季节ARIMA模型与LSTM神经网络预测的比较

Comparison Between Seasonal ARIMA Model and LSTM Neural Network Forecast

宏观经济预测一直是经济领域研究的热点和难点,随着深度学习技术的发展,长短期记忆(LSTM)神经网络已广泛应用于经济、金融等诸多领域。文章分别用传统的统计计量方法(ARIMA模型)和深度学习方法(LSTM神经网络)建模,预测中国2019年GDP季度数据。结果表明:(1)并非所有的LSTM建模预测都优于传统统计计量方法,应根据数据特点,选择适合的预测模型,通过对GDP的预测,ARIMA模型取得了更好的效果。(2)针对结构比较简单的时间序列数据,传统计量方法能得到较好的效果;针对结构复杂的非线性多变量数据,特别是非结构化数据,LSTM神经网络可以取得不错的效果。(3)根据GDP预测结果推算,2019年中国GDP增长率为6.49%,虽比上年下降0.11个百分点,但仍保持中高速平稳增长。

Macroeconomic forecast has always been a hotspot and difficulty in economic research,and with the advancement of deep learning,long short-term memory(LSTM)neural network is widely used in the fields of economy and finance.This paper uses traditional statistical method(ARIMA model)and deep learning method(LSTM neural network)respectively for modeling to forecast the quarterly GDP data of China in 2019.The results are shown as follows:1)Not all the LSTM modeling forecasts are superior to traditional statistical methods,and the suitable forecast model should be selected according to data characteristics;ARIMA model performs better through the GDP forecast.2)For time series data with relatively simple structure,traditional quantitative methods can achieve better performance,while LSTM neural network can achieve relatively good results for nonlinear multivariable data with complex structure,especially for unstructured data.3)According to the GDP forecast results,China’s GDP growth rate in 2019 is 6.49%,maintaining the steady medium-high speed growth,despite a projected decrease of 0.11%compared with last year.