f rossi -dissertation
TRANSCRIPT
University of Kent School of Economics
MASTER IN AGRICULTURAL ECONOMICS
AN ANALYSIS OF INTERREGIONAL
MIGRATION IN ITALY
Dissertation
STUDENT SUPERVISOR Fabio Rossi Dr Bill Collier
SEPTEMBER 2012
2
ACKNOWLEDGMENT I would like to thank all the members of the School of Economics, in particular Prof.
Davidova who encouraged me to carry on with my studies in the most difficult
moment during the past year. Special thanks go also to my supervicor, Dr Collier,
whose advices and insightful comments made it possible the “climb to the top of the
mountain”.
3
INDEX ACKNOWLEDGMENT ............................................................................................... 2
Chapter 1 – Introduction ................................................................................................ 4
Chapter 2 – Italian migration history............................................................................. 6
2.1 – Historical background ...................................................................................... 6
2.2 – The “Empirical Puzzle” .................................................................................... 8
2.3 – The “Southern Question” ............................................................................... 10
Chapter 3 – Theoretical framework: a brief literature review ..................................... 14
Chapter 4 – Empirical literature .................................................................................. 21
Chapter 5 – Data and methodology ............................................................................. 24
5.1 – Empirical framework ...................................................................................... 24
5.2 – Data................................................................................................................. 26
5.3 – Descriptive Analysis ....................................................................................... 29
5.4 – Methodology................................................................................................... 31
Chapter 6 – Empirical results and discussion .............................................................. 34
6.1 – Negative Binomial Regression ....................................................................... 34
6.2 – Some remarks and discussion ......................................................................... 37
6.3 – Limitations and extensions ............................................................................. 39
Chapter 7 – Conclusion ............................................................................................... 41
References ................................................................................................................... 43
APPENDIX ................................................................................................................. 48
Word count: 12,033
4
Chapter 1 – Introduction
Migration of people across different areas has been studied as a complex
phenomenon involving mainly demographic and economic aspects (Etzo, 2008), and
is recognized to be an important mechanism through which the geographical
distribution of people changes over time (Greenwood, 1997). Italy has a long history
of internal migrations characterized by relative differences between different areas,
in particular North and South. Dualism, alongside differences in productivity and
local labour market conditions, have boosted migration and interfered in the process
of growth and convergence. Classical macroeconomic models consider migration as
an equilibrating mechanism that reduces differences among regions with respect to
key economic variables (e.g. unemployment, per capita income) (Etzo, 2008).
However, despite a history of significant structural change and intense migration
flows, the empirical evidence does not show substantial convergence between Italian
regions. This is an aspect that many researchers have for long stressed and that
remains unsolved (Capasso et al., 2011).
Italy is divided into twenty administrative regions, each charactherized by a
strong linguistic and cultural identity. Since the unification of Italy in 1860, there has
been an increasing gap between Northern and Southern regions in terms of economic
development, as well as the emergence of strong internal migration flows. Despite
significant economic differences between these two Italian areas, there has been a
marked drop in migration rates between the mid-1970s and mid-1990s, followed by a
sharp turnaround in more recent years.
Following the studies of various authors (Basile & Causi, 2005; Piras, 2010;
Etzo, 2010; Napolitano & Bonasia, 2010; Biagi et al., 2011), this dissertation
investigates the determinants of interregional migration in Italy, and consider how
migrants respond to changes in economic factors. In particular, the main interest is to
investigate the role of macroeconomic variables in determining the intensity and
direction of observed migration flows. In order to do so, we utilise an extended
Gravity model (Lee, 1966), where regional levels of GDP and unemployment are
considered alongside distance and population size as the main determinants of
5
migration. Gravity models were one of the first formal models of migration and
remain the most common theoretical framework in empirical migration analysis
concerning migration flows.
The remainder of the dissertation is organized as follows. Chapter 2 introduces
the historical background of the Italian migration and gives an insight into the
“empirical puzzle” that has characterized migration between the mid-1970s and mid-
1990s, and the “Southern Question” which has not been resolved following the
unification of Italy. Chapters 3 and 4 present a selected migration literature review,
both theoretical and empirical, dealing with the determinants of internal migration
flows. Chapter 5 describes the data and methodology employed in the empirical
analyses. Chapter 6 presents the empirical results and analysis, and provides an
economic interpretation of the key findings. Finally, chapter 7 provides some
conclusive remarks.
6
Chapter 2 – Italian migration history 2.1 – Historical background
Since the end of the Second World War the Italian economic system has been
transformed radically and a steady increase of the per capita income occurred
throughout the years in all Italian regions. A once large prevailing agriculture sector
has shrunk in favour of the industry and services sectors. However, the distinction
between advanced sectors (industry and services) and backward sectors (agriculture)
which characterises many dual economies, is accompanied in Italy by a severe
dichotomy between North and South, with the South chronically lagging behind
(Capasso et al., 2011).
Evidence of Italian migration is reported in Figure 2.1 and can be traced to the
period of unification which has characterised Italian economic history for over a
century. Between 1861 and 1985, it is estimated that more than 26 million italians
migrated abroad (Del Boca & Venturini, 2003). This large-scale migration away
from Italy is referred to as the “Italian diaspora” and it is considered the biggest mass
migration of contemporary times.
Figure 2.1 Italian migration abroad in thousands, 1876-1981. (Source: Golini & Birindelli, 1990 in Del Boca & Venturini, 2001.)
7
Over the second half of the last century, interregional migration flows in Italy
have gone through different phases. These can be summarised in three main trends.
The first one, which dominated the 1950s and 1960s, was characterized by intense
migration flows, mainly from rural to urban areas and from South to the more
industrialised North. A considerable increase in labour demand from the big
industries, located mostly in the North-West1, triggered the migration of people
working in the rural Southern regions, acting as a “pull” factor. Similarly, but in
opposite direction, the excess of labour supply in the agricultural sector played an
important role as a “push” factor, boosting the migration from the sending regions.
From the early 1970s internal migration markedly declined, a trend which
persisted till the mid 1990s. Differentials in per capita incomes and unemployment
rates, however, were still substantially high and did not decrease during this second
period (Faini et al., 1997). In fact, during this period migration rates seemed not to be
strongly correlated with unemployment and income differentials. The main feature of
the second cycle of internal migration flows, therefore, is the mismatch between
internal migration and regional disparities. Falling migration rates despite the
presence of strong regional disparities has become known as the “empirical puzzle”
and the failure of traditional theory to explain such a phenomenon has attracted the
interest of many researchers (Etzo, 2010).
Finally, after the mid 1990s internal migration flows grew again with a
significant flow of migrants from Southern to Northern regions, as reported by the
Italian National Institute of Statistics (ISTAT). Importantly, whilst the main
geographical patterns of these flows has not changed (i.e., from South to Centre-
North), its composition reveals some interesting changes. For instance, during the
period 1950-1970, migrants leaving the Southern regions of Italy were very young
and with low educational attainment. By contrast, the young migrants of more recent
years have been shown to be more skilled and five years older (on average) than
migrants during the 1960s (Etzo, 2010).
1 The so called “industrial triangle” (i.e., Turin-Genoa-Milan)
8
2.2 – The “Empirical Puzzle”
As drawn previously, the pattern of migration between the mid-1970s and mid-
1990s has been referred to as the “empirical puzzle” to indicate that, despite
increasing differentials in regional unemployment levels, mobility levels fell and
internal migration rates were surprisingly lower than in previous decades. The
reasons for this slow down in migration are still not entirely clear and various
explanations have been proposed by different studies (Biagi et al., 2011).
Attanasio et al. (1991) argue that past governments have used pensions as a
discretionary income subsidy, especially in the South, thereby increasing disposable
income in Southern households. This increase in disposable income in Southern
regions favoured by family and government support may explain the observed
decline in the propensity to migrate since any increase in net household income will
lower the expected net benefit of migration. Fachin (2007) analyses the determinants
of internal migrations during the period 1973-1996, and provides some empirical
support for this hypothesis.
Faini et al. (1997) provide a further explanation for the “empirical puzzle”,
arguing that the combination of interregional job mismatching and high mobility
costs could explain the marked fall in migration rates between the mid-1970s and
mid-1990s. Over this period, job agencies in Italy were publicly operated legal
monopolies which provided no training or information on job vacancies in other
regions (Etzo, 2010). This lack of information invariably resulted in an excessive
reliance of the unemployed on local networks when undertaking job search activities.
Punitive housing taxation and rent controls also served to increase the costs of
geographical mobility. Indeed, Faini et al. (1997) and Cannari et al. (2000) both
report evidence that housing market differentials discouraged interregional migration
in Italy between the mid-1970s and mid-1990s.
One final explanation for the empirical puzzle comes from the significant
translocation of labour demand over this period from the North-West region to the
North-East. Technological progress and changes in industrial structures required
more qualified and specialised workers than generic workers that had been hired in
9
the previous two decades. As a result, new potential migrants could not rely on the
old family networks as they did in the past (Etzo, 2010).
The “empirical puzzle” seems to come to an end during the 1990s. Indeed, as
can be seen from Figure 2.2, between the mid-1990s and 2000s, Italian South-North
migration flows rose again significantly. During this period, almost eight hundred
and fifty thousand (850,000) people moved from the South to the Centre-North. In
2000, gross migration flows reached the peak of nearly one hundred and fifty
thousand (150,000), the same level as in 1975 (Basile & Causi, 2005). Once again, a
variety of explanations for this migration turnaround have been proposed.
Figure 2.2 Change of residence from South to North 1975-2000 (Source: Basile & Causi, 2005)
First, as a result of the big financial effort that Italy made to join the European
Monetary Union, the development policies for Southern Italy were drastically
reduced (SVIMEZ, 2004) and the government support that had characterized the
1980s shrank considerably. This caused a significant decline in household disposable
income (Basile & Causi, 2005), presumably determining a change in the attitude of
family members toward the choice of migrating.
Second, the widespread use of internet among households, together with the rise
and development of private job agencies, improved the efficiency of the job
searching activities and reduced the migration risks for the potential migrants.
