Research On Risk Evaluation Of China’s Overseas Mining Investment Based On Deep Learning

One of the problems for China’s mining "go abroad" is risks facing overseas investment.For the long period of construction and payback,overseas mining investment projects often face many uncertainties,therefore it is of great significance for enterprises to control loss and protect overseas resource interests by scientific risk assessment.Previous studies have made attempts in risk evaluation,but it is a highdimensional classification problem and too complex for traditional models,and most research payed more attention to qualitative analysis.This paper introduces the idea of deep learning into the empirical research on the scientific issue of "how to use scientific methods to quantify and evaluate risks facing overseas mining investment ",The research work and innovative contributions are made as follows:(1)Setting up an index system for risk evaluation of overseas mining investment with quantitative evaluation criteria.Two data sets are proposed: the risk characteristics data set and risk label data set,and the former can be used as an open data source for subsequent research.This paper studies the characteristics of 34 risk factors and sets the measurement criteria: the measurement is scientific by reflecting the economic characteristics of various indicators;Data sources are reliable from authoritative databases such as the World Bank and the International Monetary Fund;Grading the scores of the calculation results,the results are comparable.It lays a data foundation for the subsequent work of risk evaluation.(2)An idea of deep learning is introduced to construct the risk evaluation model.With the Fraser Institute’s investment risk assessment as learning labels,and the 34-dimensional risk eigenvalues of the 21 sample mining countries during 2009-2016 as input to the deep learning model,five risk levels as output,we find the mapping relationship between risk characteristics and risk level in the deep structure and train a risk evaluation model based on deep learning.In order to solve the problem of network training caused by insufficient sample data,this paper improves the model by using parameter-based transfer learning method.Taking the deep learning model of financial risk assessment of A-share companies as the object,the model parameters trained by a large number of data in the source domain are applied to the target domain.The well trained model can make full use of the advantages of feature extraction of deep architecture,transform lower-level multi-dimensional features into higher-level and more abstract features through non-linear modules,which greatly improves the objectivity of evaluation.(3)A cluster analysis method is proposed to classify the mining countries based on investment risks.This paper combines Pearson correlation coefficient and Euclidean distance,and puts forward an algorithm of mining investment risk correlation coefficient.Countries with similar characteristics of investment risk are put into the same cluster according to the results of correlation test,to provide a clearer reference for investment decisions.(4)A method of global analysis of index contribution based on deep learning is proposed.The contribution rate of each risk indicator to the uncertainty is quantitatively analyzed and ranked.The basic idea is to obtain the weight distribution matrix of each layer of the verified deep learning model,and then design an appropriate algorithm to get the final contribution coefficient matrix.Unexpectedly we find that by the DNNbased machine learning,the contribution of each risk indicator to the evaluation results is dynamic at five risk levels.This is obviously more in line with the economic reality.It also proves from another point of view that the deep learning method has superiority in handling of complex relationships.We further optimize the indicators according to their contribution,and test the sensitivity of each selected indicator for the local dynamic monitoring on changes of investment risk in sample countries at specific risk level,to find out countries close to or exceeding the risk threshold in time.In general,this study applies deep learning to risk evaluation,which not only expands the application boundary of deep learning,but also provides a scientific solution for the risk evaluation and analysis of overseas mining investment.