深度学习神经网络能改进GDP的预测能力吗?

Can Deep-learning Neural Networks Improve GDP Forecast?

目前国内外复杂的经济形势加大了预测GDP的难度,因此,如何有效地预测GDP是值得研究的重要理论与现实问题。有鉴于此,本文构建了既具有宏观经济理论基础又符合中国宏观经济特征的指标体系,并构造了一个用于GDP预测分析的LSTM模型,将之与BVAR模型进行对比研究,以科学地判断LSTM模型是否能够提升GDP预测的精确度。研究结果表明:第一,本文选择的扩展指标能够提升BVAR模型与LSTM模型的GDP预测能力;第二,相比于BVAR模型,LSTM模型能够更好地挖掘扩展指标对GDP的非线性影响,从而提升短期GDP预测能力。鉴于LSTM模型强大的自我学习能力、良好的泛化能力以及较好的模型可调节性,LSTM模型在GDP预测领域具有广阔前景。

The prediction of GDP growth rate is an important research field of macroeconomics.However,the complex economic situation at home and abroad makes it more difficult to forecast GDP,so how to forecast GDP effectively is an important theoretical and practical problem worth studying.This paper constructed with the macroeconomic theory basis and in line with the characteristics of China s macroeconomic index system,and constructed a LSTM model for GDP forecast analysis,and compared with the BVAR model,in order to scientifically judge whether the LSTM model can enhance the accuracy of the GDP forecast.The results are shown as follows.Firstly,the expansion index selected in this paper can improve the GDP forecasting ability of the BVAR model and the LSTM model,so it is necessary to take into account the expansion index reflecting China s national conditions when forecasting China s GDP.Secondly,compared with the BVAR model,the LSTM model can better explore the nonlinear influence of extended indicators on GDP,so as to improve the short-term GDP forecast ability.On this basis,because of strong self-learning ability,good generalization ability and good model adjustability,the deep-learning neural network has a broad prospect in the field of GDP forecasting.