Neural Networks for Financial Time Series Forecasting

Area of Science:

• Quantitative Finance
• Computational Economics
• Machine Learning

Background:

• Financial and economic time series forecasting is complex due to sensitivity to external factors.
• Investors seek robust models to maximize profits and minimize losses in financial markets.
• Artificial Neural Networks (ANNs), particularly Recurrent Neural Networks (RNNs), show promise for improving predictive accuracy.

Purpose of the Study:

• To forecast the closing prices of eight stock market indexes.
• To forecast the closing prices of six USD-related currency exchange rates.
• To evaluate the performance of RNNs, Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) models.

Main Methods:

• Utilized Recurrent Neural Networks (RNNs) and its variants, LSTM and GRU.
• Applied models to forecast stock market index closing prices.
• Applied models to forecast currency exchange rate closing prices.

Main Results:

• The Gated Recurrent Unit (GRU) model demonstrated superior performance overall.
• GRU achieved the best results in univariate out-of-sample forecasting for currency exchange rates.
• GRU excelled in multivariate out-of-sample forecasting for stock market indexes.

Conclusions:

• GRU models provide enhanced predictive accuracy for financial time series.
• GRU is a highly effective tool for forecasting stock market indexes and currency exchange rates.
• The study highlights the potential of advanced RNN architectures in quantitative finance.