Estimating price and income elasticities of demand for forest products: Cluster analysis used as a tool in grouping

Well-estimated elasticities of demand are important for making long-run projections in demand for forest products. In this research, cluster analysis is used to group 180 countries contained within the Global Forest Products Model (GFPM), using cross-sectional data for per capita gross domestic product (GDP), forest coverage, and per capita consumption of forest products, for forest products including plywood, particleboard, fiberboard, newsprint, printing and writing paper, and other paper and paperboard. The application of cluster analysis prior to estimating the elasticities of demand solves the problem of data availability in estimating elasticities by grouping countries based on variables identified from economics theory and enabling the extension of elasticity estimates to countries that are similar to others in a cluster, but without data for directly estimating elasticties. Mean absolute deviation is used for data standardization, and the k-medoids approach and silhouette technique are used in cluster analysis. Statistics of clusters for every forest product show various combinations of countries with similar levels of per capita GDP, forest coverage, and consumption, such as a cluster with high per capita GDP, low forest coverage, and high consumption of the discussed forest product. The results of the cluster analysis are validated by a one-way analysis of means and multiple comparisons. Countries for panel analysis are selected based on time series data availability and quality. As implied by cluster analysis, some of the countries in the cluster can be used to represent the whole cluster. In this research, long-run static models, short-run dynamic models, and long-run dynamic models of demand are estimated using panel data analysis for countries in each cluster using data from 1992 to 2007 and 9 to 44 countries in each cluster. We found that long-run dynamic elasticities are higher than short-run dynamic estimations, and dynamic model estimations outperform static model estimations as shown in RMSE statistics. (C) 2011 Elsevier BM. All rights reserved.