placebo regressions where we randomly reassign
surnames to the descendants
and regressions exploiting rare and Florence-specific surnames – is largely
reassuring on the strength of the pseudo-links. If any, our estimated elasticity is
downward biased. Second, family survival rates – and, therefore, the likelihood of
finding descendants of Florentine families in the 15
th
century among current
taxpayers – may vary across families. If the variation in the survival rate was
correlated with current earnings and/or wealth, this would bias our estimates. To
address this issue, we simulate earnings and wealth realizations for missed
(unobserved) families, assuming that the economic outcomes of their descendants
are independent from those of their ancestors (i.e. setting the intergenerational
elasticity to the lower bound of zero) and rerun the baseline regressions. We also
adopt a more standard Heckman approach, accounting for selectivity biases due to
the survival rate. Both exercises qualitatively confirm our main findings.
To the best of our knowledge, we are the first to provide evidence on
intergenerational mobility over the very long run, linking ancestors and
descendants that are six centuries apart (i.e. about 20 generations of 30 years
each). This is the main element of novelty in the paper. Linking people through
more than one generation has rarely been done. In no other case has such a long
time span been studied. Chan and Boliver (2013) showed a statistically significant
association between grandparents’ and grandchildren’s class positions, even after
parents’ class position is taken into account. Lindahl et al. (2015) used Swedish
data that links individual earnings (and education) for three generations and found
that persistence is much stronger across three generations than predicted from
simple models for two generations. More similar to our paper, Collado et al. (2012)
and Clark and Cummins (2014) exploited the distribution of surnames to estimate
social mobility over the long run. Collado et al. (2012), using data from two
Spanish regions, found that socioeconomic status at the end of the 20th century
still depends on the socioeconomic status of one’s great-great grandparents;
however, they also suggested that the correlation vanishes after five generations.
Clark and Cummins (2014) used the distribution of rare surnames in England and
found significant correlation between the wealth of families that are five
generations apart.
4
Our empirical analysis also has other prominent strengths and elements of
novelty. First, we consider different socioeconomic outcomes, including earnings,
wealth and belonging to a profession. Indeed, most of the empirical evidence is
4
In the data used by Clark and Cummins (2014), the wealth is estimated at death, thus ignoring
inter-vivos transfers. Our data, on the contrary, have the advantage of being available when an
individual is an adult. Moreover, we can control for the evolution of the outcome variable in the
lifecycle by adding age among the controls.
7
focused on labor income, though wealth inheritance has recently attracted
renewed interest (Piketty, 2011; Piketty and Zucman, 2015). Second, ancestors’
socioeconomic status has been predicted using surnames at the city level, thus
generating more precise links across generations with respect to other studies that
use names or surnames at the national level. Moreover, the huge heterogeneity and
“localism” of Italian surnames further strengthen the quality of the pseudo-links
and represent an ideal setting for analyses that exploit the informational content of
surnames. Third, the Italian cities offer a unique background to trace family
dynasties and investigate the transmission of inequalities across the centuries. In
the 15
th
century, Florence, unanimously recognized as the cradle of the
Renaissance, was already an advanced and complex society, characterized by a
high level of economic development, a rich variety of professions and significant
occupational stratification. Today, Florence continues to display the same features,
and it can be considered as a representative city of an advanced country. Hence,
our results are, in principle, generalizable to other prosperous and developed
societies. Fourth, we are the first to provide a measure of (two-generation)
intergenerational earnings mobility in a pre-industrial society.
The rest of the paper is structured as follows. Section 2 presents the
empirical strategy. Section 3 provides background information and describes the
data and the variables. Section 4 shows the main empirical results, while Section 5
examines potential biases due to the quality of the pseudo-links and to selectivity
issues, and other robustness issues. Section 6 suggests some mechanisms behind
long run persistence. Section 7 concludes.
2. Empirical strategy
The main requirement when analyzing socioeconomic mobility is an
appropriate data set that spans over generations. Unfortunately, such a suitable
dataset is not easily available, and this is even more true if we consider generations
that are centuries apart. To overcome the problem, we adopt an approach that
combines information from two separate samples (TS2SLS) and whose properties
are discussed in Inoue and Solon (2010).
In the first sample, we have information about ancestors’ socioeconomic
outcomes (e.g. log of earnings), their surnames and some other covariates, and we
run the following regression:
????????????
????????????
????????????
= ????????????????????????
????????????
????????????
+ ????????????????????????
????????????
????????????
+ ????????????
????????????
????????????
(1)
8