Economic Shocks and Their Effect On The Schooling and Labor Participation of Youth: Evidence From The Metal Mining Price Boom in Chilean Counties

The Annals of Regional Science
https://doi.org/10.1007/s00168-021-01069-8
ORIGINAL PAPER
Economic shocks and their effect on the schooling

and labor participation of youth: evidence from the metal
mining price boom in Chilean counties
Manuel Pérez Trujillo1 · Gabriel O. Rodríguez Puello2
Received: 29 July 2020 / Accepted: 11 June 2021

© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2021
Abstract
We analyze the effect of an exogenous economic shock on youth, specifically their
incentive for preferring to participate in the labor market over continuing their
education. We examine the Chilean case and the sustained increase in the market
price of the minerals extracted there during the period of Metal Mining Price Boom
between the years 2003 and 2011. Chile is a large-scale producer of metals, being
the world’s largest producer of copper and the third largest producer of molybde-
num. The prices of such metals increased steadily during the shock, what boosted
the economic growth and the demand for labor. This shock altered the skill com-
position in the labor market, increasing the jobs for low- and medium–low-skilled
workers in the counties more exposed to the shock, thus being the metal mining
counties the most affected by its impact. According to this, our results show a nega-
tive impact of the shock on returns to schooling that simultaneously reduced school
enrollment rates while increasing youth labor participation in those counties.
JEL Classification I26 · J24 · R11 · R23
1 Introduction
We aim to analyze how an exogenous economic shock, which presents a differenti-

ated effect across the subnational units in a country, could impact on the schooling
and labor force participation of youth, mainly in those territories more exposed to
* Manuel Pérez Trujillo

[email protected]
Gabriel O. Rodríguez Puello
[email protected]
1
Departamento de Economía, Instituto de Economía Aplicada Regional (IDEAR), Universidad
Católica del Norte, Avda, Angamos, 0610 Antofagasta, Región de Antofagasta, Chile
2
Universidad Tecnológica de Bolívar, Calle del Bouquet Cra.21 #25–92 Barrio Manga,
Cartagena, Colombia
13
Vol.:(0123456789)
M. P. Trujillo, G. O. R. Puello
the shock. In this regard, the classic popular theory of human capital (Becker 1964)
is a good guide to understanding the causal mechanism between an exogenous eco-
nomic shock and the incentives for schooling. A shock that can temporarily alter
the returns for schooling in the labor market could directly affect a young individu-
al’s decision to either join the labor market or continue their education (Willis and
Rosen 1979; Zarkin 1983).
For this purpose, we examine the case of Chile between 2003 and 2011, experi-
encing that country a sustained increase in the market price of the minerals extracted
there during this period. Chile is highly specialized in the extractive industry, being
a large-scale producer of metals and the world’s largest producer of copper and the
third largest producer of molybdenum. Nevertheless, the mining activity is mainly
concentrated in the northern regions, outstanding the case of Antofagasta where
this activity represents approximately 50% of the GDP and 25% of the employment
(OLAB 2016). Between 2003 and 2011, the commodity boom meant a sustained
increase in the price of raw materials due to1: (i) the impact of investor flow driving
through speculation (Singleton 2013) and (ii) the rapidly growing demand of emerg-
ing economies, i.e., China, which increased its imports of raw materials to sustain
economic development (Radetzki et al. 2008). The positive shock on raw material
demand resulted in higher prices and significant economic growth in Chile, mainly
in mining areas.
The Chilean labor market was directly affected by the positive business cycle.
During the price boom, Chile added approximately 1,000,000 new jobs,2 growth
that was concentrated in the mining sector, and particularly in the north of the
country. Counties3 specializing in mining saw rates of employment exceed the
national average. However, the increase in employment there was primarily in low-
or medium–low-skilled positions, most of which were either directly related to the
mining sector or indirectly by the positive impact that this sector generated over the
salaries, boosting the internal demand, for example, construction—which is also an
important sector for creating the necessary infrastructure to the mining activity—as
well as retail, trade and other services (Rehner and Vergara 2014). Additionally, the
salary boom had an impact in terms of its distribution on the labor market between
skilled and unskilled workers in the territories more exposed to the shock. Pellan-
dra (2014) studied the effect of the metal mining price boom on the Chilean labor
market and specifically those provinces with high indices of mining activity. Most
of the benefits in terms of higher local wages accrued to less educated workers,
while the increase in prices and exports (of raw materials) in those provinces had a
more modest effect on the wages of college educated workers. Therefore, such gains
1
As an example, the price of copper, which is the principal metal exported from Chile, went from nearly
USD 0.80 per pound in 2003 to over USD 4.00 per pound in the second quarter of 2008. The price then
stabilized between USD 3.00 and USD 4.00 through 2011.
2
Data from the Chilean National Statistics Institute (INE) Labor Force Survey comparing the levels of
employment between the years 2003 and 2011.
3
In this paper, we will use the US term “county” as they have similar administrative features and size,
comparable in some ways to comunas. Counties can contain typically several municipalities in much the
way that some comunas do.
13
Economic shocks and their effect on the schooling and labor…
contributed to a larger reduction in skill premiums4 in the most affected provinces

(by the metal mining price boom) compared to others.
The aforementioned context makes Chile an interesting case. Some studies ana-
lyze the effects of exogenous economic shocks on the schooling within the context
of an economic crisis at national level, but not considering the potential differences
of the shock at a local level. Their results tend to be ambiguous; some studies show
that a negative shock improves schooling enrollment as a result of the reduction
of opportunity costs between education and work (Goldin 1999; Mckenzie 2003;
Schady 2004). Other studies have found that when these shocks negatively affect
household income, they result in youth preferring employment (Funkhouser 1999;
Calero et al. 2009). This ambiguity is reflected in the work of Ferreira and Schady
(2009), who found that the income level in each economy is important in under-
standing the increase/decline in schooling enrollment during a crisis. In rich coun-
tries they found that the lack of opportunities in the labor market for younger indi-
viduals encourages them to remain in education. In poorer economies there seems
to prevail an income effect between schooling and joining the labor market, which
produces a pro-cyclical pattern between schooling and growth. Thus, during times
of crisis in countries where household income tends to decline, young people drop
out of school in order to participate in the labor market and therefore minimize the
shock’s impact on the family budget.5
Much less is known on whether natural resource export booms have a negative
effect on the schooling and working decisions of youth. Black et al. (2005) (hereaf-
ter BMS (2005)) carried out a leading study on the effects of the Appalachian coal
boom on high school enrollment in Kentucky and Pennsylvania in the USA.6 The
coal boom increased the recruitment of workers in the mining sector and their sala-
ries, most of whom possessed low educational levels. The shock reduced the value
of education in the market and consequently discouraged high school enrollment.
In their empirical analysis they found evidence that an increase of 10% in the sala-
ries for low-skilled workers meant a decrease of approximately 5–7% in high school
enrollment.
Despite the significant evidence found by BMS (2005), the authors did not assess
whether the decrease in high school enrollment during the shock produced an
increase in the labor force participation of youth. Therefore, although it is expected
that the shock will encourage young people to drop out of school, there is no direct
evidence on whether this shock produced a direct incentive for them to participate
in the labor market, motivated by changes in returns to schooling. We aim to address
4
The difference of the wages paid to high-skilled workers relative to wages paid to low-skilled workers.
5
According to Ferreira and Schady (2009), ‘declines in overall income levels raise the marginal util-
ity of whatever the child can contribute to the family’s budget today ([thus, having] an “antischooling”
income effect) (p. 2).’
6
Other studies analyze the impact of mining on consumption, poverty and inequality in local commu-
nities in Peru (Loayza and Rigolini 2016), poverty rates in Chile (Álvarez et al. 2018), or the negative
externalities of mining in relation to the economic activity of other economic sectors through compe-
tition for key inputs (such as land and labor) and environmental pollution in Ghana (Aragón and Rud
2012).
13
this knowledge gap by analyzing the Chilean case, also considering the different
impact of the shock across counties.
2 Theoretical framework
It is necessary to make use of the appropriate theoretical framework in order to illus-

