Heteroscedasticity effects as component to future stock market predictions using RNN-based models.

Area of Science:

• Quantitative Finance
• Machine Learning in Finance
• Econometrics

Background:

• Stock volatility forecasting is crucial for investors and traders, providing business insights.
• Predicting stock volatility is a challenging yet vital task in financial markets.
• Existing models may not fully capture heteroscedasticity, a key element in stock price movements.

Purpose of the Study:

• To propose a hybrid model, GARCH-LSTM, for enhanced stock volatility forecasting.
• To leverage heteroscedasticity in stock prices for improved prediction accuracy.
• To evaluate the performance of the GARCH-LSTM model against traditional methods.

Main Methods:

• A hybrid approach combining Generalized Autoregressive Conditional Heteroskedasticity (GARCH) and Long Short-Term Memory (LSTM) recurrent neural network (RNN).
• GARCH model estimation followed by an ARCH effect test to identify remaining heteroscedasticity in residuals.
• Modeling the heteroscedastic dataset using the LSTM algorithm.

Main Results:

• The GARCH-LSTM model demonstrated superior performance with lower MAE, RMSE, MAPE, and HMAE values compared to a single LSTM.
• Forecasting accuracy improved by 15% and 13% with the hybrid GARCH-LSTM model over single LSTMs.
• The GARCH-LSTM model effectively utilized heteroscedasticity, outperforming GARCH models alone.

Conclusions:

• Hybrid GARCH-LSTM models are effective tools for predicting stock price movements.
• The proposed model assists investors by closely predicting and imitating observed stock price patterns and trends.
• Incorporating heteroscedasticity significantly enhances stock volatility forecasting accuracy.