Increased information is fundamental for potential migrants to assess the real
differences between their region and the possible destinations (i.e., differences in
10
income, in unemployment rates, cost of living, etc.), thereby facilitating a better
match between the labour demand and the labour supply (Etzo, 2010).
Third, Basile and Causi (2005) point to the success of Northern industrial
districts following the devaluation of the Italian currency in 1992. This devaluation
caused an increase in the international competitiveness of the Northern industrial
sector and in export sales, thus inducing benefits mostly in the Northern areas with
stronger industrial specialization. The economic expansion in these areas has
contributed to determine in the second half of the 1990s a revival of South-North
migration flows (Basile & Causi, 2005).
Finally, it may be also the case that this migration turnaround is also related to
external issues such as an emerging EU-wide pattern of interregional divergence,
driven by increased competition between regions in response to the new era of global
competition (Biagi et al., 2011).
2.3 – The “Southern Question”
At the time of unification in 1861, Italy was a relatively backward country
compared to the more advanced western European nations. Agriculture was the main
productive sector and 60 per cent of the labour force was still toiling on the land.
From then on the Italian economy underwent deep changes, both in its productive
capacity and structure. Italy caught up with the level of output per head of the most
advanced economies and today shares a similar economic structure and a low rate of
growth (Malanima & Zamagni, 2010).
By the end of the nineteenth century, the poorest regions in Italy were located in
the South. At that time politicians and intellectuals started to discuss the causes of the
backwardness of the South (also known as Mezzogiorno), starting the so-called
debate on the “Questione Meridionale” (“Southern Question”) as the backward South
was then becoming a central feature of the Italian economy and society (Malanima &
Zamagni, 2010). This debate remains vivid nowadays.
Many explanations for the origins of regional disparities, especially the North-
South divide, and their persistence throughout the twentieth century have been
proposed. These can be broadly summarized by two contrasting views.
11
The first view, prevailing up to the 1990s, emphasises the “natural poverty” of
the South which arose from a dry climate, shortness of natural resources, low levels
of human and social capital, and a feudal heritage in the land system. The North, by
contrast, in particular the North-West, was a natural candidate for industrialization
because of a more favourable geographical position and natural endowments, more
advanced human and social capital, better transport infrastructure, improved credit
markets, and the existence of crucial manufactures (Felice, 2011).
The alternative view considers the South to be the wealthiest and more advanced
part of the Italian peninsula but exploited by the North. The Kingdom of the Two
Sicilies (the actual South without Sardinia) had larger monetary resources in 1860,
which were appropriated by the Kingdom of Piedmont-Sardinia and redirected by the
new Italian state in favour of northern infrastructures, to promote the process of
northern industrialisation, and to pay the debts of the House of Savoy, the royal
family leading the newly formed Kingdom of Italy (Felice, 2011; Aprile, 2011).2 At
this time, several Southern regions displayed important signs of modernity and
dynamism, showing a potential for industrialization similar to the North-West
(Felice, 2011). However, for many years after the unification, cutting-edge Southern
industries were dismantled or allowed to perish, due either to unfair competition
from the North, or as a consequence of explicit decisions made by the newly formed
Italian State (Aprile, 2011; De Crescenzo, 2002).3
New studies have been progressively emerging for the past few decades,
resulting in a deeper knowledge of the economy after 1861, especially with the
publication of new statistical series and analysis of the organization of economic
activity, technology, economic policy, North-South disparities and changes in
income distribution. This progress has been quite remarkable (Malanima & Zamagni,
2010).
2 Before the unification, Italy was divided into seven different territories and, according to the studies of Nitti (1993), the Kingdom of the Two Sicilies had about 65.7% of all the money circulating in the peninsula, followed by Papal States (14%), Duchy of Tuscany (12.9%), Kingdom of Piedmont-Sardinia (4%), Venice (1.9%), Lombrady (1.2), Parma and Modena (0.3%). 3 For a broad review of the industrial history in the Kingdom of the Two Sicilies, see De Crescenzo (2002) and Pedio (1977).
12
Figure 2.3 Per capita GDP in North and South Italy at 1911 constant prices, 1861-2004. (Source: Daniele & Malanima, 2007).
In the light of these new statistical series, Daniele and Malanima (2007) have
suggested that disparities within the regions where much more significant than those
between regions. They highlight that the difference between the two areas as
measured by GDP per capita was not significant in 1861 and approximately only 7%
by 1891. In the subsequent decades, this figure rose substantially, reaching a peak of
44% by 1940. A process of convergence was observed between 1951 and 1973.
However, a further period of significant and sustained economic disparities occurs
between the two regions thereafter (Fig.2.3).4
In terms of agriculture, Daniele and Malanima (2007) assert that in 1891, per
capita agricultural production was 10% higher in the South than the North. By
contrast, several authors (Fenoaltea, 2003; Fenoaltea & Ciccarelli, 2010) find
evidence of a pervasive similarity across these different regions in terms of industrial
structure. Indeed, as illustrated in Figure 2.4, only by 1911 is the northern industrial
triangle of Piedmont, Liguria, and Lombardy clearly apparent.
4 See Daniele & Malanima, 2007; Malanima & Zamagni, 2010 for details.
13
Figure 2.4 Indices of relative industrialization, 1871 and 1911. (Source: Fenoaltea, 2003)
Thus, at the time of unification, there were no significant differences between
the two Italian macro-areas and the view of a backward and underdeveloped South
with dramatically low levels of literacy and lack of infrastructures and natural
resources does not appear to be adequately supported by existing data or historical
documentation. The “Southern Question” debate over regional disparities seems far
from closed, not least because the extent and development of those disparities
remains highly conjectural (Fenoaltea, 2003).
14
Chapter 3 – Theoretical framework: a brief literature review
The modern economics literature on migration often is traced to Lewis' seminal
work on economic development with unlimited supplies of labor5. The Lewis model
consists of a dual economy with both "capitalist" and a "non-capitalist" sectors. The
“capitalist” sector develops by drawing labour from the non-capitalist “subsistence”
sector. The existence of “surplus labour” in the subsistence sector ensures that during
an extended period, wages in the capitalist sector remain constant because the supply
of labour to the capitalist sector exceeds demand at the prevailing wage rate. The
surplus of output over wages is captured by the capitalists as profits (Kirkpatrick &
Barrientos, 2004). Although Lewis does not propose an explicit migration model
himself, his seminal contribution is to explain the mechanisms by which an unlimited
supply of labor in traditional sectors of less developed countries might be absorbed
through capital accumulation and savings in an expanding modern sector. In practice,
the capitalist sector has generally become identified with the urban economy and the
non-capitalist sector with agriculture or the rural economy. If the capitalist economy
is concentrated in the urban economy, labour transfer implies geographic movement,
both for workers and households, and hence rural-to-urban migration (Taylor &
Martin, 2002).
The neoclassical theory predicts a positive role of migration in the convergence
process. People migrate from low capital intensity to high capital intensity regions,
reducing the capital intensity in the destination region and increasing it in the sending
region. The latter will grow faster than the former and migration flows will continue
till the convergence process finishes (Etzo, 2008). The dominant assumptions of the
neoclassical model are full employment, market clearing and perfect competition.
According to the neoclassical analysis, rural-to-urban migration exerts upward
pressure on wages and on the marginal value product of labour in rural areas, while
putting downward pressure on urban wages, assuming that wages adjust to ensure
that both rural and urban labour markets clear (Taylor & Martin, 2002). Thus, the
5 See Lewis (1954) for details.
15
most basic neoclassical model of the labour market asserts the convergence of wages
across sectors. Furthermore, assuming full employment of labour in both rural and
urban sectors, and minimal transactions costs, inter-sectoral wage differentials should
be the primary factors driving rural out-migration. (Taylor & Martin, 2002).
However, dispite the assumptions of perfect competition, homogeneity of
workers and market-clearing equilibrium outcomes, there are clearly differences
among the wages of individuals. In fact, in reality one can observe rigid wages,
persistent inter-firm wage differentials, layoffs and large-scale unemployment, all
facts that appear to challenge the validity of the model and its assumptions. In order
to reconcile theory and empirical facts, the core model of the neo-classical paradigm
has been amended and modified in various ways.
An alternative model which attempts to overcome the limitation of the
neoclassical analysis is the Harris-Todaro model. In contrast with neoclassical
model, the Harris-Todaro model (hereafter H-T model) does not assume full
employment and is thus able to explain the continuation of rural-to-urban migration,
even in presence of rising urban unemployment (Etzo, 2008). Harris and Todaro
(1970) propose a modification of the neoclassical framework and permit
imperfections in the labour market in the context of internal rural-to-urban migration,
thus relaxing the strict assumptions found in the basic model. They assert that the
unemployment rate and wage differentials between rural and urban sectors to be the
key elements of migration. Each potential rural-to-urban migrant decides whether or
not to move to the city based on an expected-income maximization objective.
The H-T model diverges from the usual full employment, flexible wage-price
models of economic analysis, where the achievement of a full employment
equilibrium is assured through appropriate wage and price adjustments (Harris &
Todaro, 1970). In H-T model, expected urban income at a given location is the
product of the wage and the probability to find a job (proxied by employment rate).
Expected rural income is calculated similarly. Individuals are assumed to migrate if
their discounted future stream of urban-rural expected income differentials exceeds
migration costs (Bojnec & Dries, 2005).
The H-T model can be formalized as
where pu(t) is the probability of urban employment at time
given employment, the term
rural sector and r is the discounted rate. If the
values holds, then rural
assumed to remain in the rural labour market.
The H-T framework implies that rural
urban unemployment can be economically rational if expected urban income exceeds
expected rural income. This is the power of the H
continuation and, frequently, acceleration of rural
of high and rising urban unemployment (Taylor & Martin, 2002).
higher wage in a different location may not be enough to encourage migration if it is
not coupled with a low unemployment rate.
probability of obtaining a j
lower in urban than in rural areas, where the conditional wage is low but the
likelihood of employment is high. Conversely, high rural unemployment will make a
given expected urban wage more conducive
Martin, 2002).
