Highways, Shocks and Labor Market Outcomes Anita Bhide and Yiming He * September 2016 Abstract Recently developing countries have seen huge increases in spending on transporta- tion networks. Given these large investments it’s important to understand the welfare consequences. The paper will look at if access to these roads has improved agricultural household’s abilities to better cope with local productivity shocks. It looks particularly at the Golden Quadrilateral Highway system in India. Proximity to this highway allows households easier access to labor markets that are uncorrelated with local market con- ditions. We build a theoretical model with precise empirical predictions; household’s equilibrium employment in agricultural and non-agriculture should respond more to shocks when located closer to roads. The changes in wages on the other hand will be smaller. These empirical predictions are tested using the REDS dataset: a nationally representative household survey of rural India. Using a difference-in-difference strategy, we look at how responses to rainfall shocks varies as distance to the GQ increases. * We thank Melanie Morten, Pascaline Dupas, Arun Chandrasekhar, Marcel Fafchamps and participants of the Development Tea for for useful comments and feedback 1
39
Embed
Highways, Shocks and Labor Market Outcomesstanford.edu/~abhide2/BhideHeHighwaysLabor.pdf · at the Golden Quadrilateral Highway system in India. Proximity to this highway allows households
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Highways, Shocks and Labor Market Outcomes
Anita Bhide and Yiming He∗
September 2016
Abstract
Recently developing countries have seen huge increases in spending on transporta-tion networks. Given these large investments it’s important to understand the welfareconsequences. The paper will look at if access to these roads has improved agriculturalhousehold’s abilities to better cope with local productivity shocks. It looks particularlyat the Golden Quadrilateral Highway system in India. Proximity to this highway allowshouseholds easier access to labor markets that are uncorrelated with local market con-ditions. We build a theoretical model with precise empirical predictions; household’sequilibrium employment in agricultural and non-agriculture should respond more toshocks when located closer to roads. The changes in wages on the other hand will besmaller. These empirical predictions are tested using the REDS dataset: a nationallyrepresentative household survey of rural India. Using a difference-in-difference strategy,we look at how responses to rainfall shocks varies as distance to the GQ increases.
∗We thank Melanie Morten, Pascaline Dupas, Arun Chandrasekhar, Marcel Fafchamps and participantsof the Development Tea for for useful comments and feedback
1
As has been noted in low-income developing countries like India, while markets for agri-
cultural inputs and outputs are well developed, that of credit and insurance markets lag
behind. An interesting question is then how poor farmers mitigate the impact of adverse
shocks. A large literature has looked at different risk insurance strategies rural households in
developing countries employ; village level risk insurance, sub-caste level insurance and saving
mechanisms. Townsend (1994)’s seminal work looked at village level risk insurance, where
idiosyncratic risk may be insured using transfers within a village. The evidence however did
not support this picture. Others have looked at several broader groupings of individuals:
Munshi and Rosenzweig (2016) look at sub-castes (jati’s) while Rosenzweig and Stark (1989)
analyze networks formed through marriages.
In this paper we look at the use of off-farm labor markets which may afford households
an opportunity to self-insure/diversify individually. The literature so far has ascertained the
role of off-farm labor supply; Kochar (1999), Rose (2001), Cameron and Worswick (2003)
and Fernandez et al. (2014) all provide evidence in favor of the use of off-farm labor markets
as means of mitigating the impact of negative income shocks.1
In spite of this there still remain more questions. To begin with how do labor markets
even play a significant role in providing diversification to farmers? Agricultural labor mar-
kets are highly correlated with rural households’ own productivity shocks (particularly in
the case of rainfall shocks). On the other hand rural non-agricultural labor markets through
the tradable sector, such as manufacturing, or urban labor markets are likely unaffected by
rainfall shocks,d d therefore would provide suitable opportunity to diversify risk for house-
holds. Further what role does access to transportation networks play? Accessing these labor
markets is highly related to the location of households. Those closer to urban centers and
with access to transportation networks (roads, bus routes and railway lines) are likely to have
the opportunity to turn to non-agricultural labor markets to diversify and smooth consump-
tion. Asher and Novosad (2016) highlight how those closer to roads are more likely to be
engaged in non-agricultural work. Is it then the case that those with better access to labor
markets and urban centers are also better able to diversify income shocks by easily accessing
labor markets that are not correlated with the agricultural shocks? As road networks have
expanded and will continue to do so, this paper will shed light on the impact on the rural
1Kochar (1999) looks at shocks to crops, Rose (2001) at rainfall shocks, Cameron and Worswick (2003)at idiosyncratic shocks such as sudden deaths and Fernandez et al. (2014) at violent conflict shocks.
