Stock Price Forecasting Models During Crisis and High Volatility

Abstract

Stock market crisis can emerge due to variations in the economic policy of a large economy. It has been observed that the crisis may originate from a large size economy, and the impact of the crisis will affect smaller economies as well. Crisis prediction is critical for the financial market, and this attracted many researchers and academicians. However the fair value of the stock price depends on stock financial parameters. There are many financial parameters such as price to earnings, company returns, company debt, etc. Identification of relevant financial parameters is a challenging task. Therefore in this work a Hybrid Feature Selection (HFS) technique is proposed to select essen- tial financial parameters. After selecting essential financial parameters, Naive Bayes (NB) classifier is used to classify high quality stocks. Then stock price bubbles are identified using Relative Strength Index (RSI). From these bubble points, with help of moving average we selected stock crisis points. These crisis points are fed to regres- sion techniques. The performance of the model is evaluated based on Mean Squared Error(MSE), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). It is found from the experimental results that performance of the XGBoost model is better than the Deep Neural Network (DNN) model. Stock price movements forecasting is an important topic for traders and stock an- alyst. Timely prediction of stock movements yields more profits. However, there are more than a hundred technical indicators are available, so it is essential to identify the correlation between stock price and technical indicators. Because each technical indi- cator signifies different aspects of stock price, so it is crucial to have an appropriate feature selection algorithm to select the correct technical indicators. Boruta feature se- lection is used to select the best features out of the 33 technical indicators and stock price movements are classified using the DNN, Artificial Neural Network (ANN), and Support Vector Machine ( SVM). It is found from the experimental results that perfor- mance of the DNN model is better than the ANN and SVM model. The proposed DNN method improved the classification accuracy rate by 5% to 6%. Volatility is a measure that represents the rate of change in the stock price over time, and it is calculated using standard deviations. This measure helps the investors to esti- i mates the risk in the stock investments. The stock price is volatile due to many factors such as demand, supply, economic policy, and company earnings. Investing in a volatile market is riskier for stock traders. Most of the existing work considered Generalized Auto-regressive Conditional Hetroscedacity (GARCH) models to capture volatility, but this model fails to capture when the volatility is very high. We have considered stock price volatility estimation using the regime-switching models. The performance of the Markov-Switching GARCH (MSGARCH) and Self-Exciting Threshold AutoRegres- sive (SETAR) models is calculated using RMSE and Mean Absolute Percentage Error (MAPE) metrics. It is found that regime-switching models performed better than the GARCH models.