The above model is based on a certain number of restrictive assumptions and is
not without criticism. The most restrictive assumption of the model is that the
expected income is the only factor to affect the
any other form of influence that may shape potential migrants' decisions, and also
their potential impacts on rural economies, appears unduly restrictive. As a result, the
Harris-Todaro model is not able to explain why,
satisfied (i.e., urban-
only some individuals migrate whilst others do not.
potential migrants to be risk neutral, such that they are
expected rural income and an uncertain expected urban income of the same
magnitude. The validity of this assumption is questionable, since poor migrants are
likely to be risk averse and require a significantly greater expecte
is the probability of urban employment at time t, wu denotes urban wages
given employment, the term c represents the migration costs, wr
is the discounted rate. If the above relationship of discounted
values holds, then rural-urban migration will be induced. Otherwise, individuals are
assumed to remain in the rural labour market.
T framework implies that rural-urban migration in a context of high
nt can be economically rational if expected urban income exceeds
expected rural income. This is the power of the H-T model: its ability to explain the
continuation and, frequently, acceleration of rural-to-urban migration in the presence
urban unemployment (Taylor & Martin, 2002).
higher wage in a different location may not be enough to encourage migration if it is
not coupled with a low unemployment rate. A high urban wage coupled with a low
probability of obtaining a job at that wage may result in an expected wage that is
lower in urban than in rural areas, where the conditional wage is low but the
likelihood of employment is high. Conversely, high rural unemployment will make a
given expected urban wage more conducive to promoting migration (Taylor &
The above model is based on a certain number of restrictive assumptions and is
not without criticism. The most restrictive assumption of the model is that the
expected income is the only factor to affect the decision to migrate. The omission of
any other form of influence that may shape potential migrants' decisions, and also
their potential impacts on rural economies, appears unduly restrictive. As a result, the
Todaro model is not able to explain why, even when its basic condition is
-rural expected income differentials exceeds migration costs)
only some individuals migrate whilst others do not. Moreover, the model assumes
potential migrants to be risk neutral, such that they are indifferent between a certain
expected rural income and an uncertain expected urban income of the same
magnitude. The validity of this assumption is questionable, since poor migrants are
likely to be risk averse and require a significantly greater expecte
16
denotes urban wages
is the wage in the
above relationship of discounted
urban migration will be induced. Otherwise, individuals are
urban migration in a context of high
nt can be economically rational if expected urban income exceeds
T model: its ability to explain the
urban migration in the presence
urban unemployment (Taylor & Martin, 2002). More generally, a
higher wage in a different location may not be enough to encourage migration if it is
A high urban wage coupled with a low
ob at that wage may result in an expected wage that is
lower in urban than in rural areas, where the conditional wage is low but the
likelihood of employment is high. Conversely, high rural unemployment will make a
to promoting migration (Taylor &
The above model is based on a certain number of restrictive assumptions and is
not without criticism. The most restrictive assumption of the model is that the
decision to migrate. The omission of
any other form of influence that may shape potential migrants' decisions, and also
their potential impacts on rural economies, appears unduly restrictive. As a result, the
even when its basic condition is
rural expected income differentials exceeds migration costs),
Moreover, the model assumes
indifferent between a certain
expected rural income and an uncertain expected urban income of the same
magnitude. The validity of this assumption is questionable, since poor migrants are
likely to be risk averse and require a significantly greater expected urban income in
order to migrate. It is in this regard that the migration literature has evolved to
integrate the Harris-Todaro expected utility maximisation approach with a further
approach, namely that of human capital (e.g., Sjaastad, 1962;
emphasizes the role of personal characteristics (Etzo, 2008).
Human Capital Theory addresses the heterogeneous nature of the labour market,
relaxing the basic model assumption of homogeneity. The theory seeks to explain
wage differentials as a co
an individual’s marginal productivity. In fact, it is well documented that the earnings
of more educated people are almost always well above the average (Becker, 1964).
Human capital analysis expla
with better education and training. That is, individuals who invest money and time
gain skills that improves their human capital and eventually their productivity.
Sjaastad (1962) develops a microecon
is modelled explicitly as an investment in human capital. In this framework the
migration decision in the interregional migration context is explained on the basis
that each individual decides to move to a particular
total benefits to moving is greater than the present value of the cost of moving (Etzo,
2008). Thus, the decision to migrate is based on a cost
that the individual is rational and maximises
be represented by the following expression:
where i denotes the origin region and
benefits and C the total cost related to the respective region,
T is the lifetime period. If the net present value of migration (NPVM) is positive,
then the migration takes place. In this model, the benefits are represented by the
income earned by migrant in the two alternative regions, which
of personal human capital stocks. It is also important to point out the non monetary
nature of some migration costs, such as the psychic costs of leaving the origin region
(Etzo, 2008).
It is in this regard that the migration literature has evolved to
Todaro expected utility maximisation approach with a further
approach, namely that of human capital (e.g., Sjaastad, 1962; Becker, 19
emphasizes the role of personal characteristics (Etzo, 2008).
Human Capital Theory addresses the heterogeneous nature of the labour market,
relaxing the basic model assumption of homogeneity. The theory seeks to explain
wage differentials as a consequence of different human capital stocks that determine
an individual’s marginal productivity. In fact, it is well documented that the earnings
of more educated people are almost always well above the average (Becker, 1964).
Human capital analysis explains this as the increased productivity of those workers
with better education and training. That is, individuals who invest money and time
gain skills that improves their human capital and eventually their productivity.
Sjaastad (1962) develops a microeconomic model where the migration decision
is modelled explicitly as an investment in human capital. In this framework the
migration decision in the interregional migration context is explained on the basis
that each individual decides to move to a particular region if the present value of the
total benefits to moving is greater than the present value of the cost of moving (Etzo,
2008). Thus, the decision to migrate is based on a cost-benefit analysis and, assumes
that the individual is rational and maximises his/her utility function. The model can
be represented by the following expression:
denotes the origin region and j the destination region, B
the total cost related to the respective region, r is the discount rate and
is the lifetime period. If the net present value of migration (NPVM) is positive,
then the migration takes place. In this model, the benefits are represented by the
income earned by migrant in the two alternative regions, which in turn is a function
of personal human capital stocks. It is also important to point out the non monetary
nature of some migration costs, such as the psychic costs of leaving the origin region
17
It is in this regard that the migration literature has evolved to
Todaro expected utility maximisation approach with a further
Becker, 1964) which
Human Capital Theory addresses the heterogeneous nature of the labour market,
relaxing the basic model assumption of homogeneity. The theory seeks to explain
nsequence of different human capital stocks that determine
an individual’s marginal productivity. In fact, it is well documented that the earnings
of more educated people are almost always well above the average (Becker, 1964).
ins this as the increased productivity of those workers
with better education and training. That is, individuals who invest money and time
gain skills that improves their human capital and eventually their productivity.
omic model where the migration decision
is modelled explicitly as an investment in human capital. In this framework the
migration decision in the interregional migration context is explained on the basis
region if the present value of the
total benefits to moving is greater than the present value of the cost of moving (Etzo,
benefit analysis and, assumes
his/her utility function. The model can
B denotes the total
is the discount rate and
is the lifetime period. If the net present value of migration (NPVM) is positive,
then the migration takes place. In this model, the benefits are represented by the
in turn is a function
of personal human capital stocks. It is also important to point out the non monetary
nature of some migration costs, such as the psychic costs of leaving the origin region
This strand of literature predicts that young in
older ones because they can obtain greater returns to investments in migration over a
longer period of time. Moreover, individuals who have not yet invested in
themselves would have an incentive to migrate, and this partl
young migrate more than the old (Becker, 1964). Thus, age is significant in
influencing migration and must be considered in interpreting earnings differentials
over space and among occupations (Sjaastad, 1962).
The importance of location
the migration decision. Lee (1966) in his classic paper “A theory of migration”
conceptualises migration as involving a set of factors at origin and destination, and
the links between them. In Lee’s mo
relative population size are the main determinants of the migration process, which is
the result of “push” and “pull” factors discounted by the distance between the areas.
In fact, in every area there are m
or attract people to it, and there are others which tend to repel them (Lee, 1966).
Thus, the characteristics of the origin act to “push” an individual into migration,
while the attributes of the destina
Moreover, between every two locations there are some intervening obstacles which
impose a cost on migration either directly (e.g. removal cost, cost of seeking a job or
home) or indirectly, such as the amou
destination. It is clear that in Lee’s model, the distance represents an important factor
that can affect the migration process. Indeed, a greater distance results in stonger
barriers which may limit migration
history or culture between origin and destination) (Champion
theoretical framework developed by Lee (1966) can be formalized in the following
basic Gravity model (Lowry, 1966):
where the number of people
on the population size in each region (
between the two regions (
This strand of literature predicts that young individuals will be more mobile than
older ones because they can obtain greater returns to investments in migration over a
longer period of time. Moreover, individuals who have not yet invested in
themselves would have an incentive to migrate, and this partly explains why the
young migrate more than the old (Becker, 1964). Thus, age is significant in
influencing migration and must be considered in interpreting earnings differentials
over space and among occupations (Sjaastad, 1962).
The importance of location and distance can also be considered in the context of
the migration decision. Lee (1966) in his classic paper “A theory of migration”
conceptualises migration as involving a set of factors at origin and destination, and
the links between them. In Lee’s model, the distance between two locations and their
relative population size are the main determinants of the migration process, which is
result of “push” and “pull” factors discounted by the distance between the areas.
In fact, in every area there are many factors which act to hold people within the area
or attract people to it, and there are others which tend to repel them (Lee, 1966).
Thus, the characteristics of the origin act to “push” an individual into migration,
while the attributes of the destination “pull” the migrant to a particular location.
Moreover, between every two locations there are some intervening obstacles which
impose a cost on migration either directly (e.g. removal cost, cost of seeking a job or
home) or indirectly, such as the amount of information available about the area of
destination. It is clear that in Lee’s model, the distance represents an important factor
that can affect the migration process. Indeed, a greater distance results in stonger
barriers which may limit migration (e.g. national borders, differences in language,
history or culture between origin and destination) (Champion
theoretical framework developed by Lee (1966) can be formalized in the following
basic Gravity model (Lowry, 1966):
number of people Mij moving from region i to region j
on the population size in each region (Pi, Pj) and negatively on the physical distance
between the two regions (dij).