2
sector.
We start with providing a theoretical framework with testable empirical predictions for
equilibrium wages and employments. The predictions differ based on each location’s road
access. We build a two-sector economy general equilibrium model with one aggregate utility-
maximizing household. The model builds heavily on Santangelo (2016). In our model the two
sectors, agriculture and manufacturing,2 hire workers in the same labor market. But different
from Santangelo (2016) working in the manufacturing sector incurs additional commuting
cost β, which is decreasing with better road access. We solve the equilibrium wages and
quantities and derive several testable predictions. First, for agricultural labor markets, a
positive rainfall shock increases both labor and wage at equilibrium. The positive labor
response is amplified and the positive wage response is mitigated when the commuting cost
is lower. Second for manufacturing labor markets, a positive rainfall shock decreases labor
and increases wage at equilibrium. The negative labor response is amplified and the positive
wage response is mitigated when the commuting cost is lower.3
Our paper looks at how equilibrium labor market employments and wages are differen-
tially affected by productivity shocks, based on their access to road networks in the setting
of India. The productivity shocks we look at are monsoon rainfall shocks; more specifically
standardized deviations from the historical mean. The access to road networks is based on
distance from the Golden Quadrilateral (GQ). The GQ is a system of roads linking four
major cities of Mumbai, Chennai, Kolkata and Delhi in India. Figure 3 in the appendix
displays the route of GQ. The GQ project is a major upgrading project of 5846 km of ex-
isting roads to four- and six-lane roads.4 It is part of the National Highways Development
Project (NHDP) initiated by the Indian government in 2001, with the goal of improving the
overall performance of its national highway system. GQ was targeted to be completed by
year 2004. In reality, by 2004 80 % of the work was finished, and 95 % of the work was
finished by the end of 2006 (Ghani et al., 2016). A similar highway project planned during
the same period, the North-South and East-West Corridor, experienced significant delays
and only less than 10 % of the roads were completed by 2006. Therefore in our empirical
2In this paper the words manufacturing and non-agriculture are interchangeable. We are aware of thefact that the non-tradable wage jobs are important sources of income for rural households, and we plan onincorporating this labor market into our model in the future.
3Here we model the manufacturing wage to be the wage that factories need to pay workers. At equilibriumthe effective wages are the same between agriculture and manufacturing sectors.
4The total cost of the project incurred at the end of year 2013 was 5.6 billion dollars
3
setting (we use household-level and district-level data in years 1998 and 2005), we only use
GQ to construct area-specific road access variable. We consider 1998 as the pre-GQ period
and 2005 as the post-GQ period.
As section 2 will show, our estimation strategy compares labor outcomes in both agri-
cultural and non-agricultural markets to shocks in areas closer to the GQ before and after
the upgrading (1998 to 2005), to those areas far away. The identification strategy exploits
the fact that (1) Monsoon shocks are exogenous and (2) Absent the Golden Quadrilateral
project, labor responses in areas near the GQ would have followed the same linear trajectory
as those far from the GQ. Our ultimate aim is to analyze whether this leads to welfare gains
for those closer to transportation networks as they are better able to mitigate the impact of
productivity shocks by relying on non-agricultural labor markets, taking into account differ-
ential changes in wages. We use several datasets. Our data on household labor allocations
is obtained from the Rural Economic Development Survey (REDS), which provide details
on household members’ primary and secondary labor market activities, particularly in agri-
cultural and non-agricultural wage work. It also provides data on household’s demographic
variables, member characteristics and village level data. We obtain data on the Golden
Quadrilateral by georeferencing the raster map file obtained from the National Highway
Authority of India in ArcGIS. Rainfall data is obtained from the Terrestrial Air Tempera-
ture and Precipitation dataset prepared by the Center for Climate Research, University of
Delaware. Finally wage data is also obtained from both the ICRISAT Village Dynamics in
South Asia Macro-Meso Database and the Annual Survey of Industries of India.
