Intergenerational elasticities are
useful summary measures, but they may
conceal interesting details about intergenerational mobility at different points of
the distribution. Researchers have used different techniques to relax the linearity
assumption, including spline, higher-order terms or quantile regressions.
Unfortunately, the sample size at our disposal prevents us from applying these
techniques, and we rely on more traditional and simpler transition matrices,
dividing ancestors’ and descendants’ economic outcomes into three classes,
according to terciles (lower, middle and upper classes). In Table 5, we report the
transition matrix referred to earnings. For those originating from the lower class,
there are fairly similar opportunities to belong to one of the three destination
classes. For those coming from the upper class, in contrast, the probability of
falling down to the bottom of the economic ladder is relatively low. A similar “glass
floor” is observed for the wealth transition matrix (Table 6); moreover, in this case,
we also observe a “sticky floor”: more than two fifths of descendants from the
lower class remain there after centuries.
The results in Tables 3-6 suggest that the persistence of socioeconomic status
in the long run is much higher than previously thought. They are even more
striking given the huge political, demographic and economic upheavals that have
occurred in the city across the centuries. On the political ground, Florence has
passed from the capital city of a small city-state (see Figure 2) to a city within a
larger State (with the Italian unification in 1861), whose capital city is located
elsewhere. Regarding demography, the population was fairly stable between 1400
and 1800 and experienced a huge increase in the 19
th
and 20
th
centuries (see
Figure 3a). Finally, the GDP per capita was basically flat in the pre-industrial era,
while it recorded an exceptionally high growth rate during the 20
th
century (see
Figure 3b), accompanied by the industrial revolution, the tertiarization, and finally,
the technological revolution.
One might wonder whether these results can be generalized to other
societies. According to the evidence at our disposable, we argue that they can be
thoughtfully extended to other advanced countries, presumably in Western
Europe, that share a similar long run development pattern with Florence. Indeed,
Milanovic et al. (2011) showed that the gross domestic income per capita and the
Gini index in Florence in 1427 were comparable to those of other pre-industrial
societies for which we have data, such as England, Wales and Holland. Looking at
more recent evidence, according to the Eurostat data, in 2013, the purchasing
power standard GDP per inhabitant in Tuscany was just slightly above the EU28
average. Moreover, Güell et al. (2015a) provided evidence on the degree of
intergenerational mobility for all Italian provinces (the data are referred to 2005);
according to their evidence, the (simulated) intergenerational income elasticity for
15
the province of Florence would be between 0.4 and 0.5, a figure that is slightly
lower than that of Italy as a whole and broadly comparable with that of other
advanced countries, such as the United States, the United Kingdom and France
(Corak, 2013).
9
In sum, Florence does not seem to be a polar case in terms of
economic development and (static and dynamic) inequality.
5. Robustness
5.1 Imputation procedures and outliers
Table 7 provides a first set of robustness checks. First, we address the
imputation procedures. As far as earnings are concerned, the tax records, as is well
known, may suffer from a severe underestimation due to tax evasion. In the first
columns, we upwardly revise the variables from the tax records with the
correction factors suggested by Marino and Zizza (2011).
10
The results are
unchanged, and this may be explained by the fact that tax evasion is not correlated
with the pseudo-ancestors’ earnings. As far as wealth is concerned, this variable is
not directly observed in the 2011 tax records and has been obtained through an
imputation process based on the real estate income. In the third column, we
directly regress the real estate income on the ancestors’ wealth in order to avoid
our results from being driven by our imputation procedure. The results are
basically unchanged.
Second, we address the sensitivity to outliers, as the distributions of earnings
and wealth have long tails that might drive the results. In the second and fourth
columns, we trim both the dependent variable and the key regressor at the 1% and
the 99% levels, and we re-estimated equation (2): again, the estimates of positive
and significant intergenerational elasticities are fully confirmed.
5.2 Robustness of pseudo-links
Our empirical strategy relies on the assumption that the probability that one
9
Güell et al. (2015a) do not estimate intergenerational elasticity, but the Informative Content of
Surname (ICS) indicator that is monotonically related to the intergenerational elasticity. The
reported figures have been obtained by mapping the ICS values into elasticities using Figure 2 in
Güell et al. (2015b). See also Section 6.1 below.
10
Marino and Zizza (2011) compared incomes from tax records with those collected through the
Survey of Household Income and Wealth. This approach is based on the hypothesis that, as the
survey questionnaire is multipurpose and replying is not compulsory, it is likely that respondents
do not feel threatened or suspicious and would hence reply truthfully. On this basis, they provided,
for each income type, a proxy of tax evasion (as measured by the difference between the income
from the survey and the income from the fiscal source).
16