Skip to contents

GSS target factor (Swanson extended)

Source: R/data.R
gss_target.Rd

The Federal Funds Rate factor from Swanson's (2021) three-factor decomposition of high-frequency FOMC surprises, the direct extension of the original "target" factor in Gurkaynak, Sack, and Swanson (2005). Event-level factor values are summed within calendar months; months with no scheduled FOMC meeting are coded as zero.

Usage

gss_target

Format

A data frame with columns:

date

Date. First day of the observation month.

shock

numeric. GSS target factor, percentage points.

series

character. Series identifier "gss_target".

Source

Swanson, E. T. (2021). "Measuring the Effects of Federal Reserve Forward Guidance and Asset Purchases on Financial Markets." Journal of Monetary Economics 118: 32-53. doi:10.1016/j.jmoneco.2020.09.003 . Data: https://sites.socsci.uci.edu/~swanson2/. Original two-factor decomposition: Gurkaynak, R. S., Sack, B., and Swanson, E. T. (2005), International Journal of Central Banking 1(1): 55-93.

Details

Identification. Swanson (2021) computes the first three principal components of high-frequency futures surprises, then rotates them by (i) zero loading of factor 3 on the current-month fed-funds-rate surprise, (ii) minimum sum of squared factor-3 values over the pre-ZLB sample 1991-07 to 2008-12, and (iii) sign normalisation. The target and path factors are therefore conditional on the pre-ZLB window used to pin down factor 3. Extensions past the bundled span must re-estimate the rotation, not simply append new events.

Relation to GSS 2005. Pre-2009 the target factor closely tracks the original two-factor decomposition of Gurkaynak, Sack, and Swanson (2005). Post-2009 it differs because unconventional policy announcements are absorbed by a distinct LSAP factor (not bundled in v0.1.0; see Swanson's website for the full three-factor panel).

Monthly aggregation. Event-level factors are summed within calendar months. Months with no scheduled FOMC meeting are coded 0. Users who want to distinguish "no news" from "news = 0" should recode no-meeting months as NA before estimation (Bu, Rogers, and Wu 2021, Journal of Monetary Economics 118).