Multiscale Analysis Of Financial Time Series Based On EEMD

Recent years, with the development of financial market and investments, quantitative investment is gradually stepping into people’s vision, which has pushed econometric analysis to an unprecedented height. Financial time series is the real measurement of markets. The potential laws of markets can be found by quantitative analysis, which is the basis of asset pricing, strategy making, product designing and risk management. Concerning the nonlinear, non-stationary and multi-scale characters of financial time series, ensemble empirical mode decomposition (EEMD) was applied to analyze financial time series in this paper.Firstly, a multi-scale integrated model was proposed based on EEMD. The original time series was decomposed and constructed into a high frequency part, a low frequency part and a trend part based on EEMD. And the three parts were predicted by Elman neural network, SVM and GM(1,1) model respectively. Finally, the prediction results of the original time series was the superimposition of the respective prediction. The empirical study shows that the new integrating model is better than the traditional ones.Secondly, a research on volatility spillover effects between stock market and exchange market from the viewpoints of time and frequency was proposed based on EEMD. The sequences of stock price and exchange rate were decomposed and constructed into three parts:high frequency represents the short-term market fluctuation, low frequency represents the middle-term market fluctuation, and long term trend. In every scale, the granger causality test and the time-varying copula are used to describe the direction and strength of volatility spillover. The results show that:the high frequency part plays the most important part in causing the volatility of the original signal serial. The direction and strength of volatility spillover vary cross scales.Finally, a denoising arbitrage strategy was proposed based on EEMD. The spread signal was regarded as two parts:noise and trend. The trend part can be got by denoising the spread with EEMD. According to the theory of mean-reversion, the arbitrage strategy was built based on the fluctuation of spread around trend part. The back-test has obtained good effect. Besides, the new strategy avoids the problem of parameter selection compared with traditional wavelet denoise method.