Studies On The Performance Of Stock Market Predictive Models

With the development of China’s financial industry,the development of China’s stock market has become more rapid and attracted more investors to be involved.Not only investors,academic researchers are also looking for various effective forecasting models to forecast.Since its development,in the field of the stock market,there have been a large number of research literatures that use prediction models such as technical index analysis and time series models to predict the stock market.However,the effectiveness of these stock market forecasting models is often affected by the data snooping bias.Due to the data snooping bias,whether the forecasting ability of the stock market forecasting model is true or effective needs to be studied and tested.Different from the existing literature on the research of stock market forecasting models,this paper in the perspectives of data snooping bias,used the Stepwise Superior Predictive Ability Test to test the effectiveness of a total of 18,532 stock market forecasting models constructed by technical indicators and time series,in order to explore whether these forecasting models have real and effective forecasting capabilities.And also through the Step-SPA test to select the forecasting models that passed the test,hoping to provide favorable information to scholars and investors.In this paper,the daily returns data of the SH300 Index and the SZ50 Index are selected for empirical research.The research finds that the stock prediction model is indeed affected by the data snooping bias.This paper also empirically examined the forecasting capabilities of the stock forecasting model before and after the 2008 international financial crisis.The study found that under different historical stages and environmental impacts of different stock data,the effect of data snooping bias on the stock forecasting model was also differently affected.Also finds that the GARCH(1,1)and ARIMA(1,1,1)models can be selected by the Step-SPA test,both in the entire sample stage and in the empirical stage.Finally,this paper used GARCH(1,1)model as a benchmark model for testing,and finds that the prediction capabilities of other stock forecasting models have not significantly exceeded this benchmark model.