18
dividuals will be more mobile than
older ones because they can obtain greater returns to investments in migration over a
longer period of time. Moreover, individuals who have not yet invested in
y explains why the
young migrate more than the old (Becker, 1964). Thus, age is significant in
influencing migration and must be considered in interpreting earnings differentials
and distance can also be considered in the context of
the migration decision. Lee (1966) in his classic paper “A theory of migration”
conceptualises migration as involving a set of factors at origin and destination, and
distance between two locations and their
relative population size are the main determinants of the migration process, which is
result of “push” and “pull” factors discounted by the distance between the areas.
any factors which act to hold people within the area
or attract people to it, and there are others which tend to repel them (Lee, 1966).
Thus, the characteristics of the origin act to “push” an individual into migration,
tion “pull” the migrant to a particular location.
Moreover, between every two locations there are some intervening obstacles which
impose a cost on migration either directly (e.g. removal cost, cost of seeking a job or
nt of information available about the area of
destination. It is clear that in Lee’s model, the distance represents an important factor
that can affect the migration process. Indeed, a greater distance results in stonger
(e.g. national borders, differences in language,
history or culture between origin and destination) (Champion et al., 1998). The
theoretical framework developed by Lee (1966) can be formalized in the following
j depends positively
and negatively on the physical distance
19
From the previous paragraphs, we can see how migration theory has evolved
from the neoclassical model, based on “simple” wage-driven decisions, to Harris-
Todaro and Human Capital models with an increasing degree of complexity. More
recent studies of migration phenomena have led to the emergence of the New
Economics of Labour Migration (NELM), which has added explanatory power to the
neoclassical model by integrating the decision to migrate made by individuals in a
household decision-making process, in a context where the collective decision-
making not only maximizes the expected income but also minimizes risks related to
different market imperfections.
In the NELM model, pioneered by Stark and Bloom (1985), migrants are viewed
as financial intermediaries who provide their families with financial resources and
income insurance, such that migration can be seen as an opportunity to diversify risk
for the family. The argument here is that, in the presence of uncertainty of income,
migration may occur even in the absence of significant wage and unemployment
differentials (Daveri & Faini, 1999), since households may reduce total income risk
by having some of their working members relocated. The fundamental aspect of the
NELM is that migration is not solely the decision of an individual, but rather the
joint decision of members within a household. The idea that migration decisions are
taken by families rather than by a single individual is also emphasized by Mincer
(1977).
Migration may also occur in response to income inequality and relative
deprivation in origin regions, or be the consequence of asymmetric information. In
the first case, individuals are concerned about their relative social status and
migration may improve their social rankings at home. Indeed, NELM suggests that
people and households migrate not only to improve income in absolute terms, but
also to increase income relative to other households. Stark and Taylor (1989) have
argued that the decision to migrate is positively correlated with inequality in the
sending regions, and negatively correlated with inequality in the destination regions.
Under asymmetric information, low productivity workers may decide to migrate if
employers in the receiving areas are uninformed about individual workers’
productivity. In this case, employers cannot distinguish high-ability workers from
low-ability workers and the outcome is a pooling equilibrium where each worker is
20
paid according to the mean productivity rather than his unknown marginal
productivity (Daveri & Faini, 1999).
A further development of household models of migration is the dynamic
approach of networks models. Migration in these models is dynamic in the sense that
migration costs may be reduced by the increased information from previous migrants
(Filiztekin & Gokhan, 2008). The basic idea of these models is that migrants create
networks in the destination regions, which reduce both migration costs and risks for
new migrants, thereby influencing the expected income gains and uncertainty
associated with migration, and inducing future migration from the same origin
regions (Etzo, 2008).
In the context of these theoretical contributions, we can generally distinguish
between two approaches to the study of migration: a microeconomic and a
macroeconomic approach. The former focuses on the migration unit (e.g. single
individual or family) and related decision-making process. In this case, we need to
take into account those individual characteristics that might affect the decision to
migrate, such as age, gender and educational attainments. Macro approaches, instead,
focus on migration flows with respect to the spatial context and related aggregate
variables. Thus, we need to consider those factors associated with area of origin and
area of destination that affect the degree of attractiveness of a location. Since
aggregate migration flows represent the outcome of the underlying individual or
household decision-making process, the two approaches can be considered as
complementary (Etzo, 2008).
21
Chapter 4 – Empirical literature
The majority of empirical studies are based on aggregate-level analysis where
the decisions of migrants are summarized into migration flows across geographical
areas. Theoretical literature and empirical evidence suggest that interregional
migration is primarily driven by a disequilibrium mechanism in which migration is
mainly a response to spatial differences in economic factors such as wages and
employment opportunities.
Biagi et al. (2011) investigate the differences between long distance and short
distance migration within Italy. In the case of short distance migration, the authors
find that individuals give more weight to quality of life and amenities differences. By
contrast, in the case of long distance flows, economic/labour market variables
appears to be the key factors. Indeed, the results of their study show that people tend
to migrate to provinces with higher GDP per capita and lower unemployment rates.
The presence of the 20-39 age group is also important in attracting long distance
migrants. This is consistent with the finding that in many countries this is the age
group which is most mobile in response to wage signals and can obtain greater
returns to investments in migration, as suggested by Human Capital Theory.
Piras (2010) finds that relative per capita GDP and relative unemployment rates
were the most important drivers for internal migration in Italy between 1970 and
2002. However, other studies of Italian migration do not report a significant
influence of unemployment on mobility (Daveri & Faini, 1999; Fachin, 2007).
Daveri and Faini (1999) use aggregate data from the regions of Southern Italy to
study migration decisions taken by risk-averse households. They compare the out-
migration flows as shares of population from eight districts in Southern Italy into two
broad categories of destinations: abroad (outside Italy) and domestic (Northern Italy),
and find evidence to support their hypothesis that risk is a significant determinant of
the decision to migrate. They also report real wages to exert a negative effect on
internal migration while the unemployment rate is found to have no effect at all.
Similar results are reported by Fachin (2007) who finds that income growth in the
22
origin region is a significant driving force of migration, while unemployment rates
have only weak effects.
Basile and Causi (2005) analyse the determinants of net migration rates in the
Italian provinces in the period when the internal migration flows were in the
declining phase (1991-1995) and the period when the internal migration flows
increased (1996-2000). For the first period, their analyses reveal that net migration
was only weakly influenced by macroeconomic variables such as unemployment and
income. In the second period, however, migration behaviour appears more consistent
with traditional theories in which economic variables play a crucial role in explaining
internal migration. Indeed, the authors find a statistically significant negative effect
on migration for the unemployment rate and a positive and significant effect for
income.
Napolitano and Bonasia (2010) analyse internal migration flows in Italy in the
periods 1985-1995 and 1995-2006. Despite a substantial increase in regional
differentials in terms of unemployment rates and real per capita income in both
decades, migration flows in the two periods were characterized by different trends.
The authors investigate this divergent trend and consider the role of wage (proxied
by GDP per capita) and unemployment differentials, alongside house prices, and
externalities such as population density, carbon dioxide emissions and juvenile
delinquency. They find that house price differentials affected migration flows in the
first decade (characterized by falling migration) but there was no role for wages.
Conversely, in the second decade, wage differentials are found to be important
whereas unemployment and housing price differentials are not. These results
illustrate the complexity of the internal migration process in Italy and the limitations
of the H-T framework which does not permit dynamic considerations as captured in
the NELM.
Using panel data on gross migration flows between Italian regions, Etzo (2010)
analyses the role of macroeconomic determinants during the period 1996-2002. The
author distinguishes between origin and destination regions, investigating the impact
of the same macroeconomic variables in the two groups of regions. The empirical
results reveal per capita GDP and the unemployment rate to be the main economic
23
determinants. However, migrants do not respond to the unemployment rate in
destination regions. The outcome of the analysis shows that the geographical
reallocation of migrants is the result of the interaction of distance together with
“push” and “pull” forces.
Another Mediterranean country with a long record of migration is Spain. The
equilibrating role of internal migration in Spain has been investigated in various
studies. Bentolila and Dolado (1991) focus on interregional migration in the 1980s.
They analyse aggregate migration flows and find that both an increase in a region's
relative wage and a fall in its relative unemployment rate cause very little increase in
net migration to that region. Antolin & Bover (1997) and Devillanova & Garcia-
Fontes (1998) report similarly.
Regarding interregional migration, Germany is also of special interest since
there is evidence of structural differences between West and East Germany, which
results in a German “empirical puzzle” similar to the Italian case. During the 1990s,
East-West migratory movements did not fully react to regional labour market signals
as expected and, similarly to the Italian South-North migration between mid-1970s
and mid-1990s, the level of German East-West migration flows was substantially
low, despite the existence of large regional labour market disparities.6 Alecke et al.
(2009) analyse the relationship between regional disparities in labour market
variables and interregional migration flows in Germany following re-unification and
identify a clear role for regional differences in the real wage and unemployment rate
as key drivers of internal migration. Similar findings are reported by Parikh & Van
Leuvensteijn (2003) and Mitze & Reinkowski (2010).
6 Huge income transfers (politically driven) and the possibility of high East-West commuting are the likely explanations for this German ”empirical puzzle” (Alecke et al., 2009).
Chapter 5
5.1 – Empirical framework
For the analysis of
introduced in chapter 3. Recall, its basic formulation is:
which can be linearized by expressing all the variables in logarithmic form:
where the dependent variable is the gross migration from region
Pj are respectively the origin and destination region’s population,
between the main city of origin region and the main city of destination region. The
expected signs are positive for
of both the origin and destination regions is expected to affect migration positively
while distance should discourage migration. The distance is commonly used as a
proxy for all costs that, directly or indirectly, might affect migration decisions, such
as transportation costs, information costs, and psychological costs.
