based on Common set of Importers and Exporters)

Table 8 :
Regression 2 by Sector: Common set of Importer-Exporter Pair

Table 8h
: UNIDO wage is the average monthly dollar wages and salaries coming from INDSTAT3 database 2005 edition; IWW is the industrial world wage rates constructed by us based on the ILO yearly manufacturing sectoral wage data; OWW is the occupational world wage rates constructed byFreeman and Oostendorp (2000)based on the "October Inquiry" Survey of ILO wages; IWW_OWW stands for the IWW supplemented by the OWW uniform weighting wage rates. Notes