trate the causal relationship that exists between an exogenous economic shock and
schooling and youth labor force participation, using for this purpose the theory of
human capital (Becker 1964). Assume an economy comprising two sectors which use
labor only for production. Both sectors differ in terms of the intensity of the use of
human capital. The firm’s hiring decisions are defined by the maximization of their
profit function7 which equates labor productivity with salary: y = w for a firm that
operates in the jth sector.8 Additionally, let us consider that labor(productivity
( )) depends
on the level of human capital accumulated by each hired worker h 𝜃j , having:
( )
y = A ⋅ h 𝜃j (1)
( 𝜕h 𝜃 )
( )
𝜃j being the number of years of schooling obtained 𝜕𝜃 j > 0 and A the produc-
j
tion efficiency of human capital. Using the solution of the profit maximization func-
tion and Eq. (1) we can establish:
w ≡ w 𝜃j = A ⋅ h 𝜃 j
( ) ( )
(2)
Taking into account (2) and that the level of human capital increases with years
of schooling, we can assume that there will be a positive wage difference in the mar-
ket in favor of those workers who are more educated, having:
( ) ( )
Ω = w 𝜃HS − w 𝜃n−HS > 0 (3)
The relationship obtained in (3) is expected when the market operates under nor-
mal conditions, which rewards higher levels of schooling. However, in the event of
an exogenous economic shock or other type of disruption, the composition of the
skill requirements for the labor market will be altered. The effect of such a shock
will distort the value of schooling, affecting the decisions of individuals with respect
to obtaining further education or joining the labor market.
To illustrate this, let us assume that a young individual of working age is faced
with this decision.9 Let us say that such an individual has a finite time horizon work-
ing life t = 0, … , T , comprising a specific number of years in which they can take
7
For the sake of simplicity, we will assume that workers from one sector cannot move to a company
belonging to another sector. We will also assume that the market price has a unit value.
8
We will only consider two sectors in our analysis: (i) a sector which only operates with high-skilled
workers (j = HS) and (ii) the remaining sector operating with only low-skilled workers (j = n − HS).
9
In order to simplify the present analysis, we will assume that the individual cannot choose to work
and study at the same time, and that once they decide to join the labor market then they cannot return to
school.
13
part in the labor market until the age T , which corresponds to the last year worked
in the labor market. In the case of making the decision to join the labor market and
considering (2), the individual will be faced throughout their working life with an
income defined by:
T T
( ) ( )
∫ ∫
w 𝜃n−HS ⋅ e−r⋅ t ⋅ dt = A ⋅ h 𝜃n−HS ⋅ e−r⋅ t ⋅ dt (4)
0 0
( )
where w 𝜃n−HS corresponds to the salary that would result from 𝜃n−HS years of
schooling and r being the discount factor. Furthermore, if the individual decides to
continue their education, he will have to meet the cost of acquiring the desired edu-
cational level 𝜃HS plus the time spent in education and therefore not receiving a sal-
ary, being the total cost defined by C (BMS 2005). In this way, the individual will be
faced with an income expectative such as:
T T
( ) ( )
∫ ∫
−r⋅ t
w 𝜃HS ⋅ e ⋅ dt + C = A ⋅ h 𝜃HS ⋅ e−r⋅ t ⋅ dt (5)
𝜃HS +1 𝜃HS +1
The effect that shocks exert over the market has yet to be determined, however its
impact is crucial in order to understand the changes that take place in such a mar-
ket and how they affect schooling decisions. According to BMS (2005) ‘a transitory
shock that increases demand for low-skilled workers […] will temporarily raise the
wages of high school dropouts relative to high school graduates. This increases the
opportunity cost of young people to remain in school, and may decrease the wage gap
between graduates and dropouts for the first year or so of work (p. 5),’ while the same
can be said about a positive shock that increases the demand for high-skilled workers.
Thus, let us suppose that an exogenous economic shock, independent of the sec-
tor in which it operates, alters the market temporarily and with it, the equilibrium
salary level goes up due to an increase in labor
( ( demand.
)) This fact has a positive
impact on wages in the market for that sector w 𝜃j , sj , so that:
b
( ) ( ) ( )
∫
w 𝜃j , sj = w0 𝜃j + sj ⋅ Δw0 𝜃j ⋅ dt (6)
a
( )
where w0 𝜃j corresponds to the wage that a worker
( ) receives with 𝜃j years of school-
ing irrespective of the economic shock, Δw0 𝜃j represents the absolute change in
wages during the shock, sj being a variable that can adopt three different values: (a)
sj = 1, when the shock has a positive impact on income, (b) sj = −1, when the shock
has a negative impact, and (c) sj = 0, when the shock has no impact. Lastly, the
parameters a and b identify the period of time of the shock over the economic sector
and with it, its duration over the salary.
By using (3), (4), (5) and (6), we can identify the impact that the exogenous eco-
nomic shock has on the performance of schooling in the market during working life,
having:
13
� ⎛ ⎞
T
� � �
⎜∫
Ω = 𝛾 sHS ⋅ ⎜ w 𝜃HS , sHS ⋅ e−r⋅ t ⋅ dt + C⎟
⎟
⎝𝜃HS +1 ⎠
(7)
� � ⎛ � �
T b
� � ⎞
∫ ⎜ 0 j ∫
− 𝛾 sn−HS ⋅ ⎜w 𝜃 + s ⋅ Δw0 𝜃j ⋅ dt⎟ ⋅ e−r⋅ t ⋅ dt
j
⎟
0 ⎝ a ⎠
( ) ( )
where 𝛾 sHS and 𝛾 sn−HS represent the perceived probability of being employed in
the high- and low-skilled sectors, respectively.
( ) These probabilities depend positively
on the exogenous economic shock sj because it increases production in that sector
by raising the profit expectations of companies and with it the number of job vacan-
cies posted (Petrongolo and Pissarides 2001). This increase in job-vacancies raises
(the probability of being )employed during the entirety of the economic shock
𝜕𝛾 (sHS ) 𝜕𝛾 (s )
𝜕s
> 0, 𝜕s n−HS > 0 .
HS n−HS
The impact of the economic shock on (7), regardless of the sector in which it
operates, will affect the incentives for the young person who has to decide between
continuing their education and joining the labor market. Thus, the probability of
a young person remaining in school (defined by the value 𝜙 = 1) is subject to the
value of (8.1) and (8.2), having:
Pr [𝜙 = 1|Ω > 0] > 0 (8.1)
Pr [𝜙 = 1|Ω ≤ 0] = 0 (8.2)
With:
{
1 (the individual remains in school)
𝜙=
0 (the individual dropouts the school)
The metal mining price boom positively affected the low and medium–low-skilled
workforce of Chile’s extractive activities as well as other sectors closely linked to
internal demand, such as construction, trade and other services. The boost on these
sectors was due to the positive effect that mining wages had on purchasing power,
especially in those counties where mining had a greater bearing on production activ-
ity (Pellandra 2014; Rehner and Vergara 2014). This impact on sectors with a low-
and medium–low-skilled workforce may have negatively affected school enrollment
while the shock lasted due to the increase in salaries, similar to that shown in (8.2).
This decrease in school enrollment should be accompanied by an increase in youth
labor market participation, motivated by the lower skill premium for schooling. In
this sense, the probability of youth labor market participation (Pr [𝜔 = 1]])10 should
10
When 𝜔 = 0, the individual doesn’t participate in the labor market.
13
be understood as the complement of the probability of young people remaining in

school (Pr [𝜔 = 1] = 1 − Pr [𝜙 = 1]).
3 Data and previous descriptive evidence
We will make use of the Chilean National Socioeconomic Characterization Survey

