Improving Forecasts of the EGARCH Model Using EMD and K-means Clustering

This paper proposes a new semi parametric EMD-EGARCH model to solve the problem of accurate modeling of time series Volatility Prediction, and takes the stock market as the empirical object. This model is composed of EMD, K-means clustering and EGARCH model. First, we have used EMD algorithm to decompose the data into several components to increase the regularity of the data and reduce the difficulty of modeling. Second, we have combined these components into three series with K-means clustering, which represent the long-term trend, medium-term fluctuation and short-term fluctuation of the data respectively, so as to overcome the insensitivity of EMD to leverage effect. Finally, we have modeled the - three sequences using EGARCH model, and added their prediction results to get the final results. We take the stock markets of eight countries as the research objects. The empirical results show that in most markets (except China, which is too affected by the policy and the result is not ideal), compared with the classical EGARCH model, our model can significantly reduce the prediction error with the increase of the prediction period, and overcome the problem of insignificant leverage caused by the simple use of EMD method.