The International Spot Gold Price Analysis And Prediction Based On Empirical Mode Decomposition (EMD)

With the collapse of the bretton woods system, international spot gold prices gradually stepped into the stage of market-oriented pricing. Because the price of gold prices, is no longer the dollar since 2008, gold prices have also changed dramatically.Gold has the dual nature of commodity and currency, the gold price volatility mechanism is also different from ordinary commodity price changes. Therefore, the analysis of the price of gold and trend prediction has become a hot topic of research and theoretical and practical circles. Since May 2013, due to the influence of many factors,the international price of gold has experienced many times. In the new economic situation, gold investment has become one of finance and investment tools, can analyze and gold prices is extremely important. Investors only understanding the formation mechanism and gold prices to circumvent the risk of gold price fluctuations, get a good return on investment.This paper puts forward a kind of based on empirical mode decomposition(EMD)international spot gold price analysis and prediction methods. Based on May 22, 1992 to September 6, 2013, the international spot gold price sequence weekly closing price data as the research object, first of all, descriptive statistical analysis was carried out on the price of gold in order to understand the basic characteristics of the gold price, secondly using the EMD technique will price sequence is decomposed into eight different frequency components(including seven intrinsic mode function(IMF) and a residual item). First of all, in this paper, the component of the decomposed to do the following four analysis: 1, the white noise on the decomposed component inspection found that there are no random sequence, they contain certain information economics. 2, defines the fund’s volatility depicting the fund relative volatility in original sequence. 3, the IMF’s most has the "rush thick tail" form, by fitting the IMF found that the distribution of its obey the lognormal distribution. 4, analysis of the IMFs cyclical and correlation.Secondly, using the reconstruction algorithm will these different frequency components constitute three different components: low frequency components, the high frequency component and trends, from the three time scales on the international price of gold sequence through theoretical analysis and empirical analysis. This article conclusion is:the international spot gold price mainly by the long-term trend, major events and the market short-term fluctuations.Because the international spot gold price is nonlinear and non-stationary time series, so accurately predict the price trend is a more challenging job. Traditional statistical and economic models are based on data is linear assumption, it is difficult to capture the nonlinear model, hidden in the spot gold price usually can’t get accurate prediction result of international spot gold prices. In order to overcome the limitations of the traditional statistics and econometrics model. In this paper, based on the EMD and SVMs(support vector machine) nonlinear combination forecasting method. The method using the EMD technique will be the international spot gold sequence is decomposed into different frequency components, according to the frequency of high and low will makes the components in the three new sequence: represent the major events in the market price fluctuations, prices, trends; Against the three sequence, build different SVMs model respectively to forecast, forecast to sequence is obtained. Using the mean square error(RMSE) and mean absolute error(MAPE) each model the prediction error of size measurement experiments, using symmetric values direction(DS) all sequences of the model predicts the price of gold moves in the direction prediction ability. Finally, using SVMs for each sequence forecast build composite model was finally predicted. USES the international spot gold price data demonstrate the effectiveness of this method, and the international spot gold price data. Results show that this method and the single SVMs models compared with BP neural network model has higher prediction accuracy.