(CASEN), developed by the Ministry of Social Development. The survey measures
the socioeconomic conditions of the country’s households at a regional and county
level, although it only includes a certain number of counties, primarily those with
a high population. The survey provides socioeconomic information such as salary,
weekly working hours, educational level, status of the individual in the labor market,
etc. The period to be analyzed will be the one corresponding to the years 2000–2015
and will utilize the 2000, 2003, 2006, 2009, 2011, 2013 and 2015 CASEN surveys.
In spite of the fact that the theoretical model previously developed is defined at
an individual level, the empirical analysis will be carried out at a county level. Data
aggregation will allow us to identify changes that occur from before, during and
after the shock for the different Chilean counties, allowing for the creation of panel
data (similar to that of BMS (2005)).11 This will allow for an unbalanced panel for
the analyzed period with an average of 257 counties for each year of the survey. Fur-
thermore, we decide to aggregate the data at a county level since this is the smallest
administrative unit in Chile and counties are the closest territorial division to a local
labor market. Also, following Álvarez et al. (2018), we decide to aggregate the data
instead of working at a household level, because in the latter it might be more chal-
lenging to identify general equilibrium effects like spillovers to sectors not directly
related to the metal mining sector, for instance.
It is essential that we define a measure for identifying the intensity of the shock
on the economy. For this purpose, we use the measure defined by Álvarez et al.
(2018), who consider the price effect of all relevant metals in the Chilean economy
for the period analyzed. The authors define a price index that considers the five
principal metals in Chilean production (copper, silver, gold, molybdenum and iron
ore).12 They first define an average percentage change in the metal’s price for each
period t , as:
∑
5
Δpt
P̃ t,m = 𝜑2000
l,m .
l=1
pt
11
The process of individual data aggregation requires the assumption that the individual behavior theo-
retically identified in Sect. 2 is the one that defines the behavior of the population average.
12
According to Álvarez et al. (2018) ‘these metals represent over 99.5% of the production value between
1998 and 2013. Copper is by far the most important: it accounts for over 85% of the total production
value for each year (p. 4).’.
13
where l represents each of the five metals considered, 𝜑2000 l,m

is the production
value share of each metal l in 200013 in every m Chilean region (m = 1, … , 15)
–we consider regions, since there is no disaggregated data at the county level (with
∑5
𝜑2000 = 1) and pt is the nominal price of each one of the five metals. The price
i=1 i,m
and production data for the primary metals produced in Chile were taken from the
Chilean Mining Yearbook (Anuario de la Minería de Chile, AMC) provided by
SERNAGEOMIN, in order have production figures that are disaggregated by region.
The metal price index is computed as (see Appendix 1, Figure 2 for more detail
about the historical dynamic of the metal index at a national level, as compared with
the series of employment in mining industry):
( )
Pt,m = 1 + P̃ t,m ∗ Pt−1,m , with P2003,m = 100
This measure considers the intensity of the shock on the economy. However, we
also have to differentiate the impact of the shock depending on the county since
there is a greater effect on those that are metal mining producers. For this reason, we
define two types of classification for categorizing counties as metal mining produc-
ers due to the inexistence of an official classification. The first one is also defined by
Álvarez et al. (2018), who create a measure of exposure to the changes in Pt,m (for
each of the counties) taking into consideration the relative weight of employment in
the metal mining sector with respect to the total for each county in the first year of
the sample. In our case: year 2000. In this way, the interaction between both vari-
ables accounts for the effect of the metal mining boom in each county.
We create an alternative criterion to classifying metal mining counties as a
robustness check, in order to test the validity of the results obtained by applying
our first measure. We consider the data provided by the AMC14 in order to get an
approximation of the counties that are metal mining producers. The AMC provides
annual information regarding the mines in operation for every county and for any
type of resource exploited in Chile and is used for the years 2007, 2009, 2011, 2013
and 201515 in order to identify the counties that are metal mining producers. We
create a dummy variable, and a value of 1 is given to a county that has at least one
metal mine operating there during the period of analysis. Otherwise, a value of 0 is
given. With this criterion, we identify 46 counties that are metal mining producers
(see Appendix 1, Table 7 and Fig. 3).
It is important to consider that this alternative criterion only provides informa-
tion about the existence (or not) of the exploitation. It does not provide informa-
tion about the metal mining production or the level of employment in metal min-
ing sector in the county. Nevertheless, this second criterion allows us to eliminate
the effect of long-distance commuting (LDC)16 by properly identifying the counties
13
The first year of the sample is taken to avoid the potential existence of simultaneous causality in our
estimates.
14
See for example: http://www.sernageomin.cl/mineria/anuario-2016-sernageomin/.
15
Unfortunately, due to a lack of information, there is no AMC information for 2000 or 2003.
16
Long-distance commuting is a phenomenon related to individuals that work in a county/region differ-
ent from that in which they live. In Chile, this phenomenon is important, particularly in counties in the
13
that are metal mining producers, something that is not possible with the measure of
Álvarez et al. (2018). The Álvarez et al. (2018) measure considers all the workers
employed in the metal mining sector, independent of whether their job is located
in their county of residence. Including LDC to create the employment rate could
distort the accuracy of the Álvarez et al. (2018)17 measure used to identify the metal
mining counties, resulting in counties with a high relative weight of employment in
the metal mining sector with respect to the total, although there is no metal mining
exploitation there. This is something which does not affect the AMC measure.
Figure 1 uses this new definition to identify the dynamics of the average for
both the school enrollment rate and the youth labor force participation during the
period analyzed, dividing the analysis between metal mining and non-metal min-
ing producer counties. It also groups the analysis by age (a) 15–18 years and (b)
18–24 years, the first corresponding to individuals who should still be in compul-
sory education, but who can legally join the labor market. The results show that
there are greater oscillations throughout the shock (the period between 2003 and
2011) in the analyzed variables for the first age group, and that these oscillations
were particularly accentuated for metal mining producer counties. Between 2003
and 2009 the average school enrollment rate for the first age group in metal min-
ing producer counties dropped from 86 to 78.3%, which was accompanied by an
increase in labor participation from 9.6 to 15.5% in the same period. After 2009, the
school enrollment rate in these counties tends to increase until exceeding the value
reached in 2003, while labor participation is at lower levels than that year. In non-
metal mining producer counties, we observe a less abrupt oscillation, although the
school enrollment rate increased starting in 2009.
It should be noted that even though the shock started in the year 2003, its impact
on school enrollment rates was felt years later, specifically in 2009 and nearly at
the end of the shock’s duration. As the shock persists over time, the perception of it
being transitory becomes blurred and with it the perception that it will persist in the
long term. This conditions the behavior of the agents, which in this case generates a
negative effect on youth regarding their incentives for replacing education with labor
participation, something that becomes more pronounced the longer the shock lasts
(BMS 2005).
Regarding the 18–24-year-old age group, the changes tend to be smoother,
although a similar behavior in the dynamics is observed for metal mining pro-
ducer counties in the variables analyzed during and after the shock. What was
described for the age groups analyzed would previously coincide with our theoreti-
cal approach.18
Footnote 16 (continued)
north of the country where it represents approximately 20–25% of total employment (Aroca and Atienza
2008).
17
It is important to consider that the information provided by CASEN for the year 2000 does not allow
us to identify LDC workers. Therefore, there is no variable capable of improving the accuracy of the
employment rate in the metal mining sector.
18
In addition, we performed an ANOVA to test whether the differences observed in Fig. 1 between the
metal mining and non-metal mining producer counties are significant in both the school enrollment rate
and labor force participation rate of young individuals. Our results indicate that these differences are
13
4 Empirical analysis
Continuing with the formulated theoretical framework, we will now evaluate if the
events in the Chilean counties during the metal mining price boom have presented
a similar behavior to what was proposed theoretically. We expect that the economic
shock has modified the returns to schooling in the labor market for those metal min-
ing producer counties. This modification has simultaneously altered the incentives for
both schooling and youth labor participation in those counties. Thus, since the mecha-
nism that transmits the shock on schooling enrollment and labor participation are the
returns to schooling, we propose estimating a simultaneous equation model, having:
rate_school_enrvt = 𝛽0 + 𝛼m + 𝜇t + 𝛽1 ⋅ schooling_returnsvt + 𝜀vt (9.1)
rate_part_empvt = 𝜋0 + 𝜔m + 𝜏t + 𝜋1 ⋅ schooling_returnsvt + vvt (9.2)

