Designing a Forecasting Model for Stock Market under Non- Normality: Case Study of Chinese Stock Exchange

Abstract

In recent decades, the stock market and its growth have attracted investors. One of the biggest challenges investors have always faced is the high volatility of stock prices. There are several studies on the prediction of stock prices, but most of them are in consistent with the fact that data must be fitted in the normal distribution. In other words, conventional prediction of stock prices beyond the normal distribution is limited, leaving a gap due to the non-normality of data. For this reason, this research has focused on data behavior and its effect on the forecasting accuracy. Therefore, we reviewed 72 recently published papers. This review identified three forecasting models and two bootstrapping methods which have been combined into a new alternative model to develop and present an uncertainty model with deeper insight into the data. In this line, we developed a five-stage model to analyze and select the best combination for prediction. This model was coded with Python and 10 selected stocks from Chinese stock exchanges were used as input to ensure the robustness of the model. The model has 20 outputs per share. Holt's winters' models accurately reproduced the trend. And "BS1-Holt's additive damped trend", "BS1- HW" were the most accurate models. In conclusion, investors could benefit from this data-based uncertainty model to improve their forecasts and profits.

Keywords: Stock Price prediction, Bootstrap, Combination model, Data Science, Non-normality, Uncertainty model

DOI: 10.7176/EJBM/14-12-03

Publication date:June 30^th 2022