LASSO Methods And Their Applications To Stock Price Prediction

In this thesis,we mainly introduce the basic theory and the analytic methods of LASSO,then apply LASSO to variable selection and parameter estimation problems of time series modeling such that we can effectively determine the order of time series models,and avoid the subjectivity of identification of correlation diagrams.At first,we use LASSO and MLE methods to estimate parameters on simulation data of several AR(p)models.It is shown that using the LASSO methods to analyze time series data is feasible,and two methods have their own advantages and disadvantages,adaptive LASSO method is better than MLE method for lower-order AR(p)models,but for higher-order ones the MLE is better.In the analysis part,we establish a time series and multivariate time series models of Baoshan Iron & Steel stock data,and forecast the future value of the opening price,the predictions of LASSO model are better than those of ARIMA model.These results show that LASSO methods have a good performance in modeling complex stock price time series,and making accurate predictions.