Research On Iron Ore Futures Price Prediction And Quantitative Strategy Based On Machine Learning Methods

In recent years,with the development of computers,machine learning methods have been introduced into the field of quantitative investment,and machine learning algorithms have been applied to the price prediction of financial assets.Latent source model is a machine learning method based on Bayesian regression,which obtains the effective prediction price by obtaining the clustering characteristics of training data and applying bayesian regression.This method uses historical data instead of empirical data to effectively reduce the dimension of parameter estimation.HMM is a classical algorithm in machine learning methods.It is a special case of dynamic Bayesian network and has the characteristics of effectively reflecting the dynamic change characteristics of time series.Its internal recursive algorithm that can call dynamic Bayesian network can effectively avoid overfitting and reduce the amount of training.The main contents of this paper are as follows :(1)build the potential source model,use Kmeans clustering algorithm to cluster the training set to obtain the prototype vector containing the price trend characteristics,and then perform bayesian regression according to the prototype vector to get the predicted price changes and get the predicted value of the price sequence.(2)the HMM model was constructed,the entire section of the test sets as the model forecast results were input in the form of observation value generation and was improved,the improvement for the use of training set rolling division,each division range using the HMM model fitting,set the same length of the test interval to find the nearest historical prices and take the historical prices sequences,applies as a prediction or forecast.(3)Build the potential source-HMM model,and use the predicted values obtained by the rolling HMM fitting method as the input of the potential source model to obtain the new prototype vector and then perform Bayesian regression to obtain the final prediction results.(4)The above models are tested back respectively,and the random optimal control method is used to dynamically adjust positions to ensure the optimal performance index in the process of testing back,and the paired transaction in THE CTA strategy is used as the reference to the machine learning algorithm.The results show that the prediction result of quantitative strategy combined with latent source model and HMM model is better than that of the two models alone.