基于混合人工神经网络的人民币汇率预测研究——兼与ARMA、ARCH、GARCH的比较

Research on Prediction of RMB Exchange Rate Based on Hybrid Artificial Neural Network: A Comparison with ARMA,ARCH and GARCH

随着人民币汇率市场化程度不断提高,其波动程度也不断增大,对人民币的预测显得越来越重要。近几年来,人工智能在许多领域都取得了巨大的成功,证明了自身的优越性,作为其主要组成部分的人工神经网络(ANN)模型已经逐渐被引入金融资产价格的预测研究中。本文将原本仅适用于二值型数据的Adaboost算法进行了优化,使其也能适应连续型数据,并用其确定混合模型的权重,解决了过往大多数研究中混合模型权重设定较为主观和随意的问题。在此基础上,本文融合了广义回归神经网络(GRNN)撞长预测趋势因素,而误差反传神经网络(BPNN)擅长预测随机因素的优点,组成了比单一神经网络模型更为强大的GR_BP_Adaboost强预测模型。最后,以均方误差(MSE)、平均绝对误差(MAE)和DM检验为标准,将GR_BP_Adaboost模型对人民币汇率的预测结果与传统的ARMA、ARCH和GARCH模型进行了对比,所有结果均表明GR_BP_Adaboost强预测模型的预测能力显著优于其他模型,说明人工智能预测技术相较于传统方法具有较大优势,也说明汇率市场不是弱式有效。

With the continuous improvement of the marketization degree of RMB exchange rate,the volatility of RMB exchange rate also increases and the prediction of RMB becomes more and more important.In the recent years,artificial intelligence has achieved great success in many fields and proved its superiority.The ANN model,which is the main component of artificial intelligence,has been gradually introduced into the prediction research on financial asset prices.In this paper,the Adaboost algorithm which is only applicable to binary data is optimized,so that it can adapt to continuous data and beused to determine the weight of the mixed model.It solves the problem that the weight of the mixed model is more subjective and random in most of the previous studies.On this basis,the paper combines the advantages of GRNN which is good at predicting the trend factors and BPNN which is good at predicting the random factors,and constitutes the GR_BP_Adaboost strong prediction model which is tronger than the single neural network model.Finally,based on MSE,MAE and DM tests,the paper indicates that GR_BP_Adaboost is more accurate than traditional models of ARMA,ARCH and GARCH,revealing that the artificial intelligence prediction technology has a greater advantage than the traditional method and the exchange rate market is not weakly effective.