NEW ESTIMATES OF TOTAL FACTOR PRODUCTIVITY GROWTH IN INDONESIAN MANUFACTURING

Abstract

This paper estimates new elasticities of value added with respect to labour and capital in Indonesian manufacturing, controlling for the simultaneity problem that potentially exists between the choice of input levels and a productivity shock (such as an increase in productivity due to new production processes), for plant exit, and for quasi-constant unobservable plant characteristics. It does so by applying the Levinsohn and Petrin (2003) production function estimator to plant-level value added, fixed assets, labour, and electricity consumption data over the period 1988–95. This methodology allows us to revisit the previously used growth accounting based elasticities, and thereby improves total factor productivity (TFP) estimates. The results show that, in the period under study, aggregate TFP growth in Indonesian manufacturing was higher than had previously been estimated.

I am heavily indebted to Kai Kaiser for data provision, and to Nick Crafts for his support. I thank participants in workshops on economic history at the London School of Economics, and on ‘Innovation, competition, and productivity: evidence from firm-level datasets’, Sophia-Antipolis (Nice-Côte-d'Azur), 19–20 December 2005, as well as two anonymous referees and the editor, for very valuable comments.

Notes

1 The backcast series are econometric estimates.

2 As I have no data on changes in electricity prices over time, or differences in electricity production costs, purchase costs or sales revenues, I carry out a sensitivity analysis using electricity consumption measured in kilowatts. The elasticities estimated using this measure are not significantly different from those estimated using electricity consumption measured in rupiah.

3 As an alternative, I used fuel consumption as a proxy for capital stock instead of electricity consumption, and found very similar results. Fuel consumption is in rupiah, deflated using the implicit deflator for GDP (World Bank, World Development Indicators).

4 Baltagi (1995) indicates that random effect estimation is to be preferred whenever the number of plants is large compared to the period covered by the panel dataset. Additionally, I perform a Hausman test comparing the fixed effect and the random effect estimation. The random effect model leads to more consistent estimates.

5 Results of the estimation of the relationship between fixed assets and output are not reported here, but are available from the author upon request.

6 I use Stata software. The Random Effect Generalised Least Squares panel data regression using instrumental variables is carried out using the ‘xtivreg’ command, the Hausman test is carried out using the ‘hausman’ command, and prediction is carried out using the ‘predict’ command.

Additional information

Notes on contributors

Virginie Vial

[email protected], Euromed Marseille School of Management, Domaine de Luminy, BP 921, 13288 Marseille cedex 9, France.