The basic Gravity model has been criticized for not addressing the causes of
migration and for not taking
(Fan, 2005). It is possible to reduce this weakness by adding variables that represent
the socioeconomic conditions of the origin and the destination (Lowry, 1966).
Accordingly, we extend the Gravity mo
regional labour markets and income levels.
Regional GDP per capita is used as a proxy for income or regional development
to account for the overall level of prosperity of each region. Regional unemployment
rates account for labour market conditions and the probability of finding job.
According to economic theory, per capita GDP is expected to be inversely correlated
with migration flows in the sending region. Conversely, an increase in per capita
GDP should “pull” migrants in the destination region. The unemployment rate is
Chapter 5 – Data and methodology
Empirical framework
For the analysis of internal migration in Italy, we adopt the Gravity model
introduced in chapter 3. Recall, its basic formulation is:
which can be linearized by expressing all the variables in logarithmic form:
the dependent variable is the gross migration from region i
are respectively the origin and destination region’s population,
between the main city of origin region and the main city of destination region. The
cted signs are positive for a1 and a2 and negative for a3. That is, population size
of both the origin and destination regions is expected to affect migration positively
while distance should discourage migration. The distance is commonly used as a
or all costs that, directly or indirectly, might affect migration decisions, such
as transportation costs, information costs, and psychological costs.
The basic Gravity model has been criticized for not addressing the causes of
migration and for not taking into account the decision-making process of migrants
(Fan, 2005). It is possible to reduce this weakness by adding variables that represent
the socioeconomic conditions of the origin and the destination (Lowry, 1966).
Accordingly, we extend the Gravity model framework to account for differences in
regional labour markets and income levels.
Regional GDP per capita is used as a proxy for income or regional development
to account for the overall level of prosperity of each region. Regional unemployment
account for labour market conditions and the probability of finding job.
According to economic theory, per capita GDP is expected to be inversely correlated
with migration flows in the sending region. Conversely, an increase in per capita
migrants in the destination region. The unemployment rate is
24
Data and methodology
internal migration in Italy, we adopt the Gravity model
which can be linearized by expressing all the variables in logarithmic form:
i to region j, Pi and
are respectively the origin and destination region’s population, dij is the distance
between the main city of origin region and the main city of destination region. The
. That is, population size
of both the origin and destination regions is expected to affect migration positively
while distance should discourage migration. The distance is commonly used as a
or all costs that, directly or indirectly, might affect migration decisions, such
as transportation costs, information costs, and psychological costs.
The basic Gravity model has been criticized for not addressing the causes of
making process of migrants
(Fan, 2005). It is possible to reduce this weakness by adding variables that represent
the socioeconomic conditions of the origin and the destination (Lowry, 1966).
del framework to account for differences in
Regional GDP per capita is used as a proxy for income or regional development
to account for the overall level of prosperity of each region. Regional unemployment
account for labour market conditions and the probability of finding job.
According to economic theory, per capita GDP is expected to be inversely correlated
with migration flows in the sending region. Conversely, an increase in per capita
migrants in the destination region. The unemployment rate is
25
expected to have a positive effect in the sending region and a negative effect in the
destination region.
In the simple gravity model, distance is used to proxy travel costs associated
with migration between regions. In this context, Cartesian co-ordinates are often used
to determine the spatial distance between locations. However, alternative measures
of travel cost that capture travel time and mode of transport may also be considered.
For example, travel time can capture the quality of the transport network and the
differential effects of road, rail and maritime transportation, the latter of which is
particularly relevant for the Italian regions of Sardegna and Sicilia which represent
the largest and most important islands in the Mediterranean Sea.
Finally, as discussed earlier, migrants may also consider amenity-related
characteristics that make a region more or less attractive relative to other regions.
Accordinlgy, we adopt the approach of Etzo (2010) and utilise a composite index
that controls for regional differences in infrastructure endowment.
Thus, the empirical specification for our extended gravity model takes the form:
migr = a0 + a1 popo + a2 popd + a3 dist + a4 gdpo + a5 gdpd + a6 unempo + a7
unempd + a8 infrao + a9 infrad + εit
A description of these variables is reported in Table 5.1. Descriptive statistics
are reported in Table 5.2. In particular, Table 5.2 reports the mean values for GDP
per capita (20,639 Euros), unemployment rate (8.8%) and infrastructures index
(92.6), as well as distance and travel time.
However, these figures are very different with regards to macroareas. Indeed,
Table A1 of the Appendix reports the minimum, maximum and mean values of the
variables used for the empirical analysis, distinguished between Centre-North (C-N)
and South (S). This allows us to make a comparison between the two macroareas.
The mean values show the existing substantial gap in GDP per capita (approx.
24,000 Euros in C-N and 15,000 in S), unemployment rate (approx. 5% in C-N and
15% in S), level of infrastructures (approx. 106 in C-N and 72 in S).
26
Variables Definitions
migr Interregional migration flows from region i to region j
popo Population in sending region
popd Population in destination region
gdpo Per capita GDP in origin region
gdpd Per capita GDP in destination region
unempo Unemployment rate in origin region
unempd Unemployment rate in destination region
infrao Infrastructure index for sending region
infrad Infrastructure index for receiving region
trtm Railroad travel time in hours between capital in region i and capital in region j
rdtm Road travel time in hours between capital in region i and capital in region j
dist Road distance in kilometers between capital in region i and capital in region j
εit Stochastic error term
Table 5.1 Variables descriptions
Variable Mean Std. Dev. Min Max
Interr Migrarion Flows 864 1319.8 1 9225
Population 2,853,366 2,256,302 119,410 9,071,124
GDP per capita 20,639 5,063.1 13,438 28,067
Unemployment Rate 8.8 5.6 2.5 22
Infrastructures Index 92.6 33.5 43.3 183.8
Road distance in Km 622.6 349.2 115 1659
Road Travel time in hours 7.15 4.17 1.35 17.25
Train Travel time in hours 7.16 4.59 0.37 24.21
Table 5.2 Summary statistics
5.2 – Data
The data come from the Italian National Institute of Statistics (ISTAT). Data on
gross migration flows are derived from the “Migratory movements of resident
population” published by the Italian National Institute of Statistics (ISTAT, 2006) for
the years 2001 and 2002. The vector representing the dependent variable is obtained
using the matrix of interregional movements. Since we have 20 Italian regions, the
27
matrix has a dimension of 20x20 with null values in the diagonal.7 Thus, we have
380 observations on gross migration flows for each year (2001 and 2002). Migration
flows indicate the number of people that, during each year, have changed their
official residency from one region to another (ISTAT, 2006). Gross migration allows
us to identify those “push” and “pull” factors that affect migration in origin and
destination regions.
Data on regional population, real per capita GDP and the unemployment rate
also come from ISTAT. In particular, regional population come from “Annuario
statistico italiano” (Istat, 2011), regional GDP from “Conti economici regionali”
(Istat, 2010) and regional unemployment rate from “Rilevazione sulle forze di
lavoro” (Istat, 2011). Population size is expressed as the annual average number of
people living (domicile) in each region. The unemployment rate is the ratio between
the unemployed males and females (aged 15 years and more) and the total labour
force (Istat, 2011).
The infrastructure index is provided by Istituto G. Tagliacarne (2001) and is
reported in Table A2 of the Appendix. The index includes all the main infrastructures
and considers both quantitative and qualitative aspects.8 Following Etzo (2010), this
index is computed as the ratio between the endowment measure and the demand
measure (expressed by the population and the region’s geographical extension).
Finally, road travel time and distance between Italian regions were obtained
using the Michelin Route Planner based on the Global Positioning System (GPS).
The capital of each region is chosen as the representative location, since in Italy
regional capitals are also those cities with the largest population within the regional
borders. Railroad travel times have been collected from Trenitalia who are the
primary train operator in Italy and owned by the Italian Government. Maritime
measures between Sardegna and Sicilia consider both overall distance and travel time
across the sea.
7 The Italian regions can be grouped into three macroareas (North, Centre and South). Piemonte, Val D’Aosta, Lombardia, Liguria, Trentino-Alto Adige, Veneto, Friuli Venezia Giulia and Emilia Romagna (North); Toscana, Umbria, Marche and Lazio (Centre); Abruzzo, Molise, Campania, Puglia, Basilicata, Calabria, Sicilia and Sardegna (South plus Islands). 8 Road network, railroad, seaports, airports, power plants, communication networks, banks, amenities, educational and cultural centres, health centres.
28
29
5.3 – Descriptive Analysis
Table 5.3 reports the population of each region, as well as regional GDP per
capita, unemployment rates and volumes of in-, out- and net-migration for 2001-
2002. Among the regions, Emilia-Romagna and Lombardia are the strongest gainers
in migration with high positive net migration. Lombardia accounts for about 16% of
Italy’s population and 17% of interregional in-migration in 2001. With regards to
Emilia-Romagna, these figures are 7% and 12% respectively.
Regions displaying negative net migration are all located in South. Campania
which accounts for approximately 10% of the Italian population and 14% of
interregional out-migration in 2001 is by far the leading donor region. Sicily and
Puglia are the next two largest donor regions, as further evidenced by their large
volumes of out-migration. A clearer pattern of interregional migration is illustrated in
Figure 5.1, which maps the data across the twenty Italian regions. It is clear that the
geographical patterns of interregional migration follow the direction from South to
Centre-North.
Figure 5.1 Gainer-loser regions and patterns of the largest out-migration flows (2001-2002). Regional GDP per capita (year 2001). (Source: own calculation).
30
Table 5.4 reports the 21 largest out-migration flows with the corresponding
region of origin and destination. The data reinforce what is evident in Figure 5.1,
namely, that interregional migration flows originate substantially from Southern
regions (Sicilia, Campania, Puglia, Calabria) and the main regions of destination are
located in the Centre-North (Lombardia, Emilia-Romagna, Lazio).
