基于WASD神经网络的日本国债预测结果与分析

Forecast Results and Analysis of Japanese Total Central Government Debt Based on WASD Neural Network

日本是发达国家中国债和GDP的比值最高的国家。在过去的20年中,日本国债规模增长迅速。根据日本财务省官方公布的数据显示,截至2020年6月底,日本国债总额已达到1159.0289万亿日元。为了预估日本中央政府未来的债务风险,对日本国债规模进行预测是有必要的。研究利用权值和结构确定(WASD)神经网络对日本2017年3月—2020年6月一共14个季度的国债规模成功进行了预测,预测结果与官方数据的相对误差在4%以内。此外,利用校验误差最小的神经网络更进一步地完成了2020年9月—2022年12月一共10个季度的短期预测。结果表明,短期内日本国债规模仍将缓慢上升。

Japan is the country with the highest debt-to-GDP ratio among developed countries.Japanese total central government debt(TCGD)has grown rapidly over the past 20 years.According to the official data released by the Ministry of Finance of Japan,the TCGD has reached 1159.0289 trillion yen by June 2020.In order to estimate the future debt risk of Central Government of Japan,it is necessary to predict the amount of the TCGD.In this paper,a short-term prediction of TCGD for 14 quarters from March 2017 to June 2020 is completed with the weights and structure determination(WASD)neural network,and the relative errors between the predicted results and the official data are below 4%.In addition,it makes a 10-quarter prediction from September 2020 to December 2022 by the neural network with the minimum validation error.The results show that the scale of Japanese central government debt will continue to rise slowly in the short term.