14
XGSi ∑ XGSi
Balassa index: χ44 = -10-------i i=1 10-------
∑XGS1,s ∕∑∑XGS1,s
s=1 / i=1 s=1
Economies of scale: x5t = Y
t 13
∑ Y
s=1
where sectoral indices are omitted for reasons of expositional clarity and with energy prices
( x11ta ) or wages ( x11tb ) included, respectively, in case of explaining energy-productivity growth
or labour-productivity growth. We expect energy prices and wages to be positively correlated
with, respectively, energy- and labour-productivity growth. We took a three-year moving
average for the energy price and wages to avoid capturing the effect of short-term price
fluctuations, assuming that investments in energy- and labour-augmenting technologies do
respond to a structural trend in energy price/wage developments rather than to short term
fluctuations. By including the investment share as an explanatory variable we test for the so-
called embodiment hypothesis or a vintage effect, assuming that higher investment will
contribute to increasing energy- and labour-productivity growth via technological change
embodied in new capital goods (see, for example, Howarth et al. 1991 and Mulder et al.
2003). We expect openness to have a positive impact on productivity growth, since an open
sector faces relatively strong competition as well as exchange of knowledge, which we both
assume to have a stimulating effect on productivity growth. The Balassa index is an indictor
measuring relative specialization patterns. We expect that if a country specializes in a
particular sector, that that sector will be technologically relatively advanced, and hence we
expect a positive effect on productivity. Finally, including an indicator for the relative size of
a sector within a country captures the potential effect of economies of scale on productivity
growth, assuming that a large sector is able to invest relatively much in R&D and in new
capital goods and, hence, might be a technological leader displaying relatively high
productivity growth rates.
The results of regressing average energy-productivity growth rates on initial energy
productivity levels and these additional explanatory variables, according to equation 6, are
presented in Table 4.17
17 We also controlled for different specifications of energy prices (current prices, 5-year moving average, and
log 3-year and log 5-year moving average), investment share ((I/Y)t_1 (I/K), (I/K)t_1 and ln(I/K)t_1), as well as an
interaction term of investment share and log initial energy productivity (ln(Y/E)0* (I/Y)). All these specifications
did not substantially alter the estimates. Details are available upon request.
15