Futures Index Forecast Based On Scene Superposition And Support Vector Machine

With the development of China's economy,the demand for raw materials is growing.As an important part of the financial market,the futures market trades for raw material futures,and the futures market has the functions of risk aversion,price discovery and asset allocation.Therefore,it is very important to study the trend of price change in futures market for the steady development of enterprises.The time series data of futures price is the historical transaction data formed by the exchange of futures market after a certain period of time.It is formed by multiple factors.It is the external expression of the movement characteristics of futures market in the past period.This paper chooses the time series prediction of futures index as the research problem.Copper is an active futures,so the paper takes Shanghai Copper Index of Shanghai Futures Exchange as the research object,extracts the 17-year historical closing price trading data of the futures from 2000 to 2017,and carries out empirical research on this data,makes full-time prediction,aiming at pursuing higher yield.Based on the idea of scene superpostion,the paper establishes a combination model to forecast futures index.The main tasks are as follows:I A forecasting method of scene superpostion is proposed.By summarizing the literature,time series forecasting methods can be divided into three categories:traditional model forecasting,intelligent model forecasting and combination model forecasting.Single model can play a good prediction effect in some specific and short-term scenarios.Combination model prediction can make up for the defect of single model prediction.But before forecasting research,time series are often processed in segments,and the results of different sub-series are synthesized.The way of scene superposition is to define different scenarios by strict parameters,classify the sample points of time series according to scenarios,establish descriptive variables to describe and study,and finally overlap the results of scenarios,so as to avoid segmentation.II Choose the appropriate fitting model by multi-dimensional nearest neighbor comparison method.In the paper,moving average model,exponential smoothing model and artificial neural network model are used to fit the time series.The futures market is forecasted according to the fitting inflection point of the fitting curve.When the maximum inflection point occurs,the futures market price is forecasted to decline in the future,and when the minimum inflection point occurs,the futures market price is forecasted.Rising;Calculating the annualized rate of return of these three models under different parameters,obtaining the optimal parameters of the model through annualized rate of return;By comparing the annualized rate of return and Sharp ratio of these three models under the optimal parameters,screening the model,arranging the artificial neural network model with lower annual rate of return and Sharp ratio.In addition,because the moving average model and the exponential smoothing model have similar index effects,the multi-dimensional nearest neighbor comparison method,which includes the dispersion degree of excellent results,the dispersion degree of corresponding parameters of excellent results and the average value of excellent results,is adopted.Finally,the moving average model is selected as the most suitable fitting model.? Find and analyze the scenario factors that lead to the decline of yield.In order to solve the problem of the decline of yield curve,the paper counts the number of yield withdrawals with different ranges,calculates the standard deviation of time series,observes the relationship between the decline of yield and the scenario according to the position relationship between the withdrawal point of return and the standard deviation curve of time series,and finds that the price spikes often appear in the index series near the withdrawal point of return.Peak and narrow-amplitude shocks are two scenarios.In this paper,the peak scenario is the focus of research.Fitting time series with fast moving average model and exponential smoothing model,defining peak inflection point and effective peak inflection point,designing peak search algorithm and model evaluation parameters.By comparing the number of peaks and efficiency under different thresholds,an effective parameter selection strategy and an optional parameter scheme are finally formed.On the choice of these two models,the moving average model can achieve better results at the lower peak,while the exponential smoothing model at the upper peak is slightly better.? Establish decision tree model and support vector machine model to predict the future direction of the peak inflection point.This paper chooses the fast moving average model to confirm the peak inflection point.The other parameters are slightly wider than the reasonable range determined above.The characteristic vectors of the peak inflection point are designed and calculated.The training and prediction schemes are constructed according to the year.The model is trained and predicted,and the effective peak inflection point predicted by the model is fitted with the fitting curve.The inflection points are superimposed to forecast the futures market together and compare with the annual rate of return obtained by fitting the inflection points.By comparison,the effect of pruning decision tree is better than that of common decision tree model,and the prediction effect of support vector machine model is better than that of pruning decision tree model.Then the pruning decision tree model and the support vector machine model are superimposed,and the effect is basically the same as that of the support vector machine model.The results of this study show that two innovative ideas,scene superposition and multi-dimensional nearest neighbor comparison,can establish an effective combination model prediction scheme.Although this paper only uses Shanghai Copper Index to study,it also has applicability to other futures varieties prediction research.