Appendix 5: Data and Modeling Strategy
- Nagwa Riad, Luca Errico, Christian Henn, Christian Saborowski, Mika Saito, and Jarkko Turunen
- Published Date:
- January 2012
The objective of this exercise is to quantify at a high level of product detail the implications of sectoral and supply chain linkages on the trade impact of relative price changes. The model employed for the analysis uses a comprehensive data set including information on imports by trading partner at the HS2002 six-digit level sourced from UN Comtrade. We have chosen the year 2008 for our analysis because (i) a larger number of countries have to date reported data for 2008 at this level of disaggregation than for 2009 and (ii) trade flow composition was presumably less distorted by the crisis-induced collapse in trade.
We model trade response to changes in relative prices in two stages. In the first stage, a partial equilibrium model is used to determine change in consumer demand for import in every import market. The second stage uses the input-output tables discussed in the main text to adjust imports in response to the trade-induced change in aggregate demand.
This modeling approach offers three key advantages: (i) the aggregation bias implicit in CGE models using aggregate trade flows is avoided; (ii) the model is simple and transparent in its assumptions; and (iii) the simulation exercise illustrates changing trade patterns at a high level of product detail which allows quantifying sectoral and supply chain linkages and their importance for changes in trade patterns. In response to a change in relative prices in a given economy, the model produces the impact on global trade flows at the HS six-digit level. The results can then be aggregated to any given level of product detail and used to compute ex ante and ex post indicators of RCA, export similarity (ESI), and the technology content in a country’s exports and imports.
The greater product detail and transparency of this modeling approach comes with a cost that renders the model useful primarily for the analysis of sectoral and supply chain linkages. Each product is modeled as a separate market and in isolation from other markets, and inter- and intrasectoral linkages or economy-wide impacts of changes in relative prices that are likely to take place are not fully captured. These limitations need to be kept in mind when examining the model’s predictions at the aggregate level.
First stage: A partial equilibrium model of import demand
In determining the change of export prices as a result of a real exchange rate shift, we take account of the fact that these are not affected symmetrically across sectors. We reiterate the argument made in the main text, namely that the foreign content (intermediates and value added) in exports differs not only by exporting country but also by sector. In particular, we highlighted that the foreign content in Chinese exports is higher in sectors that are commonly associated with high-technology goods. This implies that exports in these sectors are relatively less affected by an exchange rate shift. We integrate this argument into our methodology by assuming that export prices in a given sector change by the share of domestic content in the export value multiplied by the exchange rate shift. For instance, a 10 percent exchange rate depreciation leads to a 5 percent fall in export prices in a sector with 50 percent domestic content in exports, and by 8 percent in a sector with 80 percent domestic content.42
We utilize a simple partial equilibrium setup that is similar to the model used in Brenton and others (2011) and Lim and Saborowski (2010) and extend it to a multicountry setting. The framework allows analyzing the response of trade flows to changes in relative prices in a transparent way and at a high level of product detail. Our analysis therefore refrains from using a multicountry CGE model which would require modeling complex interactions for a large number of variables and countries and sacrifice our high level of product disaggregation. Our approach has the advantage that it requires only a limited set of parameters to be determined, namely the trade elasticities involved.
The model focuses on the import market of every product in every economy in isolation. The setup is based on a representative consumer with Armington (1969)-style preferences, who makes choices over imported goods in response to price changes in two consecutive steps: first, she substitutes between different exporters’ national varieties following relative price changes, and second, she changes her overall demand for the good in question as a result of the change in the average price of the product.43 The ex ante price of all product varieties is normalized to unity. Thus, if the percentage change in relative prices is x, the consumer price of each variety becomes 1 + x.44 A similar setup has been widely adopted in applied trade models, including single- and multicountry CGE models.
The model relies on six core assumptions. First, as is standard in consumer demand theory, sector-level elasticities are used to determine the magnitude of the demand response of trade flows to relative price changes. Second, the calculations are based on the standard Armington (1969) assumption of imperfect substitutability between imports from different trading partners (within each product category). Third, a change in relative prices is defined as a change in relative prices facing the consumer in each importing economy.45 Fourth, no direct substitution between different products is allowed (i.e. each product is modeled as a separate market and in isolation from other markets). Fifth, our parameterization of the model is aimed at computing long-term impacts of relative price changes. Finally, and given the partial equilibrium nature of the exercise, inter- and intrasectoral linkages (e.g., factor reallocation) or economy-wide impacts of changes in relative prices cannot be considered.46
Sector-level elasticities of substitution are used to determine the magnitude of substitution between exporters of a given good in each import market. The literature on elasticities of substitution is rich but provides estimates that differ widely in magnitude. Broda and Weinstein (2006) and Broda, Greenfield, and Weinstein (2006) provide the most comprehensive set of elasticities by importer and at the five-digit SITC product level. However, a more recent literature that allows for firm heterogeneity in structural models (e.g., Crozet and Koenig, 2010) suggests that elasticities may be lower on average and may lie in the range of –2 to –3. This range is also suggested by Obstfeld and Rogoff (2005) who use these values for their analysis of the exchange rate change needed to close the U.S. current account deficit. In general, this range of values is more consistent with what is typically found in studies focusing on aggregate impacts of changes in relative prices (Gagnon, 2007). In the light of these findings, we use the Broda and Weinstein (2006) and Broda, Greenfield, and Weinstein (2006) elasticities scaled to a mean of 2.25 as attained in Crozet and Koenig (2010).47 This results in an import-weighted mean elasticity of 2.4.48
Country-specific import demand elasticities at a high level of product detail are used to determine the magnitude of the demand response to relative price changes in each import market. Kee, Nicita, and Olarreaga (2008) provide a comprehensive set of price elasticities of import demand by import market and six-digit HS level of product detail. Notwithstanding the fact that their estimates are somewhat larger than is typically found in the literature, their study provides a carefully estimated and comprehensive set of elasticities unmatched in its high level of product detail. We therefore use their elasticities but scale them to a mean of –1 which results in an import weighted mean of –0.79.49 This value lies within the consensus range established by the empirical literature.50
Second stage: Adjusting imports for the trade-induced change in aggregate demand
Changes in exports have an impact on total value added and aggregate demand. In the first stage of our modeling framework, an exchange rate shock in a given country affects both imports and exports. But an important link is missing, namely the impact of changes in aggregate demand (resulting from falling or rising exports) on imports. We account for this shortcoming in the second stage of the analysis. The predicted changes in imports and exports from the first stage are used as a starting point.
We use input-output tables to determine how changing exports affect value added as well as imports of intermediates and final goods (see Appendix 3). We first use input-output tables to determine the fall in value added that is consistent with the change in exports resulting from the first stage of the analysis. The same tables are then used to determine the change in imports of intermediates and final goods (by sector) that results from falling/rising aggregate demand. The input-output tables distinguish two-digit ISIC sectors in the analysis as well as intermediate and final goods therein. Since the level of disaggregation of our trade data is higher, the resulting sectoral impacts are then split up across subsectors according to market share.
We do not adjust the magnitude of the import price change in the country under consideration by the share of intermediates in its imports and in other countries’ value added embodied therein, due to lack of reliable data on this phenomenon. This limitation is likely not to distort the results severely as long as this intermediate content is relatively small.
The total price change for a given good is computed as a weighted average (by market share) of the price changes of the different product varieties.
In other words, a change in relative prices is not the same as an exchange rate change whenever pass-through is incomplete.
Implicit in our model is one additional technical assumption: since demand responses are based on elasticities, there will never be market entry by new exporters as a result of price changes (zero trade flows will always remain unchanged at zero).
We remove outliers by capping the elasticities at a value equal to the mean plus 2 standard deviations. We then scale all elasticities to arrive at the desired mean of –2.25.
The literature on substitution elasticities (e.g., Broda and Weinstein, 2006) has shown that estimates tend to be significantly higher when estimated at a higher level of product detail. These elasticities provide correct changes in import quantities at the high level of disaggregation at which they are estimated. However, changes in a country’s total imports obtained via simple summation of these changes at high disaggregation levels would be considerably biased upward. This is because the summation cannot account for cross-product substitution in response to relative price changes. To minimize this possible bias, we therefore work with elasticities estimated at the more aggregate SITC three-digit level at which cross-product substitution is likely to be minimal. Crozet and Koenig (2010) find a mean elasticity of –2.25 at the three-digit level in a theory framework with heterogeneous firms set out by Melitz (2003), whereas Broda and Weinstein (2006) find one of –4. The values found in Crozet and Koenig (2010) also lie closer to what is typically found in time series estimation (Gallaway, McDaniel, and Rivera, 2003; Saito, 2004) and to the values used in Brenton and others (2011) and Lim and Saborowski (2010).
We initially remove outliers by capping the elasticities at a value equal to the mean plus 2 standard deviations and fill in missing observations using product elasticities at higher aggregation levels, at the country level, where available. We then scale all elasticities to arrive at the desired mean of –1.
Goldstein and Khan (1985) give a comprehensive survey of the early literature on price elasticities of import demand. Their conclusion is that the average long-run import demand elasticity lies somewhere between –0.5 and –1. Reinhart (1995) estimates long-run import demand elasticities for 12 developing countries from 1970 to 1991 using cointegration techniques. She obtains an average elasticity of –0.6. Aziz and Li (2007) find an import demand elasticity of –0.9 for China. They use quarterly data from 1995 to 2006 on total Chinese imports (from all trading partners and products) as the dependent variable. Hong (1999) provides sample import price elasticities used in the LINK modeling system for different countries. They range between –0.4 and –1. Brenton and others (2009) and Lim and Saborowski (2010) use an elasticity of –0.5, albeit in a short-term setup.