Global value chain challenges conventional wisdom on how we look at economic globalization and the policies that we developed under the old perspective of international trade. The first chapter develops a GVC-type CGE model that slices up the value-added embodied in final products with two key assumptions: 1) substitutability between internationally outsourced tasks and 2) capital-skill complementarity while labor is disaggregated into high-skilled, median-skilled, and low-skilled labors. The model is applied to WIOD dataset of 2014 to explore the question of whether protectionism helps domestic workers. Under the scenarios of the trade war between the US and China, simulation results find that a 10% tariff rate on Chinese consumer products would eventually hurt US workers in terms of the increased unemployment rate and the shrink of domestic outputs. The more the affected sector depends on outsourcing to China, the more losses the US bears if tariffs were imposed. Multi-region CGE models have widely adopted Armington's assumption of substitutability between imports from different countries, and used a two-level nested CES composition function. However, there are very few readily available 'micro' Armington elasticity estimations that are suitable for large-scale multi-country CGE models in the literature. The second chapter uses WIOD dataset to estimate Armington elasticity with and without adding domestic supplies in the commodity bundle. A modified Arellano-Bond GMM estimation shows that the micro elasticity which governs the substitution between different origins of imports is slightly larger than the macro one, which governs the substitution between domestic supplies and all the imports. Long run elasticities are in average five times larger than short run elasticities. Firms are relatively inelastic to importing origins in the short run, but get more elastic than consumers in the long run. After dealing with the issues of model structure and parameter uncertainty, the third chapter tries to combine machine learning with CGE model to cope with conflicting policy goals. Globally sustainable national policy design must consider multiple often conflicting sustainability goals. Computational constraints limit decision-support based on complex economic models and scenario analyses. The third chapter assesses whether surrogate modelling using machine learning can efficiently identify combinations of national policy choices that optimize global policy coordination problems. An Artificial Neural Network is trained using data generated from a global Computable General Equilibrium model. The trained ANN is used as a surrogate model to identify a quasi-optimal global policy mix for a combination of sustainability criteria. Different policy goals related to the global bioeconomy are considered in scenario analysis. These attempts can be used in the future to build a sustainability metrics system which is immediately applicable for CGE models and to examine the 1) the sustainable comparative advantage for each country, 2) impacts of global diversified bioeconomy policies and the trade-offs of conflicting sustainability goals (economic, environmental and social trade-offs). The main objective of combining machine learning techniques with sensitivity analysis based on CGE models, is to analyze the trajectory of bioeconomy growth and its implications on sustainability in a more holistic and comprehensive way.