摘要
投资者关注在股市中的作用是近年来研究的热门问题之一.文章创新性地将百度指数作为中国市场投资者关注度指标加入以差分进化算法优化的极限学习机(DE-ELM)中,研究百度指数对中国股票指数的预测能力.实证结果显示,差分进化算法极限学习机(DE-ELM)模型的预测能力较传统计量模型ARIMA模型和传统神经网络模型BP神经网络模型显著提高,且加入百度指数能够提升DE-ELM模型对股指收益率的预测精度,其中以加入“牛市”、“熊市”和“金融危机” 3个百度指数的差分进化算法极限学习机(DE-ELM)预测精度最高,结果最稳定.
Financial time series forecasting models based on traditional statistic theories have been applied to forecast stock indexes with the search volume.However,these models cannot identify the nonlinear relationship in the real stock market.This paper constructs an extreme learning machine(ELM) optimized with the differential evolutionary algorithm(DE) to examine the Baidu Indexes’ forecasting ability.Three market-related Baidu Indexes are selected to predict four different Chinese stock indexes’ returns.Empirical results suggest that DE-ELM model has a better prediction ability.Furthermore,Baidu indexes can improve the prediction accuracy of DE-ELM model,and the DE-ELM model extended with the Baidu index triplet has the highest forecasting accuracy.
作者
唐旻
黄志刚
TANG Min;HUANG Zhigang(School of Economics and Management,Fazhou University,Fuzhou 350108;Fujian Fintech Innovation Major Laboratory,Fuzhou 350108)
出处
《系统科学与数学》
CSCD
北大核心
2022年第6期1503-1518,共16页
Journal of Systems Science and Mathematical Sciences
