BDI Prediction Based On EMD-XGBoost Model

With the development of international trade,shipping industry plays an increasingly prominent role in the trade market,among which dry bulk shipping market has become the most important part of the international shipping market.Dry bulk cargo transportation has become a widely favored mode of transportation due to its advantages of high efficiency,variety and low cost,and has undertaken more than 40% of the transportation tasks of maritime trade.Dry bulk shipping market is seasonal,cyclical,highly volatile and capital intensive,which makes the fluctuations of dry bulk shipping market complex and diverse.The Baltic Dry bulk freight Index(BDI)has become the most important index of the shipping market,which can not only reflect the spot market of the dry bulk shipping market,but also reflect the prosperity degree of international trade.It is of great guiding significance to analyze the dry bulk shipping market deeply and judge the future trend of BDI index and the influence of other influencing factors on BDI index.Based on May 1,2013 to October 31,2020 the BDI index as the research object,the first systematic analysis of the influence factors of BDI index,select the metal market price,agricultural products market price,the price of dry bulk ships,energy market price and economic environment,and other indicators,fully describe the influence factors of BDI index set.Secondly,XGBoost model is used to mine the correlation and importance degree between various influencing factors and BDI index.Data from May 1,2013 to May 3,2019 are used as training set,and data from May 3,2019 to October 30,2020 are used as test set to predict BDI index.Then,the BDI index was divided into different IMF components according to frequency by EMD decomposition algorithm,and the components were reconstructed into the high frequency part,low frequency part and trend term of BDI index,and the BDI index was predicted again based on EMD-XGBoost model.Finally,through comparative analysis of other machine learning models such as support vector machine,multi-layer perceptron and random forest,it is found that the determination coefficient of BDI index predicted by XGBoost model is 81%,and the root mean square error is 121.27,which is more outstanding than other machine learning models.After decomposition and reconstruction of EMD algorithm,the predicted determination coefficient of EMD-XGBoost model reaches 87%,and the root mean square error is 86.24.Other machine learning models after decomposition and reconstruction of EMD algorithm also have different degrees of improvement.The empirical results of this thesis show that :(1)the price indices in different markets all have an impact on BDI index,among which the indices with higher importance are WTI crude oil price index,62% iron ore price index and thermal coal price index.(2)XGBoost model can better capture data features when predicting BDI index,and has better fitting effect and smaller prediction error compared with other machine learning models.(3)The different parts of BDI index after EMD decomposition and reconstruction are more stable and have lower volatility,which can further improve the prediction accuracy of XGBoost and other machine learning models.This indicates that XGBoost model and other machine learning models can achieve better results in predicting BDI index after decomposition and reconstruction by EMD algorithm.