Speakers
Description
Stock market price forecasting represents a challenge due to the inherent complexity, non-linearity, and volatility present in financial time series. Classical methods such as ARIMA have long been employed for forecasting financial data due to their ease of interpretation and solid theoretical foundations. However, such linear models face limitations in accurately capturing complex and nonlinear market dynamics. Forecasting stock market prices is often challenging because financial time series exhibit complex dynamics, nonlinear behaviour, and significant volatility. Traditional models, particularly ARIMA, are frequently applied in financial forecasting given their theoretical foundations and interpretability. Nevertheless, linear methods like ARIMA often face challenges to adequately capture the nonlinear structures that are present in stock market data. To address these limitations, this study proposes an LSTM-based approach, given its capability to effectively model temporal dependencies and complex patterns in sequential data. Because of that argument we hypothesize that the LSTM model will yield superior forecasting accuracy compared to the ARIMA model due to its enhanced ability to handle non-linear and complex relationships within the data. The methodology involved collecting historical closing price data from the stock market, with a particular attention to the European defence industry. The dataset was preprocessed and partitioned into training and testing subsets. The ARIMA model was developed using standard identification, estimation, and diagnostic checking procedures to identify an optimal set of parameters. For the LSTM network, a supervised learning strategy was applied where the data was structured into sequences suitable for training recurrent neural networks. Various architectures and hyperparameters, including the number of hidden units, layers, and training epochs, were tested and optimized. Both the ARIMA and LSTM models were implemented using Python, taking advantage of its extensive analytical libraries suited for financial time series forecasting. Preliminary results indicate that the LSTM approach outperforms ARIMA across the selected evaluation metrics, which include Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). The performance improvement observed with the LSTM model confirms its advantage in effectively capturing nonlinear dynamics within stock price movements, which traditional linear models like ARIMA may inadequately represent. These preliminary findings showed the practical relevance and potential applicability of deep learning methods, specifically LSTM networks, in increasing the accuracy of financial time-series predictions.
