Abstract:
Accurately forecasting stock prices movements can lead to financial gains, making it a highly sought-after area of study. In recent studies, Temporal Convolutional Network (TCN) has risen in popularity due to its use of dilated convolutions, which are adept at capturing temporal dependencies within time series data. DeepTCN, a variation of TCN designed specifically for probabilistic forecasting, is said to outperform other models in time series forecasting. As far as we know, no extensive research has been conducted to evaluate the performance of DeepTCN compared to TCN. This study conducted a comparative analysis to assess the performance of both TCN and DeepTCN in Indonesian stock price prediction. Both models will be evaluated using Mean Squared Error (MSE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) scores. The result from this comparative analysis shows that DeepTCN is superior to TCN in predicting stock prices. DeepTCN consistently outperforms TCN, with lower values of MSE, RMSE, and MAPE. This improved performance lies in the parametric approach used in DeepTCN, which allows it to better capture and adapt to fluctuations in stock trends. The findings from this comparative analysis emphasize the need to assess forecast objectives and dataset requirements when choosing between TCN and DeepTCN.