Region of origin Region of destination Out-migration
Sicilia (South) Lombardia (North) 9225 Sicilia (South) Lombardia (North) 9037
Campania (South) Emilia-Romagna (North) 8953 Campania (South) Emilia-Romagna (North) 8882 Campania (South) Lombardia (North) 8591 Campania (South) Lombardia (North) 8204 Campania (South) Lazio (Centre) 6956 Campania (South) Lazio (Centre) 6718
Puglia (South) Lombardia (North) 6013 Puglia (South) Lombardia (North) 5693 Sicilia (South) Emilia-Romagna (North) 5601
Lombardia (North) Piemonte (North) 5595 Piemonte (North) Lombardia (North) 5538
Lombardia (North) Emilia-Romagna (North) 5341 Lombardia (North) Piemonte (North) 5322
Calabria (South) Lombardia (North) 5308 Puglia (South) Emilia-Romagna (North) 5171
Calabria (South) Lombardia (North) 5117 Puglia (South) Emilia-Romagna (North) 5117
Lombardia (North) Emilia-Romagna (North) 5107 Sicilia (South) Emilia-Romagna (North) 4941
Table 5.4 The first 21 largest out-migration flows in 2001-2002. (Source: own calculations)
The disparity in economic development among the regions appears to be
correlated with the geographical patterns of interregional migration. Sending regions
are relatively poor, and most of the destination regions are economically more
developed with markedly lower unemployment rates. Using GDP per capita as an
indicator of regional economic development, Table 5.3 reveals Lombardia to be the
most developed region (27,929 Euros in 2001 and 28,067 Euros in 2002), closely
followed by other Northern regions. Southern regions hold the lowest levels of GDP
per capita, within the range 13,000-19,000 Euros (Fig.5.1).
Figure 5.2 reveals that a similar story can be told regarding unemployment.
Southern regions show high levels of unemployment rates. Among them, Sicily
31
holds the highest rate (22.0% in 2001 and 20.6% in 2002). The opposite is true for
the Northern regions, with Emilia-Romagna leading the ranking (3.2% in 2001 and
2.5% in 2002), closely followed by Trentino-Alto Adige and Lombardia.
Figure 5.2 Regional unemployment rate, year 2001. (Source: own calculation).
5.4 – Methodology
We utilise an econometric approach that takes into account that migration flows
are best considered as a count process that captures the number of times an event
occurs. Employing Ordinary Least Squares (OLS) in this context is unlikely to be
appropriate since the dependent variable can only take nonnegative integer values (0,
1, 2, ..., n). The truncated nature of count data implies that count data cannot have a
normal distribution such that OLS estimation is likely to be inconsistent.
Consequently, models that employ a probability distribution that can account for the
non-normal nature of count data are required. Two such models are the Poisson
Rergression Model and the Negative Binomial Regression Model.
The most popular specification of count data is the Poisson regression model, in
which the dependent variable is assumed to take integer values with probability
where Yi is the count variable,
particular count value,
count and can be related to a set of regressors:
A characteristic of the Poisson distribu
variance (equidispersion assumption):
This assumption is restrictive and is often violated in those applications where the
variance is found to be greater than the mean, with the consequence that standard
errors of the parameter estimates will be biased downwards (Wooldridge, 2009). This
problem, known as overdispersion, motivates alternatives to the Poisson Regression
Model which generalize the Poisson process in a variety of ways.
Descriptive analysis of our depende
variance to be greater than the mean.
generalized version of the Poisson regression model, namely the Negative Binomial
Regression Model. This version of the Poisson Model per
adjust the variance independently of the mean. More precisely, count data are now
assumed to be generated by a Poisson process
where
in which exp(εi) is drawn from a gamma distribution,
variance αβ2. 9 Variance = 1741951; Mean = 863.9
The most popular specification of count data is the Poisson regression model, in
which the dependent variable is assumed to take integer values with probability
is the count variable, yi is a strictly nonnegative number and represents a
particular count value, λi is the sole parameter representing the expected value of the
count and can be related to a set of regressors:
A characteristic of the Poisson distribution is that its mean is equal to its
variance (equidispersion assumption):
This assumption is restrictive and is often violated in those applications where the
variance is found to be greater than the mean, with the consequence that standard
he parameter estimates will be biased downwards (Wooldridge, 2009). This
problem, known as overdispersion, motivates alternatives to the Poisson Regression
Model which generalize the Poisson process in a variety of ways.
Descriptive analysis of our dependent variable (migration flows) reveals the
variance to be greater than the mean.9 Accordingly, we proceed to estimate a
generalized version of the Poisson regression model, namely the Negative Binomial
Regression Model. This version of the Poisson Model permits a second parameter to
adjust the variance independently of the mean. More precisely, count data are now
assumed to be generated by a Poisson process
is drawn from a gamma distribution, Gamma(α,β)
Variance = 1741951; Mean = 863.9
32
The most popular specification of count data is the Poisson regression model, in
which the dependent variable is assumed to take integer values with probability
is a strictly nonnegative number and represents a
is the sole parameter representing the expected value of the
tion is that its mean is equal to its
This assumption is restrictive and is often violated in those applications where the
variance is found to be greater than the mean, with the consequence that standard
he parameter estimates will be biased downwards (Wooldridge, 2009). This
problem, known as overdispersion, motivates alternatives to the Poisson Regression
nt variable (migration flows) reveals the
Accordingly, we proceed to estimate a
generalized version of the Poisson regression model, namely the Negative Binomial
mits a second parameter to
adjust the variance independently of the mean. More precisely, count data are now
Gamma(α,β), with mean αβ and
33
In the Negative Binomial model, the expected value of the count variable
remains λi = xi β and its variance becomes (λi + λi2α). That is, the variance is inflated
in order to address the overdispersion. The parameter α is called the overdispersion
parameter and the larger α, the greater the overdispersion. Therefore, if the
overdispersion parameter is significantly greater than zero, the data are over
dispersed and are better estimated using a Negative Binomial model than a Poisson
model. Notably, if the overdispersion parameter equals zero, the Negative Binomial
model reduces to the Poisson model. It is in this sense that we can consider the
Negative Binomial Model as a generalization of the Poisson Model.
34
Chapter 6 – Empirical results and discussion
6.1 – Negative Binomial Regression
The empirical analysis is based on the Negative Binomial regression. The
Negative Binomial regression models the logarithm of the expected count variable as
a function of the regressors. All regressors are in logarithms, thus allowing for a
direct interpretation of the results in terms of elasticities. Using the descriptive
information given in the previous chapter, we begin the analysis of the econometric
results presented in Table 6.1 by considering first the basic Gravity model (Model
A).
All three coefficients in Model A are statistically significant and have the
expected signs, as suggested by migration theory. The positive sign for both origin
and destination population indicates that an increase in population size acts both as a
“push” and a “pull” factor, resulting in more people to leave but also more migrants
from other regions. This also explains that the more populated regions experience the
highest migration flows. The origin population appears to be slightly stronger in
magnitude than the destination population, thus determining a negative net effect.
Distance between regions is identified as a deterrent to interregional migration
implying that factors such as travel costs and uncertainty regarding the destination
region impact negatively on internal migration flows between regions.
The basic Gravity model considers only the impacts of population and distance,
and as such fails to consider the effects of regional economic disparities.
Accordingly, Model B, considers also the impact of GDP per capita in origin and
destination regions. GDP per capita appears statistically significant for both sending
and receiving regions and shows, respectively, a negative and a positive sign. This
result is consistent with economic theory and suggests that migrants move mostly
from less to more economically developed regions.
Model C utilises an alternative macroeconomic indicator, namely the
unemployment rate, to capture observable economic disparities between regions. The
coefficients for unemployment rate are statistically significant and appear to be
35
correctly signed. An increase in the unemployment rate in the origin region causes an
increase in migration flows, whereas an increase in the unemployment rate in the
destination region causes a decrease in migration flows, ceteris paribus.
Table 6.1 Negative Binomial estimation
Variable Model A Model B Model C Model D
Origin Population 0.971*** 0.956*** 0.929*** 0.965***
Destination Population 0.932*** 0.933*** 0.952*** 0.907***
Distance in km - 0.322*** - 0.351*** - 0.365*** - 0.362***
Origin GDP per capita - 0.625*** 0.579*
Dest GDP per capita 0.527*** 0.297
Origin Unemployment rate 0.307*** 0.493***
Dest Unemployment rate - 0.186*** - 0.052
Origin Infrastructure index - 0.212*
Dest Infrastructure index 0.159
Constant - 19.484*** - 18.153*** - 19.170*** - 28.135***
Log of the dispersion parameter alpha - 0.749*** - 0.870*** - 0.887*** - 0.901***
Obs 760 760 760 760
Pseudo R2 0.1032 0.1117 0.1129 0.1140
Log likelihood - 5217.743 - 5168.013 - 5161.030 - 5154.694
Note: Dependent variable = interregional migration flows. All independent variables are in natural log. Likelihood ratio (LR) test: Reject H0 of equidispersion (p = 0.000) in all four models. Legend: *significant at the 0.05 level; **significant at the 0.01 level; ***significant at 0.001
Finally, Model D extends the Gravity framework to consider both
macroeconomics indicators and the role of amenities as proxied by an infrastracture
index for each region. The coefficients for population and distance remain
36
statistically significant and with the expected signs. However, the inclusion of both
the unemployment rate and GDP per capita leads to results which are rather difficult
to interpret in the light of migration and economic theory, but which are most likely
the consequence of multicollinearity between the macroeconomic variables as
revealed in Table A3 of the Appendix. The results reveal unemployment in the origin
region to exert a strong positive impact on migration flows. A 1% increase in the
unemployment rate of the origin (sending) region results in an increase of migration
flows of approximately 0.5%, ceteris paribus. On the other hand, unemployment rate
and GDP per capita do not exert any pulling effect in the destination region, as their
estimated coefficients are not statistically significant. That is, migrants decide to
leave if an increase in unemployment rate in their region occurs, regardless the
economic and labour market conditions in the destination region. This can be
explained by the fact that the strongest migration flows are from the poorer Southern
areas to the wealthier Centre-North, in a context of high differentials in terms of
unemployment rate and GDP per capita between the two macroareas. This would
explain why migrants do not react to unemployment rate and per capita GDP
variations in the destination regions.