The theoretical predictions hold for primary agricultural activity: a 1 unit increase in
standardized deviation in monsoon rainfall results in a 0.090 higher likelihood of having at
least 1 member primarily in agricultural activity in the year. It will also be associated with
an average increase in the number of household members involved in agricultural wage work.
More importantly being located close to the GQ post upgrade leads to an even greater re-
sponse of agricultural work. This provides suggestive evidence in favor of a lower commuting
cost and thus better diversification. Unfortunately secondary agricultural activity responds
in the opposite direction. Non-agricultural results are supported by the secondary activity
employment. Unfortunately the differential effect for primary work attenuates the impact of
The eventual aim of this paper is to link improvements in transport infrastructure to welfare
gains due to increased employment opportunities. To do so we have to look at both wage
changes to ascertain income changes, and look at changes in consumption patterns. For
wages we turn to two sources of data. Daily agricultural wages are derived from the ICRISAT
Village Dynamics in South Asia Macro-Meso Database which is obtained from various official
government data sources. The database covers 308 districts from 1966 to 2009 using the 1966
district boundaries ensure consistency over time. Daily factory wages are obtained from the
Annual Survey of Industry (ASI). The ASI surveys manufacturing factories across the entire
India. It includes all factories employing 10 or more workers using power and all factories
employing 20 or more workers. It also surveys a random third of unregistered factories.
3.5 Descriptive Statistics
Summary statistics for the main variables used are reported in Tables 5 and 6, for both 1998
and 2005 as well as those villages close and far. It is clear that in primary activity a greater
proportion of family members work in agriculture in 1998; While a quarter of families have
at least one member whose primary activity is agricultural wage work, only a fifth have it
at least one member primarily in non-agricultural wage work. This difference does even out
in 2005.
An ideal test of the identifying assumption would be a placebo test where we would run
equation ??, but look at years prior to the GQ project. The REDS dataset does include
previous year’s data: 1969-1971 and 1981.12 One concern is that there is a 17 year gap
between 1981 and 1999, and it would be hard to ascertain the parallel trends assumption.
However we feel that the similarity in primary labor market activity in 1998 for those close
and far is suggestive that far away villages are a good control group. Looking at column
(3) of Table 5, for the primary activity status the differences between those close to the
Golden Quadrilateral and those far is negligible in 1998 (prior to the GQ project) 13. This
suggests that the labor market patterns of those who are close to the GQ and those far were
not significantly different. Secondary activity status does vary significantly in the baseline
12We are currently working on compiling this to run the placebo test13The total number who are primarily in agriculture is significantly higher for those further than those
closer
20
year. Though this does not invalidate our identification strategy, we feel more confident in
primary activity status.
There are some changes over time. Primary agricultural wage work sees a decline from
1998 to 2005; the proportion of families with at least one member declines from 0.25 to 0.15
for those close and 0.26 to 0.18 for those far. There is also an increase in Secondary Activity
Wage work in both agriculture and non-agriculture.
Table 6 on the other hand depicts the descriptive statistics for the key control variables.
Looking at column (3) it is clear that there are significant differences in households close
and far from the GQ, which motivates our use of controls in all regressions. A reassuring
fact is that the distance to the GQ does not change too much over the years. Highlighting
that the GQ was not started in 1999, what this implies is the composition of participants in
the years has not changed significantly.
