EDUCATIONAL OUTCOMES IN OECD COUNTRIES
28
above). By contrast, in the augmented neoclassical growth model, changes in test scores
lead to higher steady-state levels of income but do not affect the long-run growth path.
Our empirical growth model captures the conditional convergence implied by the
neoclassical model - but also by a set of endogenous growth models - through including
the initial GDP level as a control variable. An alternative approach for the projections is
thus to interpret the model in the neoclassical rather than endogenous-growth framework
and have educational reforms affect the steady-state level of income but not its long-run
growth.
To do so, we re-estimate our growth model with the logarithmic (rather than linear)
per-capita GDP as control. The test-score coefficient hardly changes in this specification
(1.718 rather than 1.864), and the coefficient on log initial income is -1.835. This
estimated convergence rate of 1.8 percent is very close to the one expected under
standard parameter assumptions in the augmented neoclassical growth model (Mankiw,
Romer, and Weil (1992)). It means that (approximating around the steady state) an
economy moves halfway to its steady state in about 38 years. Including this
convergence process in our simulations allows us to perform projections that are in line
with neoclassical growth theory. In these projections, growth rates with and without
education reform will differ only during the transition to the new balanced growth path.
In the long run, the economy will grow at the same rate after the reform as without the
reform.
We use the estimates from this model to simulate the trajectory from the old to the new
balanced growth path in each year during our time horizon. To implement the idea that
the world technological frontier grows at 1.5 percent in the absence of education reform
in this model, we assume that in the aggregate the three countries with the largest shares
of patents in the world - the United States, Japan, and Germany - grow at 1.5 percent
without reform. Together, the three countries currently account for over 70 percent of
worldwide patents (measured in triadic patent families, Organisation for Economic Co-
operation and Development (2008)). We thus choose a constant growth parameter for
each future year that has the weighted average of the three countries grow at 1.5 percent
each year, where the weights are each country’s share in their combined GDP in the
previous year.21
Table 8 shows the results of the projections based on the neoclassical model
specification. In reform Scenario I, where each country increases by 25 PISA points, the
value of the reform - the discounted value of the future increases in GDP - amounts to
$90 trillion in present value terms. While this provides a neoclassical lower bound to
our previous projection of $123 trillion, the noteworthy fact is that over the time horizon
of our projections until 2090 (and for a reform that takes until 2070 to take full force in
21 We also experimented with alternative ways to implement the growth of the technological frontier, and results fall into the
same ballpark. The alternatives include holding the following growth rates constant at 1.5 percent: the (weighted) OECD
average growth rate, the U.S. growth rate, a simple or weighted average of the U.S. and Finish growth rates, and a specification
where “technological leadership” (as depicted in per-capita GDP) turns from the U.S. to Finland in the 2030s. The latter
specifications reflect the fact that Finland, the country with the highest test scores, is projected to have the highest steady-state
level of per-capita GDP in our model. Note that - consistent with such a model - in terms of patenting new technologies,
Finland is currently already the country with the largest revealed technological advantage in ICT, as measured by ICT-related
patents in total patents (Organisation for Economic Co-operation and Development (2008)).