Comparative Study of Traditional and Modern Models in Time Series Forecasting for Inflation Prediction

Abstract

Time series forecasting plays a crucial role in economic analysis, particularly in anticipating inflation and policy planning. This study compares the performance of seven different time series forecasting models, namely ARIMA, SARIMA, ETS, Prophet, LSTM, XGBoost, and TCN, in predicting inflation rates. Each model was applied to four years of inflation data to test its accuracy and reliability. The evaluation was conducted using Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) to measure the performance of each model. The results indicate that deep learning models, particularly LSTM and TCN, achieved the highest accuracy with the lowest MSE and RMSE values, specifically 0.0008 and 0.0015 for LSTM, and 0.0007 and 0.0013 for TCN, indicating their capability in capturing complex temporal patterns. Traditional models such as ARIMA and SARIMA, while effective in capturing trends and seasonality, showed limitations in handling non-linear patterns and sudden changes, with MSE and RMSE values of 0.0012 and 0.0024 for ARIMA, and 0.0011 and 0.0023 for SARIMA, respectively. ETS, with the highest MSE and RMSE values of 0.0013 and 0.0025, demonstrated limitations in dealing with the complexity of inflation data. XGBoost also showed good performance with MSE and RMSE values of 0.0009 and 0.0018, combining flexibility and robustness in handling complex data. Prophet achieved an MSE of 0.0010 and RMSE of 0.0020, indicating that while it effectively captures seasonal trends, there is room for improvement in handling rapid inflation increases. This research provides in-depth insights into the strengths and weaknesses of each model, as well as recommendations for practical applications in inflation forecasting. By presenting a comprehensive comparative analysis, this study aims to assist researchers and practitioners in selecting the most suitable forecasting model for their specific needs