چكيده به لاتين
Stock price forecasting has always attracted the attention of many researchers as a challenging issue. With the progress of science and technology, new methods were always used to compete with the old methods and achieve more accurate predictions. In the last few years, Tehran Stock Exchange shares have had illogical and very complicated movement trends, which are more difficult to predict than has passed selection of variable and selection of prediction method are two very effective issues in increasing the accuracy of predictions. In this research, by constructing sixteen new and innovative variables and using a combination of deep learning methods and advanced regressions, we examine three important hypotheses related to the prediction of Tehran Stock Exchange stocks. We introduce the best variable selection method and the best forecasting method for one day and thirty days by designing thirty-nine combined one-day price forecasting models and fourteen thirty-day price forecasting combined models and comparing them. The results show that our innovative variables had an effective role in increasing the accuracy of predictions and the designed combined methods of advanced lasso and ridge regressions have more predictive accuracy than the combined methods based on neural network and lasso regression. In this research, systematic risk, ratio analysis, and multi-time frame analysis have been used in price forecasting, and it has led to a significant increase in forecasting accuracy. The best models of this research, i.e. the combined models based on Lasso variable selection and the ridge forecasting method, were able to describe 96% of the variability of the response variable in one-day forecasts and 63% in thirty-day forecasts for the test data.