[ ] [ ]
schooling_returnsvt = 𝛾m + 𝜆1 ⋅ zvt + 𝜆2 ⋅ Ln Pt,m + 𝜆3 ⋅ zvt ⋅ Ln Pt,m
(9.3)
+ 𝜆4 ⋅ emp_rate_difvt + uvt
where the variable rate_school_enrit , corresponds to the school

enrollment rate for youth in the vth county and the period
t = 2000, 2003, 2006, 2009, 2011, 2013, and 2015, rate_part_empvt is the labor
force participation rate, and schooling_returnsvt represents the returns to school-
ing. In order to measure this last variable, we consider two different approaches:
(1) by applying a Mincerian wage equation,19 and (2) the difference between the
Footnote 18 (continued)
significant, in that metal mining counties have a higher rate of school enrolment and a lower rate of labor
force participation in comparison with the other group. We have also performed an additional ANOVA
to test if the two groups of counties considered present significant differences in the level of poverty.
This new analysis is performed under the logic that poorer households might face budget constraints
in keeping their children enrolled at school (Carneiro and Heckman 2002). The outcomes of this new
analysis show that non-metal mining producer counties have a higher (and significant) level of both pov-
erty and extreme poverty than the metal mining producers. These new results combined with the previ-
ous ANOVA indicate that the socioeconomic conditions in non-metal mining counties, particularly the
higher level of poverty, is likely to affect the school attendance and labor participation of young individu-
als in these counties and might explain the differences observed between groups of counties in Fig. 1.
19
To calculate this variable the following Mincerian wage equation will be estimated for each county,
using the information available in CASEN for all the individuals that earn a salary in the county, correct-
ing by sample selection bias:
ln wi,v = 𝛼0 + 𝛼1 ⋅ Year_Schooli,v + 𝛼2 ⋅ Agei,v + 𝛼3 ⋅ Age2i,v + 𝛼4 ⋅ Genderi,v + 𝜃h,v + 𝜆i,v + ui,v
i being each worker between 15 years old or over resident in a county v. The estimate includes years
of schooling (Year_Schooli,v), age ( Agei,v) -as a proxy of the potential working experience in the labor
market (Mincer 1974) and its square, individual’s gender (Genderi,v), the hth economic sector in which
he/she participates (𝜃h,v) and the Mills inverse ratio associated to the probability of being employed
in the county (𝜆i,v). To calculate this last variable a probit model is applied to identify the probability
of being employed, using: years of schooling, age and its quadratic mean, gender and the number of
people per household as regressors. This regression is applied to each of the counties in the different
CASEN surveys used in the model; the predicted value being 𝛼̂ 1 which will be identified as the variable
schooling_returnsvt in (9.1, 9.2 and 9.3) (see Appendix 1, Table 9 which shows the yearly regional aver-
age of schooling returns estimated).
13
Fig. 1 Comparison of the dynamics for averages of school enrollment rates and labor force participa-
tion rates in both metal and non-metal mining producer counties between 2000 and 2015, using AMC
criterion. The analysis groups the population by age (a) 15–18 years old and (b) 18–24 years old. Source
Self-elaborated using CASEN database for the period 2000, 2003, 2006, 2009, 2011, 2013 and 2015.
Averages only consider counties with a population greater than 25,000 individuals, since the data pro-
vided by the CASEN for small-sized counties might not be representative (Modrego & Berdegué, 2015).
As a result, an average of 19 counties have a population of at least 25,000 individuals and are identified
as metal mining producers between 2000 and 2015
average salary obtained for high- and low-skilled workers in each county. In this
sense, a high-skilled worker is classified as someone who has completed a univer-
sity or technical education. Low-skilled are those that have completed at most a
high school education. Therefore, a different estimate of Eqs. (9.1), (9.2) and (9.3)
will be applied with each of the defined measures. The variables 𝛼m, 𝜔m, and 𝛾m are
regional fixed effects, 𝜇t , and 𝜏t being time fixed effects. The variable zvt identifies
each one of[ the ]two criteria previously considered to define a metal mining county.
Finally, Ln Pt,m represents the price index defined in logs by region and the vari-
able emp_rate_difvt defines the ratio between the employment rate of high- and low-
skilled workers, this being a proxy of the probability that a high-skilled worker has
of being employed with respect to a low-skilled worker.
13
In addition,
[ it] is important to highlight that (9.3) considers the exogenous economic
shock (Ln Pt,m ) as an intervention impacting the country’s metal mining counties
(zvt). Equations (9.1), (9.2) and (9.3) define a structural simultaneous equations system,
rate_school_enrvt, rate_part_empvt, and schooling_returnsvt being the endogenous
variables in the model. The system of equations presented in its structural form is very
useful, since it allows us to estimate the impact of the shock on both school enroll-
ment and youth labor participation through the returns to schooling. We use the 3SLS
(instead of the 2SLS) methodology for solving this system of equations as it allows us
to exploit the correlation of the disturbances across equations since there may be differ-
ent unobserved elements that jointly affect evt, vvt, and uvt, thus improving the efficiency
of the estimates.
Our main interest is focused on the coefficient [ ] associated with the interaction
between the exogenous economic shock (Ln Pt,m ) and being a metal mining county
(zvt), i.e., the estimation of 𝜆3, since this isolates the impact of the shock via the returns
to schooling in those counties. However, this coefficient does not completely measure
the effect of the shock on both school enrollment and youth labor force participation.
In order to have a measure of both effects, we need to estimate the product of 𝜆3 ⋅ 𝛽1
and 𝜆3 ⋅ 𝜋1, which is performed in the estimates through a non-linear combination of
estimated coefficients in the model. The estimates of the simultaneous equations model
are calculated for two different age groups: 15–18 and 18–24, divided by the same age
group’s population defined previously in the analysis of Fig. 1.
On the other hand, the estimates defined require controlling for a potential source of
measurement error bias. This is because the sampling carried out by the CASEN for
the counties with a small population size might not be representative of the selected
variables in the analysis (Modrego and Berdegué 2015). This can introduce bias into
the estimates, particularly when the regressors used show measurement errors. Thus,
considering the following model for yvt:
yvt = 𝛾1 ⋅ x̃ vt + 𝜀vt
( )
where, x̃ vt = xvt + uvt Nvt is an approximate
( measurement
) of the real value for xvt
which includes an error measurement uvt Nvt , which depends on the size of the
population where the sample is taken from, having to:
( )
lim uvt Nvt = 0
Nvt →∞
Taking into consideration the aforementioned, the estimator 𝛾̂1 works (Greene
2012):
𝛾1
plim 𝛾̂1 =
𝜎u2
vt
1+ 𝜎x2
vt
( )
Therefore, the lower the population size Nvt of observable units, i.e., counties,
in the study, the greater the measurement error bias, and hence the limit of the cal-
culated estimator 𝛾̂1 which will converge in probability to a value close to 0. Usu-
ally, the way to correct a measurement error bias is the application of instrumental
13
variables (Angrist and Krueger 2001). For this, appropriate instruments must be
found which can correct the potential source of correlation between the endogenous
variables and the model’s residuals. This is extremely complex for the present study,
since the source of error is in the sample developed for counties with a small popu-
lation and not due to a wrong quantification of the variables used in the analysis.20
To solve this potential source of endogeneity, three different estimates of the simul-
taneous equations model are proposed in terms of the population size of the counties
considered in the study: (a) using the whole available sample, (b) considering only
those counties with at least 25,000 inhabitants and (c) considering only those coun-
ties with at least 50,000 inhabitants. As a result, it is expected that the greater the
population size of the analyzed observable units, the more stable the estimated coef-
ficients are likely to be, reducing the impact of measurement error bias.21
5 Results
Table 1 shows the estimates of (9.1), (9.2) and (9.3) for the 15–18-year-old age
group,22 taking into consideration the relative weight of the employment held in the
metal mining sector with respect to the total for each county in the first year of the
sample, as a measure of a county’s exposure to the economic shock (similar to Álva-
rez et al. (2018)). Likewise, this table takes into account the two measures proposed
to define the returns to schooling (schooling_returnsvt ), either through (1) its estima-
tion via the calculation of the Mincer salary equation or (2) the difference between
the average salary obtained for high- and low-skilled workers in each county. The
estimates apply for three different sample subgroups (the entire sample of countries
and those counties with a population of at least 25,000 and at least 50,000), in order
to deal with the aforementioned measurement error bias.
All of the estimations show that the school enrollment rates and labor force par-
ticipation rates were significantly affected by the returns to schooling, regardless of
the measure used to calculate this variable. We observe that the increase in returns
to schooling in the market improves enrollment for the 15–18-year-old age group
in the period analyzed (see the estimates for Eq. (9.1)). At the same time, it also
statistically and significantly discourages their participation in the labor market (see
the estimates for Eq. (9.2)). These results are consistent with what is theoretically
expected.
20
The measurement error is due to the sample itself, therefore any variable (nvt ) originating from the
analyzed database and used as an instrument would also carry with(it a potential problem of error meas-
urement, which would violate the orthogonality of the residues: E nvt ⋅ 𝜀vt ≠ 0, which would limit the
)
use of this technique.
21
The elimination of counties with small populations from the estimates carried out will produce results
that can only be valid for the sample under consideration. However, other authors have applied similar
distinctions in their applied studies. Card (2001) analyzes the impact of migration on the local labor mar-
kets in the U.S. and includes only 175 of the biggest cities (out of a total of 324), due to a lack of migra-
tion data.
22
See Appendix 1, Table 8 for descriptive statistics.
13
Table 1 3SLS estimates for 15–18-year-old individuals, using the relative weight of employment in the metal mining sector with respect to the total employment for each
county in 2000 as the criterion to identify the counties most exposed to the shock (Álvarez et al., 2018). Source Self-elaborated using CASEN database for the period
2000, 2003, 2006, 2009, 2011, 2013 and 2015.
13
Population (1) (2)
Sch_returnit estimated by Mincerian wage equation Sch_returnit calculated as the difference between the average salary obtained
for high- and low-skilled workers
Whole sample (≥ 25, 000) (≥ 50, 000) Whole sample (≥ 25, 000) (≥ 50, 000)
Estimates of Eq. (9.1): rate_school_enrvt

