| How is
EuroCOIN constructed?
Two Dimensions to the Data:
First,
start with the data. GDP – and the EuroCOIN indicator –
are time series, sets of observations over time. But the
underlying data set has another dimension – cross section
data, observations across countries and sectors. This
additional cross section dimension turns out to be very
important in building the indicator: the use of innovative
econometric techniques allows the cross section and time
series data to be combined. Surprisingly, using the additional
information from the cross section dimension solves two
important time series problems: the unavailability of some
recent data because of publication lags; and the difficulty
filtering out short run noise from the data while at the same
time using this data to produce forecasts.
Levels versus Growth Rates –
DeTrending the Data:
The
first step in building the indicator is important, but its
importance is easily overlooked. Statistical and econometric
theory is only applicable to data which are
"stationary". Most economic time series are trending
and so are non-stationary. So the first step is to remove the
trend from each variable to ensure they are stationary. This
is important, because it means that the indicator will track
changes in GDP, not its level. Since GDP data are available on
a quarterly basis, the indicator is designed to track the
quarterly growth rate of GDP for the euro area.
Removing Measurement Error,
Local and Sectoral Shocks:
The
next step in building the indicator is to decompose each
variable into two uncorrelated components: a "common
component" and an "idiosyncratic component".
The idiosyncratic component is in effect a residual, which
captures the effect of shocks affecting only that variable.
The common component of a variable is "common" in
the sense that it depends on a small number of common shocks,
which affect all the variables. The same set of shocks drive
all the common components, but the values of the common
component at a point in time will differ across variables
because the common shocks affect each variable with different
weights and lags. Isolating the common components is a key
step in building the indicator: in fact, the coincident
indicator is formally defined as the common component of the
GDP growth rate, after this variable is filtered to eliminate
high frequency variation (i.e. at frequencies of less than
fourteen months).
Why does the indicator focus on
the common component and ignore the idiosyncratic component?
The idiosyncratic component captures both sector-specific
shocks, such as shocks affecting output in a particular
industrial sector, and locality-specific shocks, such as a
natural disaster, which may have large but geographically
concentrated effects. Eliminating the idiosyncratic component
will produce a signal that is more useful for policy makers.
Shocks originating from a local or sectoral sources generate
dynamics that should, of course, be monitored by local or
sectoral policy makers. Such shocks are, however unlikely to
explain a large fraction of the European GDP, because they
will tend to offset each other when aggregated. By contrast,
common, Europe-wide policy should monitor the dynamics
generated by common shocks. Hence, an index of the euro area
business cycle— the reference indicator for common European
policy— should not include the idiosyncratic component.
Another important reason for removing the idiosyncratic
component from GDP is that it is likely to include most of the
measurement errors, since these are likely to be uncorrelated
across sectors or countries.
Removing Seasonal and High Frequency Noise:
The
idiosyncratic component is not the only noise affecting the
variables that must be removed. GDP growth will be affected by
short-run movements
(including seasonal and very short-run, high-frequency
changes).
A cyclical indicator will be
more useful once such short-run changes are removed, in
order to reveal the underlying medium- and long-run tendency
of the economy. The common components can be decomposed into
the sum of waves of different periodicity (the so-called
"spectral decomposition"). So the next step is to
then decompose the common component in turn into two
subcomponents: a cyclical, medium- and long-run component and
a non-cyclical, short-run component. The first subcomponent
includes waves of more than 14 months duration, and the
second, waves of shorter duration. Our cyclical indicator is
the first subcomponent of the euro area GDP variable. Previous
research has relied on "two-sided filters" to
eliminate the high frequency noise. These
filters work well in the middle of the sample, but entail problems
at the end of the sample, since they require knowledge of the future
values of GDP, which of course we do not have. These
two-sided filters can be dispensed with, however, by using
instead the information in the cross-sectional dimension of
the data to eliminate the high frequency dynamics.
Using the information in the
cross section dimension has another important advantage: A
good indicator of the business cycle should be up-to-date.
Each month we aim to be able to produce an estimate of the
indicator for the previous month, so that we can describe what
is happening in the economy now, not three or four months ago.
For some series, however, data are not yet available for
the most recent three to five months and so we
need to predict their values in order to construct the
indicator. The predictions will have errors, and it is
important to minimize these errors. The role of the
information from the cross-section dimension (and particularly
from the leading variables) is vital in reducing the
prediction error. For more
details, read the technical description
of the indicator.
|
|
| What data was used to produce
EuroCOIN?
The database used to construct EuroCOIN is organized into eleven blocks:
- industrial production
- producer prices
- consumer prices
- monetary aggregates
- interest rates
- financial variables
- exchange rates
- surveys by the European Commission
- surveys by national institutes
- external trade
- labour markets
Each block contains time series for Germany, France, Italy, Spain, The Netherlands, Belgium and
(when available), for the euro area as a whole. The six
countries
account for more than 90% of the euro area GDP. The
database includes almost 1000 time series whose homogeneity
over time and across countries has been verified. The
dataset spans the period January 1987 to the most recent data
releases. Although many time series are available for a longer period, the decision to set the starting date in 1987 is the result of a trade-off between obtaining richer time series information and maintaining a large cross-sectional dimension for the dataset.
|
|