Calibrado de filtros no paramétricos y de Hodrick-Prescott para aproximar la tendencia-ciclo del EMAE[Calibration of Nonparametric Filters and Hodrick-Prescott for Approximating the Trend-Cycle of the EMAE]

The report proposes four alternatives to the standard X-13ARIMA-SEATS procedure (which uses the Henderson filter) for extracting the trend-cycle from the INDEC’s Monthly Estimator of Economic Activity (EMAE). The smoothing parameters of these alternatives were calibrated with the aim of emulating the official EMAE trend-cycle in two periods: 2004--2019 (excluding the impact of the pandemic) and 2004--2025 (including the pandemic). The results suggest using the Nadaraya-Watson (NW) kernel with bandwidths between 8 and 10, and Local Linear Regression (LLR) with bandwidths between 10 and 12. Both alternatives clearly outperform the LOESS method and deliver highly competitive results -- albeit slightly inferior -- compared to the Hodrick-Prescott filter (with $\lambda$ between 100 and 500). Additionally, it was observed that the nonparametric filters (NW and LLR) provide more stable and realistic adjustments in the presence of extreme shocks, such as the COVID-19 pandemic. This is because they are not affected by the outlier exclusion procedure performed by X-13ARIMA-SEATS on the seasonally adjusted series. At the ends of the series, Local Linear Regression (LLR) shows better behavior than the NW kernel, which suffers from truncation bias. In conclusion, it is recommended to extract the trend-cycle using the NW method when working with volatile series or those with frequent shocks, and to use LLR for more stable series or when greater accuracy at the final end of the series is prioritized.