Pronóstico jerárquico del IPC del Ecuador con ponderadores oficiales del INEC: comparación de modelos ETS y ARIMA mediante agregación bottom-up

Abstract

This study develops a hierarchical forecasting approach for Ecuador’s consumer price index (CPI). It uses 108 monthly series disaggregated by city and expenditure division, together with the official weights published by INEC, which define the weighted-average structure of the national CPI. Unlike the traditional hierarchical forecasting literature, which focuses on additive hierarchies where upper levels are obtained as sums of lower--level components, this work adopts a strategy compatible with the weighted nature of the CPI. ETS and ARIMA models are automatically estimated on each disaggregated series, and forecasts are then coherently aggregated through a bottom-up scheme to obtain the total CPI. Performance is compared against direct ETS and ARIMA models estimated on the aggregate CPI, evaluating both a single origin and 12-month rolling windows. The bottom-up approach with ETS achieves the lowest errors, with a MAPE of 0.70 % versus 1.67 % for the best direct model. The improvement, however, is not uniform across expenditure divisions or forecast horizons. Overall, the results suggest that, for this case study, the disaggregated approach outperforms direct aggregate models and allows generating consistent projections for lower levels of the hierarchy.