Notably, origin GDP per capita has a positive sign in this latter specification
suggesting the higher is the income in the origin region, the larger are observed
migration flows. A 1% increase in GDP per capita in the sending region determines
an increase of migration flows by approximately 0.6%, ceteris paribus. Again, this
might be explained by the presence of high differentials between South and Centre-
North. Moreover, migration flows from South to Centre-North involve mostly young
unemployed migrants (Fig.A1 of the Appendix). Therefore, an increase in GDP per
capita would encourage people, especially young and financially supported by their
family, to take the risk of migration. It might also be argued that more disposal
income could help to finance the costs of moving, particularly in the presence of
widening gap between Southern regions and the rest of Italy (Faini et al., 1997; Etzo,
2010).10
10 A widening gap between the economic development of South and Centre-North has also been signalled in the 2005 report on the Southern Italian economy, published by the Association for the Development of Industry in the Mezzogiorno (Svimez, 2005).
37
In comparison, the coefficients for the infrastructure index in origin and
destination regions have the expected signs. However, only infrastructure in the
origin region is statistically significant at conventional levels. This implies that those
origin regions with higher level of infrastructures experience lower migration flows.
Increasing the infrastructures index in the origin (sending) region by 1% reduces
migration flows by approximately 0.2%, ceteris paribus.
In terms of model diagnostics, the Negative Binomial regression does not have a
measure of goodness of fit analogous to the R2 measure of Ordinary Least Squares
(OLS) estimation. However, the Pseudo R2 statistic can be used as one such measure.
Adding the macroeconomic variables and infrastructure index for sending and
receiving regions to the standard gravity model increases the Pseudo R2 from 0.1032
to 0.1140, indicating some improvement in the model fit. The overall value of the
Pseudo R2 appears low but this is not uncommon for models using Maximum
Likelihood Estimation. Finally, we perform a formal test of the null hypothesis of
equidispersion (α=0) against the alternative of overdispersion utilising a likelihood
ratio (LR) test. The outcome for each model reveals the presence of significant
overdispersion. Thus, our preference for the Negative Binomial Regression Model
over the Poisson Regression Model is supported.
As discussed in section 5.1, travel time may be a more appropriate measure of
travel cost than a simple measure based on Cartesian distance. Accordingly, we test
the robustness of our results further and re-estimate each of the four models using
travel time as a proxy for distance between regions. The results are qualitatively and
quantitatively similar to those presented and discussed above. Accordingly, we do
not discuss them further here though the resutls are reported in Table A4 of the
Appendix.
6.2 – Some remarks and discussion
Our analysis confirms some of the findings from previous studies (Basile &
Causi, 2005; Etzo, 2010; Napolitano & Bonasia, 2010). First, the results show that
unemployment rate plays a crucial role in determining interregional migration flows
in 2001-2002, thereby confirming the end of the “empirical puzzle” which
characterized internal migration up until 1994. Second, the gravity model approach is
38
relevant and effective for describing and explaining internal migration in Italy,
although the interpretation of different model specifications is not always
straightforward and suggests an avenue for further research. Finally, the analysis
represents a snapshot of the Italian economic situation in 2001-2002 at regional level.
It is evident that Italy is still characterized by relevant differences between South and
Centre-North, not only in terms of GDP per capita and unemployment rate, but also
in the level of infrastructure endowment. This huge disparity between the two
macroareas explains why most of the spatial patterns of internal migration in Italy
follows the direction from South to Centre-North.
In the presence of such dramatic economic disparities, it has often been
remarked that one way to reduce that gap and reverse the negative trend in the
economy of Southern Italy is to promote local development (SVIMEZ, 2002, 2004,
2007, 2011). This can be achieved by implementing a selective fiscal policy, funding
research centres, promoting technological innovation, giving incentives to private
companies that invest in the most disadvantaged areas, and providing those necessary
infrastructures that Southern regions are disperately in need. Obviously, all this
requires conspicuous financial resources and the success of a radical change in the
Southern economic system will depend primarily by the capacity of making a good
use of the endowments that already exist. To this purpose, it is also essential to
attract and keep the best human resources, reducing the phenomenon of “brain drain”
which might have detrimental effects not only for local labour market performances
but also for prospective local growth. However, financial and human resources are
not enough without a long-term vision and a strong political willing towards an
internal cohesion policy. Despite the most relevant recommendations, in the past ten
years there has been a substantial reduction in funds assigned to public investments
and a lack of strategic development policies. It has been recorded a general economic
impoverishment of Southern regions with an increasing number of unemployed
people. To some extent, this pertains also the rest of Italy, particularly following the
severe global recession that occurred in 2008-2009 and, then, the European financial
crisis.
39
6.3 – Limitations and extensions
The analysis highlights the role of regional disparity in determining the patterns
of interregional migration flows in Italy, and we have found important relationships
between migration flows and macroeconomic variables. Regional GDP per capita,
the unemployment rate and level of infrastructure in sending regions are shown to be
key determinants of migration flows in Italy, alongside the classical gravity
consideration of population and distance. Nonetheless, the empirical analysis is not
without limitation.
A first limitation is represented by the aggregate nature of the gravity model.
Indeed, while the gravity model is very successful in explaining the choice outcome
of a large number of individuals (e.g. migration flows), it is less likely to be able to
explain the decision choices of individuals or households at the microeconomic level
since it cannot sufficiently capture the heterogeneity of these groups or their
decision-making processes.
A second limitation concerns data availability. We utilise pooled cross-section
data for the years 2001 and 2002. Unfortunately, there are no such data for more
recent years. Moreover, the empirical analysis focuses only on a limited set of
macroeconomic indicators, in addition to those most commonly utilised within the
gravity model framework. Thus, our analysis may suffer from a lack of information
both in terms of explanatory variables for consideration and, in particular, any
dynamic effects of regional disparities on interregional migration behaviour.
Several steps can be made to improve model specification and the empirical
analyses. As mentioned, increased data coverage across a longer time frame would
permit one to estimate a dynamic version of the extended gravity model and test the
presence of migration networks effects, since present migration flows can be affected
by past migration flows. This could be achieved by including a lagged dependent
variable representing the number of migrants from the origin to the destination
region in the previous period.
Moreover, the empirical framework could be extended to consider additional
explanatory variables which relate to the stock of human and social capital, as well as
the quality of life available across different regions. Specifically, one could consider
40
the individual characteristics of migrants such as age, educational attainment, marital
status and employment status to analyse potential “brain drain” effects on
interregional migration flows.
Finally, it would be interesting to analyse our model in the context of migration
flows between the two core macroareas (i.e. South to Centre-North migration) and
also by excluding migration flows within the two macroareas and between close-by
regions. This may help to emphasize the role played by the main macroeconomic
variables in determining migration streams.
41
Chapter 7 – Conclusion
In this dissertation we have investigated the main determinants of interregional
migration flows in Italy in 2001-2002. The geographical patterns of internal
migration in Italy suggest that gravity variables (e.g. population and distance) are
important as well as macroeconomic variables, which enable us to take into account
regional economic disparities. Following previous studies in migration (Basile &
Causi, 2005; Fan, 2005; Etzo, 2010; Napolitano & Bonasia, 2010; Biagi et al., 2011),
different specifications of the gravity model have been estimated. The analysis uses
gross migration flows between each pair of Italian regions, which allows us to
identify the effects of our explanatory variables both in sending and receiving
regions. However, the analysis suffers from a limitation in the data (available only
for the years 2001 and 2002) and ignores the fact that migration can be strongly
affected by past migration flows. Moreover, other important variables are excluded,
particularly those related to human capital (e.g. age, education).
The gravity variables have the expected signs and are statistically significant in
all models’ specifications, whereas the coefficients for the macroeconomic variables
show unclear results. In fact, both their signs and significance are not consistent in
the different estimations. The origin unemployment rate appears to be the only
variable highly significant and consistent, acting as a strong push factor with an high
effect on migration flows. Considering that the strongest migration flows are
between Southern and Northern regions, the important role played by unemployment
rate can be connected to the dramatic gap in unemployment rate differentials
between South and Centre-North.
The empirical analysis demonstrates that models based on gravity principles are
appropriate for describing internal migration in Italy. However, it shows also that the
explanation of migration phenomenon is not straightforward for the Italian case.
What emerges clearly is the huge gap in terms of unemployment rate and GDP per
capita between the two macroareas of the country, causing an increase in
interregional migration flows.
42
In order to address the persistent regional economic disparities between South
and Centre-North Italy, it has been stressed the importance of promoting regional
development by implementing a selective fiscal policy, funding research centres,
promoting technological innovation, and providing those necessary infrastructures
that Southern regions are disperately in need. Despite the advice of many political
and academic observers to act promtly and effectively, South Italy has long been
waiting for the right policies to be put in place. Yet, the above mentioned measures
and investments seem far from being realized, especially in a context aggravated by
the harsh financial and economic crisis that all Europe has been experiencing for the
last few years.
43
References Alecke B., Mitze T., Untiedt G., (2009), “Internal Migration, Regional Labour
Market Dynamics and Implications for German East-West Disparities”, Ruhr
Economic Papers #96.
Antolin P., Bover O., (1997), "Regional Migration in Spain: the Effect of Personal
Characteristics and of Unemployment Wage and House Price Differentials Using
Pooled Cross-Sections", Oxford Bulletin of Economics and Statistics, vol. 59, n. 2.
Aprile P., “Terroni. All That Has Been Done To Ensure That The Italians of The
South Became ‘Southerners’”, 2011, Bordighera Press, New York.
Attanasio, Orazio P., Padoa Schioppa F., (1991), “Regional inequalities, migration
and mismatch in Italy, 1960–86”, in F. Padoa Schioppa (a cura di), “Mismatch and
Labour Mobility”, Cambridge University Press, Cambridge.
Basile R., Causi M., (2005), “Le determinanti dei flussi migratori nelle province
italiane: 1991-2001”, Università degli Studi “Roma Tre”, Dipartimento di
Economia, Working Papers No. 49.
Baum C.F., “Introduction to Modern Econometrics Using STATA”, 2006, STATA
Press.
Becker G., (1964), “Human Capital”, New York, Columbia University Press.
