Artificial Neural Network Based Non-linear Transformation of High-Frequency Returns for Volatility Forecasting

Area of Science:

• Quantitative Finance
• Machine Learning
• Financial Forecasting

Background:

• Accurate volatility forecasting is crucial for financial risk management and trading strategies.
• Traditional models often struggle to capture complex patterns in high-frequency financial data.

Purpose of the Study:

• To investigate the efficacy of Long Short-Term Memory Recurrent Neural Networks (LSTM RNNs) in extracting information from intraday returns for daily volatility forecasting.
• To compare LSTM RNN-based forecasting performance against models that do not use extracted information and established alternative methods.

Main Methods:

• Utilized Long Short-Term Memory Recurrent Neural Networks (LSTM RNNs) to process intraday high-frequency stock returns.
• Applied the developed models to forecast the daily volatility of IBM stock.
• Compared forecasting accuracy with baseline models and Mixed Data Sampling (MIDAS) alternatives.

Main Results:

• Models incorporating information extracted by LSTM RNNs demonstrated significant improvements in daily volatility forecasting accuracy for IBM stock.
• The LSTM RNN approach outperformed models that omitted the extracted information.
• LSTM RNNs proved superior to two Mixed Data Sampling (MIDAS) alternatives in this forecasting task.

Conclusions:

• LSTM RNNs are effective tools for extracting predictive information from high-frequency trading data.
• The proposed LSTM RNN methodology offers a superior approach to daily volatility forecasting compared to existing popular methods.
• This research highlights the potential of advanced deep learning techniques in financial market analysis.