1. For single country VARs, Creel and Poilon (2008 Creel, Jerome and Poilon, Gwenaelle. 2008. Is public capital productive in Europe?. International Review of Applied Economics, 22(6): 673–91. [Taylor & Francis Online] , [Google Scholar]) do not conduct unit root tests. For their panel estimates, Creel and Poilon use the Levin–Lin–Chu test – which assumes all countries follow the same unit process – and find that the relevant variables are I(0) when the variables are expressed as ratios to total employment.

2. Alternatively, the exponential coefficients in equation 1 would sum to one.

3. The impact of prior productivity changes on current levels of productivity is implicit in the dynamic empirical model. Theoretical justifications have been given for arguing that past productive activities can have an impact on future productivity, particularly in the endogenous growth literature. For example, Arrow’s (1962 Arrow, Kenneth. 1962. The economic implications of learning by doing. Review of Economic Studies, 29: 155–73. [Crossref], [Web of Science ®] , [Google Scholar]) ‘learning by doing’ model of endogenous productivity improvements suggests that future productivity growth may be a by‐product of current productive activities.

4. The cointegrating relationship can be estimated directly from the error‐correction specifications of equations 6a and 6b. An alternative two‐step procedure involves estimating the cointegrating relationship and the ECM separately, using OLS estimators. Both estimators possess similar asymptotic properties. See Stock (1987 Stock, James H. 1987. Asymptotic properties of least squares estimators of cointegrating vectors. Econometrica, 55(5): 1035–56. [Crossref], [Web of Science ®] , [Google Scholar]) for a fuller discussion.

5. In addition, investment in assets such as school buildings also responds to broad demographic trends (e.g. increases in the school‐age population) and may be de‐linked from economic development policies.

6. We use the term ‘unproductive’ in the sense that we expect their impact on private productivity to be negligible. The remaining public assets in the category of ‘core infrastructure’ include roads and highways; water, sewer and power infrastructure; other transportation infrastructure; and other structures and buildings.

7. In the error‐correction models presented in equation 6a, the coefficient on the labor input variable expressed in levels (L _t−1) is λ ₁+λ ₂. Similarly, the coefficient on the public capital variable (P _t−1) is γ ₁ + γ ₂.

8. The elasticity of output with respect to the public capital stock can be written as . If we multiply the elasticity by the ratio of output (Y) to public capital stock (P) we get or the change in output with respect to a change in the stock of public capital.

9. For this simple exercise, we do not attempt to account for the impact of prior changes in private capital productivity on future capital productivity as captured by the coefficient on the lagged capital productivity variable in the estimated models.