Evaluation of time series models for stock price prediction

Lim, Jing Hao (2023) Evaluation of time series models for stock price prediction. Master dissertation/thesis, UTAR.

Preview

PDF Download (1727Kb) | Preview

Abstract

This project aims to compare and analyse the performance of five time-series forecasting model—ARIMA, SARIMA, Prophet, Holt Winters, and LSTM—in predicting stock prices for the healthcare and technology sectors. The evaluation focuses on the Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) metrics across various data ranges, including 1 year, 3 years, 5 years, and 7 years. The findings indicate that the LSTM model consistently achieves the lowest MAE and RMSE values, suggesting superior forecasting accuracy compared to the other models. The SARIMA model ranks second in performance, followed by Prophet, ARIMA, and Holt Winters. These results offer valuable insights for researchers, practitioners, and investors seeking to forecast stock prices using time series model. By understanding the strengths and weaknesses of different models, stakeholders can make betterinformed decisions, improve overall market efficiency, and enhance risk management strategies. Future research can explore the effects of data pre-processing, feature engineering, and hyperparameter tuning on forecasting accuracy, as well as expand the analysis to other sectors to assess the generalizability of the findings.

Actions (login required)