schooling_returnsvt 1.3223 1.4580 1.7251 0.0052 0.0026 0.0025
(coefficient estimated: 𝛽1) (0.2255)*** (0.2565)*** (0.3459)*** (0.0010)*** (0.0006)*** (0.0006)***
𝛽0 71.0079 72.3807 (4.8509)*** 68.9958 70.2890 71.1841 71.1694
(2.9031)*** (5.8733)*** (3.8072)*** (4.1963)*** (4.1026)***
Estimates of Eq. (9.2):rate_part_empvt
schooling_returnsvt −0.6277 −0.6308 −0.6386 −0.0021 −0.0010 −0.0008
(coefficient estimated: 𝜋1) (0.1593)*** (0.1636)*** (0.1949)*** (0.0005)*** (0.0004)*** (0.0003)**
𝜋0 14.7942 11.7417 11.9769 14.5512 11.7637 10.3787
(2.0869)*** (3.1920)*** (3.3950)*** (2.0123)*** (2.7710)*** (2.5275)***
Estimates of Eq. (9.3):schooling_returnsvt
zvt(using Álvarez et al. 0.5582 0.7008 0.7595 117.5399 317.6370 (110.3000)*** 296.3672 (136.6847)**
(2018) definition) (0.1508)*** (0.2318)*** (0.2499)*** (78.5453)
Ln Pt,m 0.7881 0.6517 1.1040 226.3621 630.0525 766.4116 (194.7276)***
(0.1911)*** (0.3373)* (0.3324)*** (147.4855) (172.8364)***
[ ]
emp_rate_difvt 0.0350 0.0275 0.0217 8.3392 13.9172 (1.9923)*** 13.2545

(0.0044)*** (0.0039)*** (0.0038)*** (3.4433)** (2.0722)***
zvt ⋅ Ln Pt,m −0.0812 −0.0944 −0.1180 −16.8049 −45.4890 (19.6813)** −48.3861 (23.9045)**
(0.0273)*** (0.0419)** (0.0436)*** (14.1302)
[ ]
(coefficient estimated: 𝜆3)
Observations 1798 861 539 1954 898 555
Table 1 (continued)
Population (1) (2)
Sch_returnit estimated by Mincerian wage equation Sch_returnit calculated as the difference between the average salary obtained
for high- and low-skilled workers
Interactions
𝛽1 ⋅ 𝜆3 −0.1074 −0.1376 −0.2036 −0.0873 −0.1210 −0.1221
(0.0412)*** (0.0671)** (0.0836)** (0.0754) (0.0569)** (0.0635)*
𝜋1 ⋅ 𝜆 3 0.0510 0.0595 0.0753 0.0354 0.0495 0.0394
(0.0217)** (0.0311)* (0.0355)** (0.0312) (0.0269) (0.0258)
*: Statistically significant at the .10 level, **: at the 0.05 level, ***: at the 0.01 level. Equations (9.1) and (9.2) include both regional and time fixed effects in their esti-
mates. Meanwhile, (9.3) only incorporates regional fixed effects
13
On the other hand, when analyzing the impact on the returns to schooling that the
interaction between the economic shock and the relative weight of the employment
held in the metal mining sector for the first year of the sample in each county, we
observe a negative and significant effect for practically all the estimates (see the esti-
mates for Eq. (9.3)). The effect of the shock combined with being a county that pre-
viously had high exposure to it significantly reduced the skill premium for school-
ing in the labor market during the period analyzed. Likewise, when analyzing the
significance associated with the impact of the shock via returns to schooling using
the school enrollment rate (the coefficient associated with the interaction 𝛽1 ⋅ 𝜆3) and
the youth labor force participation rate (𝜋1 ⋅ 𝜆3)), the impact was negative and sig-
nificant in almost all the estimates for the coefficient associated with the interaction.
An increase of one unit in the interaction between the weight of mining employment
in the year 2000 and the metal price index (in logs) would reduce the school enroll-
ment rate in the period analyzed between −0.0873 and −0.2036 percentage points
on average.
To understand this estimated relationship, let’s propose an example (see Table 2).
Consider the Calama23 and Diego de Almagro counties, which are those that pre-
sented the greatest percentage of metal mining employment in 2000, reaching 28.2%
and 27.7%, respectively, and increases in the metal price index by a log of 2.22 and
2.46 units between 2003 and 2011 (the start and end of the shock). By multiplying
the respective values of
each county by the coefficient associated with 𝛽1 ⋅ 𝜆3 in each
of the estimates made ((1) and (2)), we observe between −5.46 and −12.73 percent-
age point decrease in the school enrollment rate for Calama, and between −5.95 and
−9.38 decrease for Diego de Almagro. These values are close to the decreases that
were actually observed: −8.46 and −8.69 percentage points, respectively; figures
that are equivalent to the reduction in the school enrollment rate between 2003 and
2011 for the 15–18-year-old age group in both counties.
The shock encouraged the youngest age group to enter the labor market, this
effect being significant only when using the Mincer salary equation to calculate
returns to schooling. This could be due to the fact that this measure is more accu-
rate measuring the value of an education in the labor market with respect to the
average salary difference obtained between high- and low-skilled workers. A 1 per-
cent increase in the weight of mining employment in 2000 and the metal price index
(in logs) would increase the labor participation rate on average between 0.0354 and
0.0753 percentage points during the period analyzed.
Likewise, when the same estimates are analyzed for the 18–24-year-old age
group (see Table 3), a behavior similar to that described for the 15–18-year-old age
group is observed, with a decrease in the school enrollment rate and a simultaneous
increase in the labor force participation rate. However, the coefficients associated
with the interactions 𝛽1 ⋅ 𝜆3 and 𝜋1 ⋅ 𝜆3 tend to be higher. If we compare the cases
of Calama and Diego de Almagro again (see Table 2), we observe that the school
enrollment rate for the 18–24-year old group in Calama decreased between −14.78
and −28.9 percentage points (Table 4), with an increase between 6.39 and 12.01
23
County where the well-known Chuquicamata mine is located.
13
Table 2 Impact estimated for the Chilean counties of Calama and Diego de Almagro, two of the top-5 counties with the highest percentage of employment corresponding
to the metal mining sector with respect to the total employment in 2000
County Employment Growth Returns to Variation in Variation in
held in metal- Ln Pt,m schooling(x) rate_school_enrvt rate_part_enrvt
[ ]in
mining sector between 2003 2003–2011 2003–2011
in 2000 (%) and 2011 (𝛽1 ⋅ 𝜆3) (𝜋1 ⋅ 𝜆3)
Estimates Table 1 Real varia- Estimates Table 2 Real variation

tion
Whole ≥ 25, 000 ≥ 50, 000 Whole ≥ 25, 000 ≥ 50, 000
sample sample
Population between 15 and 18 years old

Calama 28.2 2.218084 (1) −6.71 −8.60 −12.73 −8.46 3.19 3.72 4.71 8.52
(2) −5.46 −7.56 −7.63 2.21 3.09 2.46
Diego de 27.7 2.462636 (1) −7.32 −9.38 −5.95 −8.69 3.47 4.05 5.13 14.43
Almagro (2) −5.95 −8.25 −8.33 2.41 3.37 2.68

