Forecasting Stock Prices of Taiwan Semiconductor Manufacturing Company (TSMC) Using Recurrent Neural Networks: Evaluating Predictive Performance in a Volatile Market

Abstract

Accurate stock price prediction plays a critical role in guiding investment strategies, particularly in dynamic industries such as semiconductors, where price volatility is high. This study investigates the effectiveness of Recurrent Neural Networks (RNN) in predicting the stock prices of Taiwan Semiconductor Manufacturing Company (TSMC), a global leader in the semiconductor sector. Using daily closing price data from January 2020 to January 2023, the RNN model was developed and trained to forecast future stock prices. The data was preprocessed with feature scaling to ensure the stability of model training, and a sliding window approach was applied to capture temporal dependencies. The model's predictive performance was evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) as key metrics. The RNN achieved an RMSE of 9.87 and a MAPE of 5.90%, indicating that the model provides reasonable accuracy in forecasting stock prices with a moderate level of deviation from actual values. Visual analysis further demonstrated the model's capacity to capture general trends in the stock price movements, although challenges were noted in predicting highly volatile periods. The study highlights the potential of RNN in financial forecasting while suggesting future improvements, such as incorporating advanced models like Long Short-Term Memory (LSTM) or external factors to enhance predictions during market volatility. These findings offer valuable insights for investors and analysts seeking to leverage machine learning in stock price prediction, particularly in industries characterized by rapid technological advancements and price fluctuations.