In the trend-cycle extraction, estimates are usually derived from moving average (also called linear filters) methods. In the center of the series, symmetric filters are applied. However, due to the lack of future observations, real-time estimates must rely on asymmetric moving averages. Classic asymmetric filters minimize revisions errors but introduce delays in the detecting turning points.

This presentation will describe different approaches to build asymmetric filters: methods based on local polynomials, optimization of filters’ properties (Fidelity-Smoothness-Timeliness) and Reproducing Kernel Hilbert Space. It will show how to use them with the R package rjdfilters, applying them to the COVID-19 crisis