Bentolila S., Dolado J.J., (1991), “Mismatch and internal migration in Spain 1962-
86”, in F. Padoa Schioppa (ed.), “Mismatch and Labour Mobility”, Cambridge:
Cambridge University Press.
Biagi B., Faggian A., McCann P., (2011), “Long and Short Distance Migration in
Italy: The Role of Economic, Social and Environmental Characteristics”, Spatial
Economic Analysis, Vol. 6, No. 1.
Bojnec S., Dries L., (2005), “Causes of Changes in Agricultural Employment in
Slovenia: Evidence from Micro-data”, Journal of Agricultural Economics, Vol. 56,
No. 3, 399–416.
44
Cannari L., Nucci F., Sestito P., (2000), “Geografic labour mobility and the cost of
housing: evidence from Italy”, Applied Economics, vol. 32 (14), pp. 1899-1906.
Capasso S., Carillo M.R., De Siano R., (2011), “Migration Flows, Structural
Change and Growth Convergence. A panel data analysis of the Italian Regions”.
Department of Economic Studies, University of Naples “Parthenope”, Discussion
Paper No. 7/2011.
Champion T., Fotheringham S., Rees P., Boyle P., Stillwell J., “The Determinants
of Migration Flows in England: A review of existing data and evidence”, The
Department of Geography University of Newcastle, 1998.
Daniele V., Malanima P., (2007), “Il prodotto delle regioni e il divario Nord-Sud in
Italia (1861-2004)”, Rivista di politica economica, 97, n. 3-4, pp. 267-315.
Daveri F., Faini R., (1999). “Where do migrants go?”, Oxford Economic Papers, 51
(October), 595-622.
De Crescenzo G., “Le industrie del Regno di Napoli”, Grimaldi & C. Editori, Napoli
2002.
Del Boca D., Venturini A., (2003), “Italian Migration”, IZA Discussion Papers No.
938.
Devillanova C., Garcia-Fontes W., (1998), ”Migration across Spanish Provinces:
Evidence from the Social Security Records (1978-1992)”, Department of Economics
and Business, Universitat Pompeu Fabra, Economics Working Papers No. 318.
Etzo I., (2008), ”Internal migration: a review of the literature”, MPRA Paper No.
8783.
Etzo I., (2010), ”The determinants of the recent interregional migration flows in
Italy: A panel data analysis”, MPRA Paper No. 26245.
Fachin S., (2007), ”Long-run Trends in Internal Migrations in Italy: A Study on
Panel Cointegration with Dependent Units”, Journal of Applied Econometrics, Vol.
51, No. 4, pp.401-428.
45
Faini R., Galli G., Gennari P., Rossi F., (1997), "An empirical puzzle: Falling
migration and growing unemployment differentials among Italian regions",
European Economic Review, Elsevier, vol. 41(3-5), pages 571-579.
Fan C. Cindy, (2005), “Modeling Interprovincial Migration in China, 1985–2000”,
Eurasian Geography and Economics, 46, No. 3, pp. 165-184.
Felice E., (2011), “Regional value added in Italy, 1891-2001, and the foundation of
a long-term picture”, Economic History Review, 64, 3 (2011), pp. 929-950.
Fenoaltea S., (2003), “Peeking Backward: Regional Aspects of Industrial Growth in
Post-Unification Italy”, The Journal of Economic History, 63, n. 4, pp. 1059-1102.
Fenoaltea S., Ciccarelli C., (2010), “Attraverso la lente d’ingrandimento: aspetti
provinciali della crescita industriale nell’Italia postunitaria”, Quaderni di Storia
Economica (Economic History Working Papers), n.4, Banca d’Italia.
Filiztekin A., Gokhan A., (2008), “The Determinants of Internal Migration In
Turkey”, Conference EcoMod2008, Berlin, http://www.ecomod.org/files/papers/749.pdf
Greenwood M.J., (1985), “Human migration: Theory, models and empirical
studies”, Journal of Regional Science, 25: 521–544.
Harris J.R., M. Todaro, (1970), “Migration, unemployment and development: A
two-sector analysis”, American Economic Review, 60, 126-142.
ISTAT, Annuario statistico italiano, 2011
ISTAT, Conti economici regionali, Comunicato stampa, 28 settembre 2010,
www.istat.it/it/archivio/12718
ISTAT, Rilevazione sulle forze di lavoro, Comunicato stampa, 1 aprile 2011
ISTAT, “Movimento migratorio della popolazione residente. Iscrizioni e
cancellazioni anagrafiche. Anni 2001-2002”, (Italian National Institute of Statistics),
Annuario, n. 14 – 2006.
Istituto G. Tagliacarne, “La dotazione di infrastrutture nelle province italiane
1997-2000”, Ottobre 2001.
46
Kirkpatrick C., Barrientos A., (2004), “The Lewis model after fifty years”,
University of Manchester, Development Economics and Public Policy, Working
papers series, Paper No 9.
Lee E., (1966), “A Theory of Migration”, Demography, Vol. 3, No. 1, pp. 47-57.
Lewis W. A., (1954), “Economic Development with Unlimited Supplies of Labour”,
The Manchester School, vol. 22, no. 2, pp. 139-191.
Lowry I., (1966), “Migration and Metropolitan Growth: Two Analytical Reports”,
San Francisco, Chandler.
Malanima P., Zamagni V., (2010), “150 years of the Italian economy, 1861–2010”,
Journal of Modern Italian Studies 15(1) 2010: 1–20.
Mincer J., (1977), “Family Migration Decisions”, NBER Working Paper Series, No.
199.
Mitze T., Reinkowski J., (2010), “Testing the Neoclassical Migration Model:
Overall and Age-Group Specific Results for German Regions”, Ruhr Economic
Papers, #226.
Napolitano O., Bonasia M., (2010), “Determinants of different internal migration
trends: the Italian experience”, University of Naples "Parthenope", MPRA Paper
No. 21734.
Nitti F. Saverio, “Nord e Sud”, Calice (collana Piccola biblioteca meridionalista),
1993.
Parikh A., Van Leuvensteijn M., (2003), ”Interregional labour mobility, inequality
and wage convergence”, Applied Economics, Vol. 35, pp.931-941.
Pedio T., “Industria, società e classe operaia nelle province napoletane nella prima
metà dell'Ottocento”, Archivio Storico Pugliese, Bari, 1977.
Piras R., (2010), “Internal Migration Across Italian regions: Macroeconomic
Determinants and Accommodating Potential for a Dualistic Economy”, Working
Paper 2010.115, Fondazione Eni Enrico Mattei.
Sjaastad L., (1962), “The costs and returns of human migration”, Journal of
Political Economy, Vol. 70, (1962) pp. 80–93.
47
Stark O., Taylor J., (1989), “Relative deprivation and international migration”,
Demography 26: 1–14.
Stark O., Bloom D., (1985), “The New Economics of Labor Migration”, American
Economic Review, vol. 75, issue 2, pages 173-78.
SVIMEZ, “Rapporto 2002 sull’Economia del Mezzogiorno”, http://www.svimez.it/.
From the same website also annual reports 2004, 2005, 2007, 2011.
Taylor E., Martin P., “Human capital: Migration and rural population change”, in
B. Gardner and G. C. Rausser (eds), Handbook of Agricultural Economics
(Amsterdam: Elsevier Science, 2002).
Wooldridge J.M., “Introductory Econometrics. A Modern Approach”, International
Student Edition, South-Western, 4e, 2009.
48
APPENDIX
Centre-North South
Variables Mean Min Max Mean Min Max
Interr Migration flows 866 1 5,595 368 7 1,187
Population 3,044,776 119,410 9,071,124 2,566,250 320,757 5,713,244
GDP per capita 24,400 20,321 28,067 14,997 13,438 18,353
Unemployment rate 4.9 2.5 10.4 14.6 8.6 22
Infrastructures Index 106.4 46.2 183.8 71.9 43.3 96.6
Road distances in Km 370 115 750 516 128 1027
Road travel time in hours 4.04 1.35 7.56 7.27 2.04 16.27
Train travel time in hours 3.73 0.37 9 8.74 2.05 18.03
Table A1. Summary statistics distinguished between C-N and S
Regions Index (Italy=100)
Piemonte 89.2 Valle d’Aosta 46.2
Lombardia 120.3 Trentino Alto Adige 62.7
Veneto 115.9 Friuli Venezia Giulia 118.6
Liguria 183.8 Emilia Romagna 107.2
Toscana 117.1 Umbria 81.8 Marche 92.5 Lazio 142.0
Abruzzo 78.5 Molise 54.3
Campania 96.6 Puglia 81.6
Basilicata 43.3 Calabria 78.0 Sicilia 86.2
Sardegna 57.0
Italia 100
Table A2. Ranking list of regions according to the infrastructure index 1997-2000.
49
Figure A1. Interregional migration rates in Italy by gender and age, 2002 (Source: Istat, in Etzo, 2008)
50
Table A4. Negative Binomial estimation with travel time
Variable Model A Model B Model C Model D
Origin Population 0.938*** 0.918*** 0.884*** 0.929***
Destination Population 0.893*** 0.887*** 0.900*** 0.862***
Travel time in hours - 0.301*** - 0.342*** - 0.370*** - 0.378***
Origin GDP per capita - 0.715*** 0.723**
Dest GDP per capita 0.429*** 0.394
Origin Unemployment rate 0.361*** 0.598***
Dest Unemployment rate - 0.128** 0.041
Origin Infrastructure index - 0.264**
Dest Infrastructure index 0.117
Constant - 19.927*** - 16.691*** - 19.628*** - 30.920***
Log of the dispersion parameter alpha - 0.742*** - 0.860*** - 0.883*** - 0.902***
Obs 760 760 760 760
Pseudo R2 0.1026 0.1110 0.1127 0.1141
Log likelihood - 5220.749 - 5171.978 - 5162.172 - 5154.081
Note: Dependent variable = interregional migration flows. All independent variables are in natural log. Likelihood ratio (LR) test: Reject H0 of equidispersion (p = 0.000) in all four models. Legend: *significant at the 0.05 level; **significant at the 0.01 level; ***significant at 0.001