Abdulrashid Jamnia; Mohammad Reza Sasouli; Emambakhsh Heidouzahi; Mohsen Dahmarde Ghaleno
Abstract
The capital or stock market along with the money market is one of the most important parts of financial sector of the nation’s economy, providing long-term financing required for efficient production and service activities. The total stock price index as reflector of stock market fluctuation is ...
Read More
The capital or stock market along with the money market is one of the most important parts of financial sector of the nation’s economy, providing long-term financing required for efficient production and service activities. The total stock price index as reflector of stock market fluctuation is important for finance practitioners and policy-makers. Therefore, in this research, a comparative investigation was presented on two superior deep-learning-based models, including long short-term memory (LSTM), and convolutional neural network long short-term memory (CNN)-LSTM, applied for analysing prediction of the total stock price index of Tehran stock exchange (TSE) market. The complete dataset utilized in the current analysis covered the period from September 23, 2011 to June 22, 2021 with a total of 3,739 trading days in the TSE market. Forecasting accuracy and performance of the two proposed models were appraised using root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) criteria. Based on the results, the CNN-LSTM showed the lowest values of the aforementioned metrics compared to the LSTM model, and it was found that the CNN-LSTM model could be effective in providing the best prediction performance of the total stock price index on the TSE market. Eventually, graphically and numerically, various prediction results obtained from the proposed models were analysed for more comprehensive analysis.
Saeid Tajdini; Farzad Jafari; Majid Lotfi Ghahroud
Abstract
According to the literature on risk, bad news induces higher volatility than good news. Although parametric procedures used for conditional variance modeling are associated with model risk, this may affect the volatility and conditional value at risk estimation process either due to estimation or misspecification ...
Read More
According to the literature on risk, bad news induces higher volatility than good news. Although parametric procedures used for conditional variance modeling are associated with model risk, this may affect the volatility and conditional value at risk estimation process either due to estimation or misspecification risks. For inferring non-linear financial time series, various parametric and non-parametric models are generally used. Since the leverage effect refers to the generally negative correlation between an asset return and its volatility, models such as GJRGARCH and EGARCH have been designed to model leverage effects. However, in some cases, like the Tehran Stock Exchange, the results are different in comparison with some famous stock exchanges such as the S&P500 index of the New York Stock Exchange and the DAX30 index of the Frankfurt Stock Exchange. The purpose of this study is to show this difference and introduce and model the "reversed leverage effect bias" in the indices and stocks in the Tehran Stock Exchange.
