Page 283 - Special Topic Session (STS) - Volume 3

STS544 Baoline C. et al.
rates of the indicators were selected. Based on the dynamic characteristics of
PCE services component and the dynamics of its indicators (or common
factors), different combinations of lagged dependent variables and current
and lagged independent variables were selected.
4.2 Results from out-of-sample prediction
Because our samples have short time spans, to efficiently use the
information from the data, we chose the recursive estimation approach to
compute pseudo-out-of-sample predictions. This means that each one-step-
ahead prediction is computed from estimated model using data up to the
quarter prior to the quarter being predicted. We measure improvement in
accuracy by reductions in the RMSR relative to the RMSR of the current
extrapolation method.
Of the 121 PCE service components, 36 component series use identical
indicator data for the first and the third estimates. For these components,
revisions are zero or close to zero with any minor revisions coming from
revisions in the indicator data. The remaining 85 components use distinct
source data to compile the first and the third estimates. Our evaluation of
improvement in accuracy is based on the changes in the RMSR of these 85
components.
The main observations from the one-step-ahead pseudo-out-of-sample
predictions are that 1) out-of-sample predictions from both bridge equation
and bridging with factors models resulted in reductions in RMSR for 63 PCE
service components that used distinct indicators for the first and the third
estimates; 2) out-of-sample predictions from estimated bridge equations
outperformed those from bridging with factors model in 48 out of 63
components (76%); 3) out-of-sample predictions from both bridge equation
and bridging with factors models led to reductions in the RMSR at the sub-
group aggregates of the PCE services, and 4) degrees of reduction in the RMSR
varied from the one-step-ahead predictions computed from different
estimated models and differed across service groups.

272 | I S I W S C 2 0 1 9