3.6 Number of factories
This paper assumes that being closer to the GQ reduces commuting costs when traveling to
local factories for work. Villages closer to the GQ will see a rise in the number of factories
both in their village and nearby, thus reducing the commuting costs compared to villages
further from the GQ. Table 7 provides evidence in favor of this. Columns 1 and 2 indicate
that after the GQ was completed, villages closer to the GQ had more factories near the
village and in total than those further away. This also matches Ghani et al. (2016) which
finds that the GQ led to a movement of indian manufacturing from urban to rural districts.
4 Results
4.1 Employment Outcomes
Table 8 in Appendix C shows the preliminary regression results using having at least 1
member whose primary activity is agriculture, non-agriculture and farm labor as outcome
variables. It is clear that monsoon has a strong impact on casual labor allocations. While
a 1 unit increase in a standardized deviation in monsoon rainfall leads to a 0.090 increase
in the likelihood of having at least 1 family member in primary agricultural wage work it
21
leads to a 0.059 decrease in the likelihood for non-agricultural work. This is the prediction
obtained from the theoretical framework. A good rainfall shock increases agricultural pro-
ductivity and thus labor demand for agricultural work. Our theoretical framework predicts
that equilibrium employment increases. On the other hand a positive rainfall shock results
in reduced labor supply for non-agricultural work through substitution towards agricultural
work. As we assume non-agricultural demand does not change, we have that equilibrium
employment reduces in the non-agricultural labor market.
Unfortunately this is not universal result. For villages close to the GQ the overall response
in 1999 is in-fact negative. This is evident by looking at the coefficient on NearGQ·Monsoon
which when added to that on Monsoon leads to a negative effect. This does not necessarily
invalidate our empirical strategy, it does contradict our theoretical prediction that the impact
of monsoon on agricultural labor is positive. This does not however occur for non-agricultural
work. Here the overall effect on non-agricultural work is negative for all villages, and is not
significantly different.
Next we turn to the key parameter for this paper: the coefficient on the triple interaction
term (β7). This captures the differential impact on households in villages close to the GQ.
Being close to the GQ increases the equilibrium response to productivity shocks for agricul-
tural work. As can be seen in column (1) of Table 8 being close to a GQ is associated with
a significantly positive impact on the likelihood of having at least 1 member in agricultural
wage work. There is an implied increase of 0.15 in the likelihood. This is in accordance with
the theoretical framework presented in section 1 which predicts that β7 should be positive.
When looking at non-agricultural wage work, being close to the GQ paradoxically mitigates
the effect of a rainfall shock, as can be seen by the positive coefficient on the triple interac-
tion in the second column of Table 8 which implies a smaller reduction in wage work closer
to the GQ. This coefficient is both positive and statistically insignificant, contradicting our
theoretical prediction.
The results are similar when using the number of individuals in each labor market activity
per household as outcome variable, depicted in Table 9. The impact of a monsoon shock
leads to an increase in the number of agricultural wage work members but a decrease in
the number of non-agricultural work members. When looking at the triple-interaction term
access to GQ amplifies the impact of a positive rainfall shock, as the coefficient in column
(1) is positive, though not significant. However once again for non-agricultural wage work
22
villages close to the GQ decrease less as the coefficient before the triple interaction term in
column (2) is positive, though not significant, which contradict our theoretical predictions.
The coefficients on the controls are largely as expected: having more males and females
between the ages of 15 and 45 increases participation in casual work and on farm work.
Obtaining education leads to reduced casual and on-farm work. More land, and in particular
irrigated land leads to greater on farm work and less casual work.
Tables 10 and 11 depict the results for secondary labor market activities. Again we look
at wage work as this refers to casual work that can respond in the short run to shocks.
The results largely contradict those found in primary agricultural labor markets. Looking
at column (1) of both tables the result is that after a rainfall shock equilibrium agricultural
wage labor falls, contradictory to both our theoretical and primary activity analysis. The
effect is however amplified by access to roads, as the coefficients before the triple-interaction
terms are negative and statistically significant. Looking at non-agricultural work in column
(2) of Tables 10 and 11 show similar results to those in primary activity, which is as theory
predicts. The impact of a monsoon shock leads to a reduction in the likelihood of having at
least 1 member in agricultural work by 0.08 and on average 0.11 people in the household.
