A Non‐parametric Examination of Capital–Skill Complementarity*
| DOI | http://doi.org/10.1111/j.1468-0084.2009.00550.x |
| Published date | 01 August 2009 |
| Date | 01 August 2009 |
519
©Blackwell Publishing Ltd and the Department of Economics, University of Oxford, 2009. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
OXFORD BULLETIN OF ECONOMICS AND STATISTICS, 71, 4 (2009) 0305-9049
doi: 10.1111/j.1468-0084.2009.00550.x
A Non-parametric Examination of Capital–Skill
ComplementarityÅ
Daniel J. Henderson
Department of Economics, State University of New York at Binghamton, Binghamton,
NY 13902-6000, USA (e-mail: djhender@binghamton.edu)
Abstract
This paper uses non-parametric kernel methods to construct observation-specific elas-
ticities of substitution for a balanced panel of 73 developed and developing countries
to examine the capital–skill complementarity hypothesis. The exercise shows some
support for capital–skill complementarity, but the strength of the evidence depends
upon the definition of skilled labour and the elasticity of substitution measure being
used. The added flexibility of the non-parametric procedure is also capable of uncover-
ing that the elasticities of substitution vary across countries, groups of countries and
time periods.
I. Introduction
Rising wage inequality has been a key feature of the US labour market since the
late 1970s. This phenomenon, both in the USA and across the world, has received
much attention in the literature. One possible explanation is provided by the
capital–skill complementarity (CSC) hypothesis. The hypothesis states that physical
capital and skilled labour are more complementary than unskilled labour and physical
capital. Assuming the hypothesis is true, an increase in physical capital, ceteris
*The author thanks two anonymous referees, the editor Jonathan Temple, Viera Chemlarova, David
Drukker, John Duffy, Li Gan, Subal Kumbhakar, Qi Li, Daniel Millimet, Salvatore Modica, Chris Papa-
georgiou, Chris Parmeter, Fidel Perez-Sebastian, Jeff Racine and Robert Russell for suggestions which led to
improving an earlier version of this paper. Comments from participants in the TexasA&M University Econo-
metrics Seminar Series, the Sam Houston State University Department Seminar Series, the University of
Houston/Rice University Macroeconomics Seminar Series, the Fifteenth Annual Meeting of the Midwestern
Econometrics Group (Carbondale, IL) and the Institutional and Social Dynamics of Growth and Distribution
Conference (Lucca, Italy) are also greatly appreciated. Finally,the author thanks Ozkan Eren and Ayfer Gurun
for excellent research assistance.
JEL Classification numbers: C14, C23, D2.
520 Bulletin
paribus, will increase the demand for skilled labour (and thus wages for skilled
labourers). Capital deepening seen across many economies in recent years combined
with CSC could be one such explanation for rising wage inequality.Thus, if significant
increases in physical capital have been made and CSC is shown to hold for a particular
economy, policymakers could use this information to possibly find ways to decrease
inequality.
Griliches (1969) finds empirical evidence that physical capital and skilled labour
are less substitutable than physical capital and unskilled labour and concludes that
the CSC hypothesis holds using a data set of US manufacturers. Since Griliches
(1969), the CSC hypothesis has been empirically studied in great detail. While some
authors debate the specification of the model, others debate the type of data which
should be studied. Fallon and Layard (1975), and others, study the CSC hypothesis on
an international scale. Specifically, they piece together data from 22 developed and
developing countries for the year 1963. They find mild evidence in favour of the CSC
hypothesis. Duffy, Papageorgiou and Perez-Sebastian (2004), hereafter DPP, extend
the work of Fallon and Layard (1975) to a balanced data set of 73 countries over
a 25-year period (1965–90). They use a two-level constant elasticity of substitution
production function specification and use nonlinear estimation methods which allow
them to relax the assumption of perfectly competitive markets. Further, they pres-
ent five alternative ways of dividing the labour force into two categories, ‘skilled’
and ‘unskilled’.1This allows them to find the greatest support for CSC when the
category for skilled labourers is defined as those who have attained some second-
ary education, those who have completed primary education or as those who gained
some primary education. Part of their purpose for looking at an international panel
was to find evidence of CSC over long periods of time and across countries at differ-
ent stages of development. This strategy was partly influenced by Goldin and Katz
(1998) who note that physical capital and skilled labour have not always been viewed
as relative complements. In particular, they suggest that transitions between pro-
duction processes change the relative demand for skill; thus, different economies at
different times may or may not possess CSC.
Most studies simply give a single conclusion for all observations. This could prove
to be detrimental. For example, suppose a researcher found that CSC exists in a panel
of countries. Then if countries take that information as given, it may affect their
policy. If CSC holds true for that economy, it could increase spending on education
to potentially reduce the impact of advancing technology on inequality. However,
if CSC does not exist, those resources spent on education may have been better
allocated.
Although observation-specific estimates seem logical, the aforementioned papers
simply give a single estimate for each elasticity.2An increasingly popular method
to obtain observation-specific estimates is to use non-parametric kernel methods. In
1These categories or thresholds will be described in greater detail in section III.
2There is, however, some work being done with translog cost functions (which require price data) that allow
for observation-specific estimates (Bergstr¨omand Panas, 1992; Ruiz-Arranz, 2002).
©Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2009
To continue reading
Request your trial