Gold futures price forecast research----A combinatorial prediction method based on CEEMDAN-GARCH-SVR

Abstract

In recent years, the international form has become more and more complex, and various emergencies have constantly impacted the world economy. In this context, countries in the world need to maintain the stability of the domestic economic market and reasonably avoid financial risks. Gold plays a major role for both countries and individuals, and predicting the gold price is a prerequisite for making important decisions.In this paper, with the closing price of gold futures AU9999 from January 2,2008 to February 13,2023 as the research object. This paper mainly uses the traditional time series model for modeling, combined with the machine learning method, and introduces the decomposition reconstruction algorithm, hoping to achieve a better prediction effect. The traditional time-series model-error t-based ARIMA (2,1,2) -EGARCH (1,1) was established, and then the machine learning method SVR was used to model the data. In order to improve the performance of the model, this paper uses the CEEMDAN method to decompose the price sequence, and then decompose the IMF according to the high frequency sequence judgment condition summary into high frequency, low frequency and remaining term three sequence, make the original complicated sequence relatively simple, then with the traditional time series model for high frequency and low frequency modeling, SVR for the remaining term. Finally, we found that the decomposition modeling algorithm proposed here has the best prediction effect and made reasonable suggestions according to the prediction results.