Calama 28.2 2.218084 (1) −16.08 −19.19 −24.90 −10.99 7.40 9.19 12.01 9.48
(2) −14.78 −17.24 −18.38 6.39 8.19 8.89
Diego de 27.7 2.462636 (1) −17.53 −20.92 −27.16 −23.42 8.07 10.02 13.10 3240
Almagro (2) −16.12 −18.80 −20.05 6.97 8.93 9.70
Source Self-elaborated with estimates from Tables 1 and 2. Both counties are selected because they present on average a population greater than 25,000 people during the
period analyzed
(x)
The method applied for calculating the returns to schooling are:
(1) schooling_returnsvt estimated by Mincerian wage equation
(2) schooling_returnsvt calculated as the difference between the average salary obtained for high- and low-skilled worker
13
Table 3 3SLS estimates for the 18–24-year-old age group, using the relative weight of employment in the metal mining sector with respect to the total employment for
each county in 2000 as the criterion to identify the counties most exposed to the shock (Álvarez et al. 2018)
Population (1) (2)
13
Sch_returnit estimated by Mincerian wage equation Sch_returnit calculated as the difference between the average salary
obtained for high- and low-skilled workers
Whole sample ≥ 25, 000 ≥ 50, 000 Whole sample ≥ 25, 000 ≥ 50, 000

(coefficient estimated: (0.5226)*** (0.6293)*** (0.9605)*** (0.0031)*** (0.0016)*** (0.0015)***
𝛽1)
𝛽0 6.98515 13.6259 (11.0009) −1.7874 −2.0293 16.2045 7.6296
(6.3069) (15.7808) (12.0476) (8.3611)* (8.8110)
(coefficient estimated: (0.2944)*** (0.3541)*** (0.5196)*** (0.0014)*** (0.0008)*** (0.0008)**
𝜋1)
𝜋0 55.9836 51.5780 58.3470 59.9418 50.9421 54.2521
(3.6764)*** (6.3972)*** (8.6463)*** (5.5191)*** (4.8661)*** (4.9414)***
zvt(using Álvarez et al. 0.3570 0.4418 0.4150 83.1168 236.2701 225.5322 (83.9456)***
(2018) definition) (0.0923)*** (0.1537)*** (0.1664)** (41.5019)** (70.9969)***
Ln Pt,m 0.6334 0.5366 1.0198 196.6434 460.7054 682.7903
(0.1695)*** (0.3037)* (0.3088)*** (127.7963) (148.6473)*** (180.5260)***
[ ]
emp_rate_difvt 0.0382 0.0298 0.0224 7.0001 16.0887 (1.7912)*** 14.9641

(0.0043)*** (0.0038)*** (0.0038)*** (2.9591)** (1.6300)***
zvt ⋅ Ln Pt,m −0.0611 −0.0660 −0.0676 −12.3330 −34.7915 −37.1090 (14.5598)**
(0.0279)** (0.0288)** (7.0651)* (12.2088)***
[ ]
(coefficient estimated: (0.0167)***
𝜆3)
Observations 1798 861 539 1954 898 555
Table 3 (continued)
Population (1) (2)
Sch_returnit estimated by Mincerian wage equation Sch_returnit calculated as the difference between the average salary
Interactions
𝛽1 ⋅ 𝜆3 −0.2571 −0.3068 −0.3982 −0.2364 −0.2757 −0.2940
(0.0800)*** (0.1411)** (0.1754)** (0.1501) (0.0942)*** (0.1058)***
𝜋1 ⋅ 𝜆 3 0.1184 0.1470 0.1921 0.1023 0.1310 0.1422
(0.0382)*** (0.0688)** (0.0860)** (0.0653) (0.0467)*** (0.0534)***
Source Self-elaborated using CASEN database for the period 2000, 2003, 2006, 2009, 2011, 2013 and 2015. *: Statistically significant at the .10 level, **: at the 0.05 level,
***: at the 0.01 level. Equations (9.1) and (9.2) include both regional and time fixed effects in their estimates. Meanwhile, (9.3) only incorporates regional fixed effects
13
percentage points in the labor force participation rate. In Diego de Almagro, these
figures are −16.12 and −27.16 percentage points for the school enrollment rate and
6.90 and 13.10 percentage points in the labor force participation rate.
These first estimates show that the shock caused a reduction in the skill premium
in the labor market, discouraging young people from continuing their education in
order to participate in the job market. In order to contrast the robustness, the AMC
criterion for classifying counties as metal mining or non-metal mining producers is
used. The results show similar results for the two age groups analyzed. However,
these tend to be mostly significant when the Mincer salary equation is used to define
the returns to schooling (schooling_returnsvt ), since it is better suited for measuring
this concept.
We observe for the 15–18-year-old age group that metal mining producer coun-
ties (under the AMC criterion) experienced a decrease in the school enrollment rate
of between −0.9122 and −4.276 percentage points for each one percentage point
increase of Pt,m. The labor force participation rate for this age group increased
between 0.1358 and 1.2549 percentage points for said increase of one percentage
point of Pt,m. Meanwhile (Table 5), for the 18–24-year-old age group we observe
that their estimates show a greater effect, with the impact of the shock on the
school enrollment rate for metal mining producer counties between −2.2031 and
−7.4714 percentage points for each percentage point increase in Pt,m. Also, this
shock produced a simultaneous increase in the labor force participation rate for
the 18–24-year-old age group of between 1.0970 and 3.9731 percentage points for
said increase of one percentage point in Pt,m (see Table 6 for the predicted values of
Diego de Almagro and Calama).
In summary, the results presented empirically validate the assumption that the
metal mining price boom had a negative impact on returns to schooling which
encouraged youth to join the labor market in those counties where metal mining
activity is present.24
6 Conclusion
We analyze the impact that exogenous economic shocks have on both the school
enrollment and labor participation rates of youth, particularly when the shock
(while it lasts) alters the skill composition required by the labor market. We
considered in our analysis the Chilean case during the metal mining price boom
between 2003 and 2011. The immediate effect that the shock had on the Chilean
economy was an increase in employment opportunities in not only the mining
sector, but also other sectors which directly (construction) or indirectly (trade or
24
Due to the importance of the copper mining industry in the Chilean economy we replicate our empiri-
cal analysis using the
[ world
] price of copper as a measure for the intensity of the shock, thus creating a
new measure for Ln Pt,j in the simultaneous equations model estimated. The main results remain robust,
but in some cases, we find a loss of significance, which can be explained by the lower variation of the
copper price. All these additional estimates are available upon request.
13
Table 4 3SLS estimates for the 15–18-year-old age group, using the AMC criterion to identify counties that are metal mining producers
Population (1) (2)
Sch_returnvt estimated by Mincerian wage equation Sch_returnvt calculated as the difference between the average salary

𝛽1)
𝛽0 68.4995 67.3705 (3.6650)*** 66.4836 72.7482 72.3312 73.9847
(2.9507)*** (3.7704)*** (2.5866)*** (2.5952)*** (2.4257)***
(coefficient estimated: (0.1630)*** (0.1414)*** (0.1462)*** (0.0003)*** (0.0002)** (0.0002)
𝜋1)
𝜋0 14.9809 12.5383 12.0345 13.1921 10.7159 9.2093
(2.0325)*** (2.3485)*** (2.3253)*** (1.5478)*** (1.8188)*** (1.6769)***

zvt(using AMC defini- 8.5557 13.0406 15.3910 2,701.1940 4,078.653 5,064.501
tion) (1.9827)*** (3.0555)*** (3.4853)*** (1,271.252)** (1,860.564)** (2,523.198)**
Ln Pt,m 1.0528 1.3106 1.5236 285.4142 631.8754 666.4089
(0.1809)*** (0.2784)*** (0.2640)*** (144.0574)** (163.0632)*** (179.6800)***
[ ]
emp_rate_difvt 0.0199 0.0179 0.0158 10.3021 (2.1872)*** 12.2355 (1.2730)*** 12.5078

(0.0026)*** (0.0023)*** (0.0021)*** (1.3759)***
zvt ⋅ Ln Pt,m −1.5485 −2.3821 −2.8225 −470.5089 −499.3041 −485.1127 (486.4499)
(0.5936)*** (0.6687)*** (239.5335)** (361.3596)
[ ]
𝜆3)
Observations 1996 894 565 2178 934 583
13
Table 4 (continued)
Population (1) (2)
13
Interactions
𝛽1 ⋅ 𝜆3 −2.2989 −3.4264 −4.2760 −1.3312 −0.9122 −0.7459
(0.6866)*** (1.0103)*** (1.1859)*** (0.6865)* (0.6839) (0.7667)
𝜋1 ⋅ 𝜆 3 0.9099 1.0345 1.2549 0.5260 0.2498 0.1358
(0.3405)*** (0.4244)** (0.5018)** (0.2997)* (0.2125) (0.1670)
Table 5 3SLS estimates for individuals of the 18–24-year-old age group, using the AMC criterion to identify the counties that are metal mining producers
Population (1) (2)

