An evaluation of gravity models and artificial neuronal networks on bilateral trade flows in wood markets

Trade fuels economic development in interwoven international wood markets, while economic shocks and structural changes jolt market response behavior. In this context, both accurate predictions and forecasts of trade flows and a deep understanding of their influencing factors are essential for policymakers and stakeholders to enhance economic planning and decision-making affecting trade policies. A popular method for analyzing bilateral trade flows is the deterministic Gravity model of trade due to its intuitive design and effectiveness. However, data-driven machine learning methods such as artificial neural networks (ANN) could enhance the accuracy of deterministic modeling approaches through their complex and potentially nonlinear nature. To the best of our knowledge, no study exists that uses an ANN approach to assess bilateral trade for different wood-based products was. Therefore, it remains unclear whether ANN is an appropriate method to predict and forecast trade flows in forest product markets or if Gravity models of trade might yield better results. This study compares the ability of Gravity models and feedforward neuronal networks (FFNN) to predict existing and forecast future bilateral trade flows of four main product categories in international wood product markets. Our findings highlight that it is essential to consider the purpose of the analysis alongside the specific product group under investigation. The FFNN approach outperforms Gravity models for predicting past and present trade flows, delivering more accurate predictions across all product categories. Looking at the accuracy of forecast, we see that the superiority of FFNNs is present but decreases as the forecast horizon increases.