LARS Algorithm On The Variable Selection And Its Application On The Economy

Variable selection is an important step in establishing an statistical linear model,Based on different needs, different constraints can be constructed to carry out variable selection. This paper will introduce the LARS-Lasso algorithm whose main purpose is to screen out some variable which have less impact on the de-pendent variable.This paper contains the contents of two aspects:the one is the knowledge about the Lasso,the other one is the Least Angel Regression("LARS") algorithm.The main idea of this paper is constructing the problem of variable se-lection through adding the absolute value constraints to the ordinary least square problem,then use the Lagrange Multiplier method,the classical Lagrange Multi-plier can be constructed as the parameters for the iterative algorithm LARS,after that,we can get the answer through modifying the iteration procedure of the LARS according to the constraints of the Lasso. On this basis,we establish a model for the national economy using the algorithm,and analyze the indicators,finally we get the variables which have more effect on the national economy and the estimation for the GDP,after that,we predicate the growth rate of the GDP in 2011. In order to study the accuracy which is caused by the size of the samples, we compare the small samples model with the big one, and get the result that the size of the samples is important to the accuracy of the algorithm.