使用差分方法提高预测精度——基于中国沪深300股票指数的实证分析

Using Difference Method to Improve the Prediction Accuracy—An Empirical Study Based on China Shanghai and Shenzhen 300 Stock Index

摘要: 当AR、ARIMA、ARMA、MA模型被用来进行时间序列未来值的预测时，由于误差项很容易被忽略掉，所以预测精度受到很大的影响。为了提高预测的精度，本文提出使用差分方法(DF)来处理自相关模型(AR)的误差项。通过对中国沪深300指数(Hushen300)进行实证研究发现：第一，AR-DF模型比单纯AR模型大大提高了HUSHEN收益指数实际值与预测值之间的相关系数；第二，AR-DF模型所得到的误差值比单纯AR模型要小很多；第三，在滞后阶数达到700时，AR-DF模型对于股票收益指数涨跌趋势的命中率比单纯AR模型提高了25.30个百分点，从50.99%提高到了62.65%。

Abstract: When prediction models such as AR, ARIMA, ARMA, and MA are usually used to forecast the future values of a time series, because the residual values are easily ignored, the prediction accuracy has been influenced a lot. For improving the prediction accuracy, this paper suggests use difference method (DF) to deal with the residual item of autoregressive model (AR). Based on the China Shanghai and Shenzhen 300 Stock Index (Hushen300), the empirical study has found: First, the AR-DF model can result in higher correlation between the real value of HUSHEN return index and its prediction value than pure AR model; second, the average residual value of AR-DF model is quite smaller than the pure AR model; Third, the hit ratio of AR-DF model has increased more than 25.30 percent points than pure AR model, which can increase from 50.99% to 62.65% when the lag order is up to 700.