Forecasting The Direction Of Stock Return Based On A New Nonparametric Method

Forecasting the direction of stock return has an important guiding significance for investors,the movement of the stock is either rise or fall,can be regarded as a binary choice problem,most scholars use Logistic regression forecasting model to forecast the direction of the stock return,however,the logistic model belongs to the generalized linear model,which make it has some limitations.The financial market is a complex dynamical system,and has flexible distribution features,which makes the linear model is difficult to capture its characteristics,the Logistic regression prediction method is difficult to achieve a good performent.Moreover,there are few classification function in existing econometric models,so it is of great significance to study the new econometric classification method.the parametric models are hard to avoid model misspecification when predicting stock return,nonparametric models can better describe features of the stock return than parametric models.Harvey&Oryshchenko(2012)studied nonparametric kernel density estimation for the time series data and put forward the time-varying density function estimation theory.In this paper,the theory is applied to forecast the direction of stock return,named as the time-varying nonparametric density function model-TVF model.Considering some variables have influence on stock return,in order to improve the prediction of the model,this paper extends the model framwork by imposing affecting variables as weighting factors on kernel density and establishs a new nonparametric prediction model,named as the time-varying factor weighted density function model-F-TVF.In the empirical study,this paper apply these models to monthlyprice data of Chinese stock market index and employ the rolling window strategy for out-of-sample forecasting.In order to comprehensively evaluate the predicting power of above models,the article use statistical evaluation method and the method of simulating trading strategy,and meanwhile do a stability test,which makes the evaluation conclusion robust and reliable.The empirical results show the new nonparametric F-TVF model has the best prediction performance in the three models,meanwhile it has significant out of sample predictive ability of the direction of stock return in statistical and economic sense;F-TVF model and TVF model both have significantly advantage than Logistic model,which reflects the nonparametric models have advantages than parametric models when forecasting the direction of stock return;In addition,this paper justify the predictability of the financial market in China to some extent.