𝛽1)
𝛽0 −1.1185 −1.6180 (8.3412) −8.1144 15.5230 (6.0709)** 11.6012 13.7303
(6.6898) (9.1672) (5.3763)** (5.2287)***
𝜋1)
𝜋0 60.1009 59.4363 63.2348 52.6340 53.4469 52.2962
(3.9022)*** (4.9744)*** (5.6958)*** (3.3722)*** (3.2737)*** (3.2922)***

zvt(using AMC defini- 5.8361 8.4459 8.6336 1432.276 3388.412 3169.551 (1,881.568)*
tion) (1.2493)*** (2.1856)*** (2.5472)*** (772.0053)* (1,375.373)**
Ln Pt,m 0.8664 1.1652 1.4084 125.4254 574.2298 609.8093
(0.1663)*** (0.2710)*** (0.2603)*** (115.7296) (159.5340)*** (177.6209)***
[ ]
emp_rate_difvt 0.0195 0.0186 0.0162 10.5049 13.7009 (1.1949)*** 14.4180

(0.0026)*** (0.0022)*** (0.0021)*** (2.1753)*** (1.2643)***
zvt ⋅ Ln Pt,m −1.0616 −1.5697 −1.5512 −242.7552 −497.5292 −302.7946 (362.5350)
(0.4246)*** (0.4834)*** (141.0171)* (267.0057)*
[ ]
𝜆3)
Observations 1,996 894 565 2,178 934 583
13
Table 5 (continued)
Population (1) (2)
13
Interactions
𝛽1 ⋅ 𝜆3 −5.0264 −6.9327 −7.4715 −2.2031 −2.8905 −1.5327
(1.3330)*** (2.0610)*** (2.4178)*** (1.1673)* (1.6098)* (1.8617)
𝜋1 ⋅ 𝜆 3 2.4198 3.3780 3.9731 1.0970 1.4248 0.8542
(0.6683)*** (1.0407)*** (1.3157)*** (0.5875)* (0.8012)* (1.0383)
Table 6 Impact estimated for the Chilean counties of Calama and Diego de Almagro, two of the top-5 counties that are metal mining producers using the AMC criterion
County Growth in Ln Pt,m Returns to Variation in rate_school_enrvt Variation in rate_part_enrvt
schooling(x) 2003–2011 2003–2011
[ ]
between 2003 and
2011 (𝛽1 ⋅ 𝜆3) (𝜋1 ⋅ 𝜆3)
Estimates Table 1 Real variation Estimates Table 2 Real variation


Calama 2.218084 (1) −6.63 −9.88 −12.33 −8.46 2.01 2.29 2.78 8.52
(2) −3.83 −2.63 −2.15 1.16 0.55 0.30
Diego de Almagro 2.462636 (1) −5.66 −8.43 −10.53 −8.69 2.24 2.54 3.09 14.43
(2) −3.27 −2.24 −1.83 1.29 0.61 0.33
Calama 2.218084 (1) −11.14 −15.37 −16.57 −10.99 5.36 7.49 8.81 9.48
(2) −4.88 −6.41 −3.39 2.43 3.16 1.89
Diego de Almagro 2.462636 (1) −12.37 −17.07 −18.40 −23.42 5.96 8.31 9.78 3240
(2) −5.42 −7.11 −3.77 2.70 3.50 2.10
Source Self-elaborated with estimates from Tables 4 and 5. Both counties are selected because they present on average a population greater than 25,000 people during the
period analyzed
(x)
The method applied for calculating the returns to schooling are:
(1) schooling_returnsvt estimated by Mincerian wage equation.
(2) schooling_returnsvt calculated as the difference between the average salary obtained for high- and low-skilled worker.
13
consumer services) depended upon it, especially in parts of the country with high
concentrations of mining. The exogenous shock not only boosted activity requir-
ing a high- to medium-skilled workforce, but it also favored job creation in the
low- or medium–low-skilled segment via a production-intensive workforce. Dif-
ferent authors (Pellandra 2014; Rehner and Vergara 2014) identify the later seg-
ment as having the most growth in employment during the shock.
We assess whether the effect of the shock on the labor market discouraged
school enrollment rates among youth while simultaneously increasing their par-
ticipation in the labor market. For this, a theoretical model was developed based
on the classical theory of human capital which is subsequently empirically evalu-
ated. The results show that the exogenous shock had a negative impact on returns
to schooling in primarily metal mining producer counties. These counties present
simultaneously both a negative impact on school enrollment rates and an increase
in youth labor market participation. Additionally, this behavior is observed
in those individuals who should legally be part of the educational system (the
15–18-year-old age group), introducing a risk for the economy primarily because
of a negative effect on human capital accumulation, coupled with the decreasing
employability of said youth in future. Something that may have boosted the level
of unemployment in the counties highly exposed to the shock once this finished.
Appendix 1
See Figs. 2, 3 and Tables 7, 8, 9.

3.5
National metal mining employment share
300
Metals Price Index (2003 = 100)
3
250
2.5
200
2
150
1.5
100
1986 1990 1994 1998 2002 2006 2010 2014

Year
National metal mining emp. share Metals Price Index
Fig. 2 Evolution for national average metal mining employment share between 1986 and 2015, and for
national metal mining price index between 1999 and 2015. Source Self-elaborated using data from the
National Labor Survey (ENE) and COCHILCO
13
Fig. 3 Chilean maps to identify the location of the metal mining producer counties, using both defi-
nitions used in the analysis. Figure 1 Relative weight of employment in the metal mining sector with
respect to the total employment (Álvarez et al. 2018, Fig. 2) AMC definition. Source: Self-elaborated
using CASEN database. Note Figure 1 shows the average county employment share for the year 2000.
“No data” registers the counties not originally considered in the sample provided by the CASEN
Table 7 Mining counties

considered in the analysis using Pozo Almonte Diego de Almagro Los Andes Camarones
the AMC classification criterion. Pica Vallenar Cabildo Putre
Source: Self-elaborated Antofagasta La Serena Petorca Chile Chico
Mejillones Coquimbo La Calera Huasco
Sierra Gorda Andacollo Nogales La Ligua
Taltal La Higuera Machalí Los Vilos
Calama Vicuña Requínoa Pencahue
Tocopilla Illapel Coihaique Rancagua
María Elena Salamanca Río Ibáñez Río Hurtado
Copiapó Ovalle Lo Barnechea San Esteban
Tierra Amarilla Monte Patria Alhué
Chañaral Punitaqui Arica
13
Table 8 Mean and standard deviation (in brackets) for the variables used in empirical analysis. Source
Self-elaborated using CASEN database for the period 2000, 2003, 2006, 2009, 2011, 2013 and 2015
Population Whole sample (≥ 25, 000) (≥ 50, 000)
rate_school_enrvt(15–18 years old) 83.13 83.83 84.76

(9.24) (8.04) (7.45)
rate_school_enrvt(18–24 years old) 35.95 40.52 43.82
(13.42) (13.57) (13.61)
schooling_returnsvt(Mincerian wage equation) 7.53 8.83 9.70
(6.13) (5.49) (5.21)
schooling_returnsvt(difference between the average 3414.66 3530.55 3783.69
salary obtained for high- and low-skilled workers) (5209.08) (3243.86) (3580.51)
rate_part_empvt(15–18 years old) 10.93 10.73 10.41
(7.44) (6.17) (5.69)
rate_part_empvt(18–24 years old) 48.08 47.84 47.69
(10.63) (9.67) (9.91)
zvt(using Álvarez et al., (2018) definition) 1.55 1.46 1.45
(4.36) (4.40) (3.99)
zvt(using AMC definition) 317a 134a 82a
(14.52%)b (14.52%)b (14.07%)b
[ ]
Ln Pt,m 4.95 4.95 4.97
(0.45) (0.40) (0.40)
emp_rate_difvt 19.89 32.83 44.11
(51.47) (76.36) (94.55)
a
Total number of cases
b
Relative frequency
Table 9 Yearly regional average of schooling returns estimated in Mincerian wage equation Source Self-
elaborated
Region CASEN
2000 2003 2006 2009 2011 2013 2015
Tarapacá 13.13 4.95 6.14 9.05 10.81 5.79 11.19

