Analyzing Heterogeneity in Estimates of the Armington Elasticity of Substitution Using Panel Data

This paper addresses the issue of neglected heterogeneity in the estimates of the Armington elasticities of substitution by means of development of an alternative model to the classical Armington approach, where values of the elasticity of substitution between domestic and foreign goods (so called “Macro” level elasticities) and elasticity of substitution between imports from different destinations (so called “Micro” level elasticities) are assumed to be homogeneous across countries. In the result of an in-depth literature review some of the determinants of Armington elasticities of substitution have been identified. These determinants could be broadly classified into 3 groups: sector-specific, country-specific and model-specific characteristics. Best practices obtained from the leading researchers in the fields of econometrics and trade analysis are employed in this paper in order to address issues of model misspecification. One of the key instruments employed to reduce omitted variable bias is Fixed Effect at importer, partners, time and group levels, capturing unobserved variables and reduce bias of the estimation results. However, some of the mentioned earlier determinants are vary across years and even within one year, causing heterogeneity in the estimates, therefore, they could not be adequately addressed by the classical model with FE. In order to capture these effects the alternative estimation model has been developed which accounts for differences in the countries’ levels of income and time horizon, addressing issues of the model misspecification. Furthermore, data from the leading institutions in the field of economics have been utilised for the estimations, contributing to the quality and reliability of the results. CHELEM and Trade Unit Values databases from CEPII provide data for trade volumes and trade prices (approximated by the unit values), and data from the World Bank gives indicators for the countries’ levels of income. The estimates have been performed only at the “micro” level as there was no data available at the level of aggregation used in the paper for the “macro” level. As the result of the analysis, estimates for 82 countries have been obtained for 43 GTAP sectors for the years 2000-2017. 29 of these estimates are significant at the .01 level. As for the alternative model estimates, variation is observed across levels of income and years, increasing number of the relevant estimates to 37 for some country groups in particular years at the same level of significance. Alternative model estimates vary in the range from 0 to 2.917, meanwhile the baseline model provides estimates ranging from 0.02 to 2.028. Despite confirmation of the original hypothesis from the preliminary results of the alternative model, an in-depth review does not support the presence of heterogeneity in the estimates of the “micro” level Armington elasticities of substitution. This is due to the average outperformance of the alternative model by 0.094 in comparison with the baseline, which is considered as a measurement error. Despite the acceptable performance of the baseline model on average, it is still recommended to employ the alternative model proposed for calibration of CGE models, as it provides higher degree of details for the larger number of sectors, allowing to obtain more detailed simulation results at sectoral level in particular year, rather than on average. Through these findings the research contributes to the field of the trade-related studies in the framework of the quantitative analysis, as it highlights the importance of calibration of CGE models, employed by the leading institutions such as the WTO, the UN, the World Bank, and other. Furthermore, the paper concludes with some limitations of this particular research and outlines scope for the further studies in this field. One of the proposed continuation of the research is to test sensitivity of GTAP Model with respect to the obtained elasticity estimates.