Page 321 - Special Topic Session (STS) - Volume 1

STS441 Andrea N.
4. The current approach and the road ahead
Banca d'Italia is currently working in collaboration with the EG-LMM group
and with the University of Perugia (professor Giovanna Ranalli) to develop a
new adjustment method that builds on the previous experiences.
The methodology focuses on the two major reasons for the macro-micro
gap (once differences in definitions are addressed): the low probability of rich
households to be captured by the survey and the underreporting behavior.
The baseline method works under the assumption that the only external
information available are the National accounts totals and a list of the richest
people in the country such as the Forbes World Billionaires list. Yet, it can be
easily extended to incorporate external information (when available).
The preliminary step consists in harmonizing as much as possible
definitions and concepts across the two data sources. For instance, the total
wealth held by non-profit institutions serving households is estimated and
then removed from FAs. Then we apply a set of sequential adjustments some
of which are iteratively repeated until a convergence criterion is met. The first
step is to split the survey sample in two groups: the “rich” households and the
“non-rich” households. The two groups are then adjusted using a different
methodology. For rich households we use the Pareto method (Vermeulen,
2016). The Pareto is a highly right-skewed distribution with a heavy tail which
has already been shown to fit the upper tail of the distribution. It requires a
preliminary estimation of the share of rich households and the choice of a
wealth threshold (above which households are classified as rich). As a result of
the method it is possible to estimate the total wealth held by rich households.
We then subtract it from the total household wealth and distribute the
remaining share among “non-rich” households using imputation methods
(proportional adjustment can be seen as a particular case). We repeat the
process until convergence (the difference between the share held by the Rich
does not change significantly over iterations).
The iterative procedure makes sure that the adjustment of the two groups
is done simultaneously and that the final results do not depend on the initial
choices relating the threshold and the parameter of the Pareto distribution.
We call this approach “Simultaneous Pareto-calibration allocation”.
Table 1 shows some preliminary results relating four countries. The
method is compared with a simple proportional allocation which consists in
estimating for each wealth component the ratio between macro and micro
estimate and then in multiplying the amount declared in the survey by
respondents to this ratio. The proportional method is quite used as a
benchmark since it is very easy to apply and preserves the univariate
distributions. The cons are that it assumes that the underreporting behavior is
equal for all households and that it does not adjust either for missing wealthy
at the top or for noreporting.

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