This is statistically identical in near and far villages as can be seen by the insignificance on
NearGQ ·Monsoon. Looking at the coefficient on the triple interaction term we can see
that as predicted access to roads amplifies the equilibrium response.
Looking at non-agricultural labor, the results in secondary labor activity are more in
line with our theoretical predictions. Monsoon shocks cause significant declines in Non-
Agricultural participation. Furthermore looking at column (2) in Table 10 β7 the coefficient
of interest is negative and significantly. The results hold in Table 11.
4.2 Wage Responses
The next section looks at responses of wages. Table 12 depicts the impact on daily agricul-
tural wages. Column (3) suggest that a 1 unit positive monsoon shock leads to an increase
in daily wages of 2.39 rupees. The positive effect applies to districts close to the GQ in 1999.
This can be seen in the fact that coefficient on GQDummy ·Rainfall is positive and the sum
β3 + β4 is positive as well. However as can be seen in 2005Dummy ·Rainfall the coefficient
is negative and the sum β3 + β5 is also negative. This implies that for districts far from the
23
GQ, agricultural wages do not rise. The coefficient on the triple interaction suggests that
wage response is smaller as districts come closer to the Golden Quadrilateral in 2005, though
not statistically significantly. This is in keeping with our theoretical framework.
Table 13 conducts the similar exercise for non-agricultural wages. These are daily factory
wages for all workers in factories, excluding those in managerial positions. Looking at column
(3) depicts that a good rainfall shock paradoxically leads to a decrease in wages. This does
not vary across districts depending on if they are close or far, though this (measured by
coefficient on GQDummy · Rainfall) is also quite noisily measured. Finally looking at
the triple interaction term we obtain a small increase. While the result is in contrary to
the theoretical predictions, the large standard errors render the data too noisy to draw a
conclusion. It is noteworthy that the overall effect for β3 + β4 + β5 + β7 is positive, though
the remaining predictions do not hold.
5 Conclusion
Can labor markets even play a significant role in providing diversification to farmers? This
was the primary question we wish to answer in this paper. We began by building a static
general equilibrium framework on rural labor markets. The framework incorporates both
non-agricultural and agricultural labor markets, and provides predictions on equilibrium
outcomes. The improved ability to diversify is captured by a lower commuting cost to non-
agricultural jobs. If farmer with better access to roads (and thus urban labor markets),
we should see differential responses to the same productivity shocks in the agricultural and
non-agricultural labor markets, implying different ability to cope with shocks. The theory
predicted that this better ability to diversify should be seen in greater equilibrium responses
to shocks in terms of employment. At the same time, wages too would adjust differentially,
and to account for this we look at wage responses separately.
Our main findings remain ambiguous. Productivity shocks do seem to change wage
work as expected at times: primary work for agriculture and secondary for non-agriculture.
Unfortunately the results are contradicted by secondary activity for agriculture and primary
activity for non-agriculture. Furthermore the results for non-agricultural work are often
insignificant again contradicting the predictions. The results on wages are noisily estimated,
therefore fail to provide solid evidence in support of the theory.
24
References
T. Allen and D. Atkin. Volatility and the gains from trade. Technical report, National
Bureau of Economic Research, 2016.
S. Asher and P. Novosad. Market access and structural transformation: Evidence from rural
roads in india. Technical report, 2016.
A. Banerjee, E. Duflo, and N. Qian. On the road: Access to transportation infrastructure and
economic growth in china. Working Paper 17897, National Bureau of Economic Research,
March 2012. URL http://www.nber.org/papers/w17897.
R. Burgess and D. Donaldson. Can openness mitigate the effects of weather shocks? evidence
from india’s famine era. American Economic Review, 100(2):449–53, May 2010. doi: