Are Armington elasticities different across countries? A cross-country study for European trade elasticities

CGE models are a widely used and accepted technique for policy evaluation and impact analysis. The modeling technique is especially useful in the analysis of trade reforms, tax reforms, energy sector reform and development policy analysis. However, the results of such models are often argued to be sensitive to the choice of exogenous parameters such as elasticities. Apart from the elasticities of substitution between production factors in the production function, the so-called Armington elasticities which determine the substitutability between domestic goods and imports are often mentioned as one of the caveats of CGE models. McDaniel & Balisteri (2002), Schuerenberg-Frosch (2012), Siddig & Grethe (2012) and others show that the choice of the so-called Armington elasticities in the import demand function has a strong influence on the simulation results. Hence, it is very important to choose these elasticities in a sensible way. Unfortunately, many CGE papers are not very transparent concerning the choice of elasticities and the sensitivity of the results with respect to this choice. As e.g. Welsch (2008) points out ”In practice, the elasticities employed are frequently based on 'guestimation' or on estimates picked from the literature.” Armington (1969) and most CGE modelers use the constant elasticity of substitution (CES) function to model demand for domestic goods and imports. The Armington elasticity is hereby defined as the proportionate change in the ratio of quantities divided by the proportionate change in the marginal rate of substitution in demand between domestic and foreign goods. There exist a number of estimations for Armington elasticities and the results of these are frequently used in CGE studies. In this paper we argue that this strategy could lead to severely biased model results as the estimated elasticities might not be applicable to either the specific model or the country in question. The reasons we give for this argument are the following: Most of the existing studies provide results only for the U.S. Even among the estimated elasticities for the U.S. there is some variance found. More importantly, the very few studies for other countries[such as Gibson (2003), Welsch (2006, 2008)] find substantially differing results. But studies for other countries are very scarce. Thus, the often formulated argument that time-series studies find rather small elasticities might simply be driven by rather small elasticities in the specific U.S. case. Moreover, the higher elasticities for other countries (e.g. Gibson (2003)) can be explained by the fact that these studies are more recent and consider the effects of globalization, market integration and increasing competition, which lead to higher substitutability between domestic and foreign goods. Welsch (2006) argues that the Armington elasticities decrease over time due to intra-industry specialization among open economies. He also finds indications for this hypothesis in French data. Thus, elasticities from older studies (e.g. from the 1990s or earlier) might not be useful in models based on recent data as the trade pattern and trade motives might have undergone important changes since then. One result that emerges quite clear from the literature is that elasticities differ depending on the level of aggregation used in the data. It is uniformly found across most of the studies that elasticities tend to be higher the more disaggregate the underlying data is. Thus, a CGE modeler should use estimated elasticities from a study with the same level of sectoral disaggregation he uses in his model. However, the mentioned studies for the U.S. have a rather high level of disaggregation with 180-200 industries included. Most CGE studies are much more aggregate. Nonetheless, as McDaniel & Balisteri (2002) points out, authors simply calculate the average elasticity across subsectors and use this number for their aggregated sector. This might lead to an aggregation bias and thus to biased CGE results. Bloningen & Wilson (1999) investigate the determinants of Armington elasticities. In addition to sector-specific effects they also find country-specific determinants such as trade policy. This implies that the usage of elasticities from another country might be misleading. In addition, a comparison of estimated elasticities across countries is very difficult as the studies often not only differ in the country but also in the degree of disaggregation, the method applied, the time horizon, the data frequency and even the underlying structural model. This paper aims at providing additional insights in the aforementioned aspects by providing estimated elasticities based on recent data for a larger group of European countries. We focus here on elasticities for CGE modeling. Thus, we aggregate our data to the 2-digit level of NACE Rev.2 which is the degree of disaggregation also used in the EU and OECD SAMs and thus used in many CGE studies for these countries. We also derive our functional form from these models. Using cointegration analysis we estimate the first order condition resulting from cost minimization or utility maximization subject to the CES subutility function in imports and domestic goods. The results show a rather big variance across sectors and countries. The elasticities are between 0 and 3.7 what indicate a plausible magnitude comparable with the recent studies of Welsh (2006) and Saito (2004). Thus, there are significant differences between the minimum and maximum values for different countries. While for Austria and France the estimates are in the interval from 0.5 to 3.7 and 3.2 respectively, the values for Italy reach only a maximum of 1.6. These large differences are also observed for individual sectors. For instance, the elasticities for wood and cork products vary from 0.9 in France to 2.11 in Finland. The same can be observed for beverages (from 1.9 in Finland to 3.7 in Austria) and paper products (from 1.6 in France to 2.95 in Finland). Our results indicate higher values of Armington elasticities in comparison to the U.S. studies from the 1980s and 1990s despite a higher level of aggregation and the usage of annual data (i.e. a lower frequency). This confirms the statement that elasticities have increased since the 1980s due to increased internationalization of production and increasing competition on world markets. Moreover, this also confirms that higher values are found for countries outside the USA, just like the finding of Gibson (2003) for South Africa. However, the significant cross-country differences illustrate clearly that it is not acceptable to use estimated elasticities for another country when specifying a CGE model - which is very often done in practical CGE work. See above See above