Agista Marshanda

Statistika, Universitas Terbuka, Tangerang Selatan, Indonesia

Harmi Sugiarti

Statistika, Universitas Terbuka, Tangerang Selatan, Indonesia

DOI: https://doi.org/10.19184/mims.v26i1.60044

ABSTRACT 

Forecasting inflation data is an important part of economic decision making. Periodic updates are needed considering changes in external factors that affect the inflation rate. This study aims to examine the performance of the Holt-Winters Exponential Smoothing method in predicting the inflation rate in Indonesia for the period January 2018 to March 2024. The results show that the Holt–Winters Exponential Smoothing method performs reasonably well in forecasting inflation in Indonesia. The additive model demonstrates superior performance compared to the multiplicative model, as indicated by a Mean Absolute Percentage Error (MAPE) value of 7.60%, which is classified as highly accurate. This indicates that the additive model provides more accurate and consistent forecasts while effectively capturing the seasonal characteristics of the data. Furthermore, lower MAPE and Root Mean Square Error (RMSE) values in the testing data compared to the training data suggest that the model has good generalization ability and does not suffer from overfitting. However, since MAPE and RMSE do not provide consistent recommendations in selecting the most appropriate model between the additive and multiplicative approaches, future research is recommended to incorporate additional performance metrics such as Mean Absolute Error (MAE), Symmetric Mean Absolute Percentage Error (SMAPE), and Theil’s U statistic to achieve a more robust and comprehensive evaluation. In addition, model improvement can be pursued through the adoption of machine learning approaches, such as Long Short-Term Memory (LSTM), which are more capable of capturing nonlinear patterns and long-term dependencies in time series data.

Keywords: Inflation, Holt-Winters, forecasting.

MSC2020: 62M10