Prediction For Shanghai Non-ferrous Metal Futures Prices Based On Support Vector Machine

Futures market is an important part of the socialist market economy, and it isplaying the important functions of price discovery and risk-averse in thenational economy. Effective prediction of the futures price can help to masterthe rules of price change, guide investment and hedging correctly and allocatethe social resources rationally. However, the complexity of the internalstructure of the futures market and the volatility of various factors has broughtabout many difficulties to futures price forecasting. Further more, the results oftraditional forecasting methods are not very satisfactory. With the rise ofartificial intelligence technology and computer technology, there are manyexcellent futures price forecasting methods emerge in response to the needsof times. Especially the development of statistical learning theory and supportvector machine, providing a new direction for futures price forecasting. Thesupport vector machine was used in pattern recognition problem at first.However, support vector regression also show an excellent performance inregression problems in recent years. At present, the application of supportvector machine of the non-ferrous metals futures price forecasting is still blank.This paper considers Shanghai non-ferrous metals futures as the researchobject, proposes a price forecasting model based on support vector machineand uses particle swarm optimization to optimize parameters of the supportvector machine model.Firstly, this paper starts with the background knowledge of the futuresmarket, study the futures varieties, futures trading functions, and futures priceforecasting methods, then it analyze various factors which impact on Shanghainon-ferrous metals futures prices. After that, this paper put forward the mainfactors as follows: supply and demand factors, the macroeconomic situation,the impact of the relevant market (including spot market and the internationalfutures market), the relevant commodities such as crude oil price fluctuations,and import and export policies related to non-ferrous metal, exchange rates,non-ferrous metal production costs, the change of non-ferrous metal application trend, the direction of fund transactions and other factors.Afterward this paper extracts many influencing factors that can be used asinput indicator of predict system and establishes a futures price forecastingmodel based on support vector machine.Secondly, in order to reduce the complexity of the model, this paperoptimizes the input vector through principal component analysis method.Principal component analysis is designed to use lower-dimensional mind toconvert many related indicators into a few highly independent or unrelatedcomprehensive indicators. Thus it not only can keep the integrity of datamessage as much as possible, but also can improve the model runningefficiency.Thirdly, with regard to parameters choice problems, this paper describes therelationship between structural risk minimization, experience risk minimizationand parameters choice by means of experiment. Then, the paper finds areasonable value of penalty coefficient C according to the experiment results.After that, the paper uses particle swarm optimization to optimize parametersof the support vector machine model. First, set a reasonable range for eachparameter according to the result of three-parameter optimization. Second,assign random value which within the range to initial particles so as to enhancethe speed of particle optimization.Finally, this paper uses Shanghai non-ferrous metals futures data forempirical studies, predicts the prices of copper futures, aluminum futures andzinc futures respectively. The empirical results show that support vectormachine prediction model which based on particle swarm optimizationalgorithm can achieve satisfactory effect.