Thursday 7 September 2017, 1.00PM to 2.00pm
Speaker(s): Hiroshi Yamada (Hiroshima U)
Abstract: This paper introduces a new trend filtering method, which has the Hodrick-Prescott (HP) filtering and the l1 trend filtering as special cases. The relation between the HP filtering, the l1 trend filtering, and the new filtering corresponds to the relation between the ridge regression, the lasso (least absolute shrinkage and selection operator) regression, and the bridge regression. For this reason, we refer to the new filtering as bridge filtering. The bridge filtering enables us to estimate the trend component of a time series with less-sudden structural changes. In this paper, after introducing the new filtering, we show some of its properties, a method for specifying its tuning parameter, and an empirical illustration of how it may be applied.
Location: A/EC202 Economics Staff Room
Admission: All welcome