The level and growth effects in the empirics of economic growth: some results with data from Guatemala

Abstract

Mankiw et al. (1992) have extended the Solow (1956) model by augmenting the production function with human capital. Its empirical success is impressive and it showed a procedure to improve the explanatory power of the neoclassical growth model. This article suggests an empirical procedure to further extend the neoclassical growth model to distinguish between the growth and level effects of shift variables like the human capital. We use time-series data from Guatemala to show that while the growth effects of education are small, they are significant and dominate the level effects.

Acknowledgements

We thank a referee for helpful suggestions and Rup Singh of the University of the South Pacific for his comments and help. However, we alone are responsible for any errors that may remain.

Notes

¹A referee has pointed out that Abramovitz (1956) was the first to point out that the residual in the growth accounting exercises is a measure of our ignorance of the determinants of growth.

²Jones (1995) has listed 10 growth factors, identified by the EGMs and these are: physical investment rates, human capital investment rates, export shares, inward orientation, the strength of property rights, government consumption, population growth and regulatory pressure.

³The transition period between the steady-state growth rates is long and therefore, the growth rate will remain above the steady state growth rate for 15–20 periods. This insight is based on the simulation results with the closed form solutions of NCGM.

⁴However, Kocherlakota and Kei-Mu Yi (1996) have found limited support for EGMs with US data. They examined the effects of seven policy variables on the growth rate and found that only nonmilitary investment and nonmilitary structural investment has some effect on the growth rate.

⁵There seem to be difficulties in estimating the deep parameters underlying the inter-temporal utility optimization models with constant risk aversion utility functions. The main finding of the empirical work based on Hall's (1978, 1988) random walk hypothesis is that, it is difficult to estimate the inter-temporal elasticity of substitution and risk aversion parameters. However, Ogaki and Reinhart (1998) proposed a method to estimate these parameters by using durable and nondurable consumption expenditures. See also Campbell and Mankiw (1989).

⁶Hoover and Perez (2004) in their survey of cross-country works, for example, list 64 variables that have been used in various EGMs.

⁷MRW (1992, p. 412) hint that factor shares from national income data can also be used to compute steady-state income.

⁸In proposing this specification we are aware that this is an empirical modification. We justify this empirical modification because of its conformity with data. It would be interesting if someone develops a theoretical justification. Essentially what we are assuming amounts to the assumption that technical progress is due to the economy-wide externalities of variables like HKI and for individual firm these effects are like manna from heaven.

⁹The Sargan test statistic is computed when there are over-identifying restrictions; with the null hypothesis that the selected instruments are exogenous i.e. they are uncorrelated with the error term. When the null is not rejected, it can be said that the chosen instruments are exogenous and valid. However, the Sargan test is appropriate for large samples whereas our sample size is modest. The G R ²is a measure of goodness-of-fit of IV estimates, developed by Pesaran and Smith (1994). It is a valid discriminator of models based on IV method.

¹⁰There is no theoretical justification for such constraints and they are imposed for purely empirical reasons i.e. to gain a few degrees of freedom and reduce the SEs. Such empirical constraints are common in the GETS approach. We thank the referee for insisting on this clarification.

¹¹When trend was included in these equations, the coefficient of HKI and HKI²became negative and insignificant.

¹²The correlation between trend and HKI is high at 0.99 and it is difficult to obtain reliable estimates of the two parameters π ₁ and π ₂ in Equation 9 – see Equation VII, Table 2. The sign of π ₂ is positive and contrary to expectation. However, the summary statistics are impressive. We have also estimated an equation with quadratic growth effects. Although this equation is well determined, the G R ²was less at 0.639 and implied that HKI has its maximum growth effects when it is 2.8, which seems to be low. This result is not reported to conserve space.