Temporal Aggregation and Smoothing Parameter Sensitivity in Exponential Smoothing: Evidence from the Jakarta Composite Index

Abstract

The study examines the structural sensitivity and predictive performance of the Single Exponential Smoothing (SES) model when applied to varying temporal aggregation levels of the Indonesian stock market index. This study employs time-series data of the Jakarta Composite Index (JCI / IHSG) spanning from January 2, 2020, to May 31, 2025, which are structurally categorized into three distinct frequency domains: daily observations, weekly aggregated averages, and monthly aggregated averages. Methodologically, the optimal smoothing parameter (?) for each aggregation tier is determined through maximum likelihood estimation by minimizing the Mean Absolute Percentage Error (MAPE). The results reveal an extreme structural behavior where the optimal ? approaches its upper asymptotic boundary across all temporal frameworks, specifically 0.9943 for daily data, 0.9999 for weekly data, and 0.9999 for monthly data. Concurrently, the predictive error amplifies systematically as the aggregation window widens, yielding MAPE values of 0.74%, 1.30%, and 2.88% for daily, weekly, and monthly frameworks, respectively. The findings of this study suggest that the JCI movement exhibits strong adherence to the random walk hypothesis, wherein the mathematical framework of SES reacts almost exclusively to the most recent historical innovation. Consequently, temporal aggregation does not induce a structural smoothing effect on parameter convergence but rather introduces informational attenuation that compromises short-term forecasting precision.