1. Introduction
♣
“Prestige is an accident that affects human beings. It
comes into being and decays inevitably. […] It reaches its
end in a single family within four successive generations”
Ibn Khaldun
“Almost all the earnings advantages or disadvantages
of ancestors are wiped out in three generations”
Gary Becker and Nigel Tomes
1
Almost all of the theoretical and empirical studies on
intergenerational
mobility have focused on the correlation in socioeconomic status between two
successive generations – parents and their children – and have shared a common
view that the economic advantages and disadvantages of ancestors vanish in a few
generations.
2
In this paper, we question this view and empirically document the
persistence of socioeconomic status across generations that are six centuries apart.
This result is even more surprising: the huge political, demographic and economic
upheavals that have occurred in the city across the centuries were not able to untie
the Gordian knot of socioeconomic inheritance.
Linking people belonging to generations that are distant from each other is
difficult because of data limitations. In this paper, we exploit a unique dataset
containing the main socioeconomic variables, at the individual level, for people
living in the Italian city of Florence in 1427. These individuals (the ancestors) have
been associated, using their surnames, to their pseudo-descendants living in
Florence in 2011. Empirically, we use a two-sample two-stage least squares
(TS2SLS) approach: first, we use the sample of ancestors and regress the log of
♣
We thank Antonio Accetturo, Anthony Atkinson, Rosario Ballatore, Miles Corak, Domenico De
Palo, Marta De Philippis, Paul Gregg, Andrea Ichino, Paolo Naticchioni, Paolo Sestito, the conference
participants at AIEL, CIRET, SIE and SIEP and the seminar participants at the Bank of Italy, EUI,
University of Ferrara, University of Florence and University of Trento for their useful comments; we
also thank Giovanna Labartino, Peter Lindert and Giovanni Pica, who kindly shared data on the
distribution of surnames across provinces, data on incomes in 1427 and estimates of
intergenerational income mobility in Florence in 2005, respectively; we finally thank Riccardo
Innocenti and Massimiliano Sifone of the municipal statistical office of Florence for providing us
with tax records by surnames. The views expressed in this paper should be referred only to the
authors and not to the institutions with which they are affiliated.
1
Ibn Khaldun was one of the greatest Arab historians, and he is considered among the founding
fathers of modern sociology, historiography and economics; the citation has been drawn from his
influential book The Muqaddimah (1377). Becker and Tomes provided, in their seminal
contributions, the theoretical framework that represented the main building block of research on
intergenerational mobility; the citation was drawn from Becker and Tomes (1986).
2
Earnings persistence has been observed in all countries studied so far, although to varying
degrees. See Black and Devereux (2011) and Corak (2013) for recent surveys. Chetty et al. (2014)
moved the analysis at the local level, providing evidence across commuting zones within the U.S.
5
earnings on a full set of surname dummies (and, in some
specifications, also on age
and gender); second, we observe the current taxpayers present in the 2011
Florence tax records and regress the log of their earnings on that of their
ancestors, as predicted by the surname in the first step. The same strategy has
been repeated using the log of real wealth or dummies for professions, instead of
the log of earnings, as dependent variables.
3
We find that the elasticity of descendants’ earnings with respect to ancestors’
earnings is positive and statistically significant, with a point estimate around 0.04.
Stated differently, being the descendants of the Bernardi family (at the 90
th
percentile of earnings distribution in 1427) instead of the Grasso family (10
th
percentile of the same distribution) would entail a 5% increase in earnings among
current taxpayers. Intergenerational wealth elasticity is significant and equals
about 0.02, though the magnitude of the implied effect is even larger: the 10
th
-90
th
exercise entails more than a 10% difference in real wealth today. Looking for non-
linearities, we find some evidence of the existence of a glass floor that protects the
descendants of the upper class from falling down the economic ladder.
These results suggest that the persistence of socioeconomic status in the long
run is much higher than previously thought. In order to rationalize these findings,
we provide two explanations. First, we show that intergenerational mobility in the
15
th
century was much lower than at present: using the
methodology recently
proposed by Güell et al. (2015b), we estimate an intergenerational earnings
elasticity (between two successive generations) between 0.8 and 0.9, thus
depicting a quasi-immobile society in 1427. Second, we find evidence of dynasties
in certain (elite) professions. This latter result is consistent with our baseline
evidence on the long-run persistence of socioeconomic status. Moreover, it also
highlights a potential channel of inheritance – related to the market and non-
market mechanisms governing access to certain professions – and it helps to
explain why earnings elasticity does not necessarily decline geometrically, as
commonly thought.
Our empirical findings may suffer from two potential sources of bias. First,
the strength of the pseudo-links may be questioned, as we work with generations
that are six centuries apart. However, a rich set of robustness checks – including
3
Björklund and Jäntti (1997) were the first to apply the TS2SLS approach to intergenerational
mobility estimation. Thenceforth, the same strategy has been adopted for many country studies,
typically using occupation, education and sector of activity to predict pseudo-fathers’ earnings. On
the contrary, Aaronson and Mazumder (2008) used state and year of birth, while Olivetti and
Paserman (2015) exploited the information conveyed by first names. Some of these variables,
however, are partly endogenous, since they are related to parental characteristics, but they may
also directly affect children’s outcomes (e.g. parents’ education or state of residence), thus leading
to an upward bias. Surnames, in contrast, are markers that are more exogenous.
6