<![CDATA[Accurate stock price forecasting requires appropriate preprocessing, particularly for time series data with high variability and nonlinear patterns. This study investigates whether applying the Box-Cox Transformation (BCT) improves forecasting performance when modeling Telkom Indonesia's stock price using Neural Network Autoregressive (NNAR) and Double Exponential Smoothing Holt (DESH) methods. The NNAR model architecture is selected based on nonlinearity testing of lag variables, while DESH parameters are optimized by minimizing mean square error. Forecasting accuracy is evaluated using Mean Absolute Percentage Error (MAPE), root Mean Square Error (RMSE), and Mean Percentage Error (MPE), comparing models built with and without BCT. Results show that BCT does not enhance forecasting accuracy for either NNAR or DESH. Moreover, the NNAR model without BCT outperforms DESH, producing approximately 50% lower MAPE, RMSE, and MPE values on the testing dataset. These findings suggest that BCT may not be necessary for time series modeling in this case, and NNAR without transformation is recommended for forecasting Telkom's stock price.]]>
Copyrights © 2024