Antofagasta 10.98 7.90 7.69 6.81 7.29 9.43 14.19
Atacama 5.54 8.42 8.78 5.06 6.57 8.51 6.18
Coquimbo 6.83 8.86 5.30 4.89 6.93 4.77 7.66
Valparaíso 8.21 7.46 5.57 5.75 7.83 6.48 9.10
O’Higgins 5.72 5.72 5.12 4.62 3.43 4.16 7.67
El Maule 8.15 5.45 5.21 5.98 3.55 5.29 7.84
El Biobío 9.35 7.95 8.27 7.44 7.06 4.64 11.45
La Araucanía 8.79 9.71 9.15 5.69 7.64 8.14 9.94
Los Lagos 8.78 6.95 7.48 9.30 10.39 9.17 8.29
Aysén 11.82 13.76 6.84 5.53 10.28 7.89 10.34
Magallanes 9.27 9.96 9.83 13.01 12.67 8.87 9.60
Metropolitana 10.24 9.80 8.47 7.61 8.29 8.34 8.67
Los Ríos 8.82 11.94 8.51 9.67 7.39 6.89 10.16
Arica y Parinacota 7.78 8.19 6.06 3.27 14.38 5.29 6.24
13
Supplementary Information The online version contains supplementary material available at https://doi.
org/10.1007/s00168-021-01069-8.
References
Álvarez R, García-Marín A, Ilabaca S (2018) Commodity price shocks and poverty reduction in Chile.
Resour Policy. https://doi.org/10.1016/j.resour pol.2018.04.004
Angrist J, Krueger A (2001) Instrumental variables and the search for identification: from supply and
demand to natural experiments. J Econ Perspect 15(4):69–85
Aragón F, Rud J (2012) Mining, pollution and agricultural productivity: evidence from Ghana. Econ
J. https://doi.org/10.1111/ecoj.12244
Aroca P, Atienza M (2008) La conmutación regional en Chile y su impacto en la Región de Antofagasta.
Revista Latinoamericana De Estudios Urbanos Regionales 34(102):97–121
Becker G (1964) Human Capital. Columbia University Press for the National Bureau of Economic
Research, New York
Black DA, McKinnish T, Sanders S (2005) Tight labor markets and the demand for education: evidence
from the coal boom and bust. Ind Labor Relat Rev 59(1):3–16
Calero C, Bedi A, Sparrow R (2009) Remittances, liquidity constraints and human capital investments in
Ecuador. World Dev 37(6):1143–1154
Card D (2001) Immigrant inflows, native outflows, and the local market impacts of higher immigration. J
Law Econ 19(1):22–64
Carneiro P, Heckman J (2002) The evidence on credit constraints in post-secondary schooling. Econ J
112:989–1018
Ferreira F, Schady N (2009) Aggregate economic shocks, child schooling, and child health. World Bank
Res Obs 24(2):147–181
Funkhouser E (1999) Cyclical economic conditions and school attendance in Costa Rica. Econ Educ Rev
18(1):31–50
Goldin C (1999) Egalitarianism and the returns to education during the great transformation of American
Education. J Polit Econ 107(6):65–94
Greene W (2012) Econometric Analysis, 7th edn. Pearson, New York
Loayza N, Rigolini J (2016) The local impact of mining on poverty and inequality: evidence from the
commodity boom in Peru. World Dev 84:219–234
Mckenzie D (2003) How do households cope with aggregate shocks? evidence from the Mexican Peso
Crisis. World Dev 31(7):1179–1199
Mincer J (1974) Schooling, Experience, and Earnings. National Bureau of Economic Research, New York
Modrego F, Berdegué J (2015) A large-scale mapping of territorial development dynamics in Latin
America. World Dev 73:11–31
Observatorio Laboral para la Región de Antofagasta (OLAB) (2016) Panorama Regional Antofagasta,
OLAB, Servicio Nacional de Capacitación y Empleo (Chile)
Pellandra A (2015) The commodity price boom and regional workers in Chile: a natural resources bless-
ing? Working Paper Lacerlacea, https://pdfs.semanticscholar.org/5ecf/4f0764da5971bbccc66bb1a8e
d8731055fa6.pdf
Petrongolo B (2001) Pissarides C (2001) Looking into the black box: a survey on the matching function.
J Econ Lit 39:390–431
Radetzki M, Eggert R, Lagos G, Lima M, Tilton J (2008) The boom in mineral markets: how long might
it last? Resour Policy 33(3):125–128
Rehner J, Vergara F (2014) Efectos recientes de la actividad exportadora sobre la reestructuración
económica urbana en Chile. Revista De Geografía Norte Grande 59:83–103
Schady N (2004) Do macroeconomic crises always slow human capital accumulation? World Bank Econ
Rev 18(2):131–154
Singleton K (2013) Investor flows and the 2008 boom/bust in oil prices. Manage Sci 60(2):300–318
Willis RJ, Rosen S. (1979) Education and Self-selection. Journal of Political Economy. Special Number
Part 2: Education and Income Distribution. 87(5): s7-s36
Zarkin GA (1983) Cobweb vs rational expectations models. Econ Lett 13:87–95
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published
maps and institutional affiliations.
13

Economic Shocks and Their Effect On The Schooling and Labor Participation of Youth: Evidence From The Metal Mining Price Boom in Chilean Counties

Uploaded by

Copyright:

Available Formats

Economic Shocks and Their Effect On The Schooling and Labor Participation of Youth: Evidence From The Metal Mining Price Boom in Chilean Counties

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Economic Shocks and Their Effect On The Schooling and Labor Participation of Youth: Evidence From The Metal Mining Price Boom in Chilean Counties

Uploaded by

Copyright:

Available Formats

The Annals of Regional Science

Economic shocks and their effect on the schooling

Manuel Pérez Trujillo1 · Gabriel O. Rodríguez Puello2

Received: 29 July 2020 / Accepted: 11 June 2021

JEL Classification I26 · J24 · R11 · R23

We aim to analyze how an exogenous economic shock, which presents a differenti-

* Manuel Pérez Trujillo

contributed to a larger reduction in skill premiums4 in the most affected provinces

It is necessary to make use of the appropriate theoretical framework in order to illus-

be understood as the complement of the probability of young people remaining in

3 Data and previous descriptive evidence

We will make use of the Chilean National Socioeconomic Characterization Survey

where l represents each of the five metals considered, 𝜑2000 l,m

rate_part_empvt = 𝜋0 + 𝜔m + 𝜏t + 𝜋1 ⋅ schooling_returnsvt + vvt (9.2)

where the variable rate_school_enrit , corresponds to the school

Estimates of Eq. (9.1): rate_school_enrvt

emp_rate_difvt 0.0350 0.0275 0.0217 8.3392 13.9172 (1.9923)*** 13.2545

Estimates Table 1 Real varia- Estimates Table 2 Real variation

Population between 15 and 18 years old

Almagro (2) −5.95 −8.25 −8.33 2.41 3.37 2.68

Estimates of Eq. (9.1): rate_school_enrvt

emp_rate_difvt 0.0382 0.0298 0.0224 7.0001 16.0887 (1.7912)*** 14.9641

Estimates of Eq. (9.1): rate_school_enrvt

(2.0325)*** (2.3485)*** (2.3253)*** (1.5478)*** (1.8188)*** (1.6769)***

emp_rate_difvt 0.0199 0.0179 0.0158 10.3021 (2.1872)*** 12.2355 (1.2730)*** 12.5078

Estimates of Eq. (9.1): rate_school_enrvt

(3.9022)*** (4.9744)*** (5.6958)*** (3.3722)*** (3.2737)*** (3.2922)***

emp_rate_difvt 0.0195 0.0186 0.0162 10.5049 13.7009 (1.1949)*** 14.4180

Estimates Table 1 Real variation Estimates Table 2 Real variation

Population between 15 and 18 years old

See Figs. 2, 3 and Tables 7, 8, 9.

1986 1990 1994 1998 2002 2006 2010 2014

National metal mining emp. share Metals Price Index

Table 7 Mining counties

rate_school_enrvt(15–18 years old) 83.13 83.83 84.76

Tarapacá 13.13 4.95 6.14 9.05 10.81 5.79 11.19

You might also like

3 Data and previous descriptive evidence

(2.0325)* (2.3485)* (2.3253)* (1.5478)* (1.8188)* (1.6769)*

emp_rate_difvt 0.0199 0.0179 0.0158 10.3021 (2.1872)* 12.2355 (1.2730)* 12.5078

(3.9022)* (4.9744)* (5.6958)* (3.3722)* (3.2737)* (3.2922)*