Model for international trade of sawnwood using machine learning models

The study tests the potential of machine learning models to analyse and forecast global bilateral trade flows of soft sawnwood by countries. The empirical trade flow data including annual import and export quantities and prices of soft sawnwood from 2000 to 2014 is obtained from FAOStat. We compare forecasting results from three methods, which can be classified as machine learning models: support vector machines (SVM), Neural networks and Random forests that is an ensemble of decision trees. The significant changes in the global sawnwood markets in the North America, Asia and Western Europe after the financial and economic crises of 2008-2009 raise the question to update the modelling of trade flows. The information on the global trade flow developments are important for decision makers involved in strategic planning and forecasting at the European Union-, national- or industry level. The aim of this study is to test methods previously quite rarely applied in the forest sector market modelling, but which could be helpful for analysts to visualize and examine a large amount of bilateral trade data to have a view on ongoing changes and to assess next years’ developments. The results also support the Finnish Forest Sector Outlook Studies, which are published biannually by the Natural Resources Institute Finland (Luke). As an example we present empirical results for Finland and for some of its main competitor countries and export destinations.201