Abstract
In the cross-country and time series studies on the determinants of the growth rate, capital stock is often proxied with the investment ratio due to lack of reliable data. While investment ratio may give good results with OLS, their robustness is doubtful. In addition there are other problems: (a) investment ratio may be a good proxy for the change in capital stock, but combining it with the level of employment to estimate a production function leads to misspecification bias; (b) it is not possible to impose any valid constraints on the coefficients of these variables and (c) if instrument variables are used to minimize any endogenous variable bias, the equation with the investment ratio seems to be fragile.
Notes
1 This does not mean that GETS is a superior technique. In fact it is hard to say which of the alternative techniques is better. Our experience shows that these alternative techniques give close results in large samples. For an interesting symposium on the merits and limitations of GETS see Hoover and Perez (Citation1999), Hendry and Krolzig (Citation1999) and especially Granger and Timmermann (Citation1999).
2 The instrument variables in (8) are: intercept, trend, ln Y t−2, ln K t−2, ln L t−2, Δln K t−1, Δln L t−1 and DUM9495 t . The same instruments, but by replacing capital with the investment ratio are used for Equation 10. However, it was found necessary to use Δ(I/Y) t −2 instead of its one period lagged value. The latter gave much worse results than those reported in (10).