Abstract
Production frontiers with technical inefficiency determinants are estimated using stochastic models for textile manufacturing in eight developing countries encompassing about 800 firms. Inefficiency determinants are considered either on an individual basis, or in the form of composite indicators reflecting in-house or managerial factors and various dimensions of the external environment. Although each of these two categories of factors is statistically significant, the former proves more influential in the explanation of the difference in efficiency between firms. Simulations are then proposed to assess the efficiency levels that would occur if firms had the opportunity to produce in the most favourable productive environments.
Acknowledgements
The authors thank anonymous referees and the editor of the journal for valuable comments. They also acknowledge very useful feedback on an earlier version of the article from Patricia Augier, Arne Bigsten, Leonardo Iacovone, Christophe Muller, Sergio Perelman, Elisabeth Sadoulet, as well as from participants to the Economic Research Forum Annual Conference in Cairo, Egypt, November 2009. The authors acknowledge with gratitude grant assistance provided to a larger research program by FEMISE (Forum Euro-Méditerranéen des Instituts de Sciences Economiques). They also acknowledge the Fondation pour les Etudes et Recherches sur le Développement International (FERDI), as well as the French Government, Programme d'investissements d'avenir, for their financial support. The views expressed in this article are those of the authors and do not represent those of the IMF or IMF policy. The usual disclaimers apply.
Notes
1. The estimated standard errors have to be corrected owing to the use of [zcirc] instead of the observed z-variables.
2. The same method cannot be adopted for the u term because the Jondrow et al. (Citation1982) estimates do not provide perfect predictions of inefficiencies. This method does not provide estimates of ui, but the mean of the distribution from which ui is generated (Greene, Citation2008).
3. Regional averages concerning characteristics of the external environment are also useful to complete missing information on non-behavioural z-factors.
4. This percentage is calculated as .
5. Commander and Svejnar (Citation2008) refer to the 2005 and 2002 Business Environment and Enterprise Performance Surveys (BEEPS) collected by the European Bank for Reconstruction and Development (EBRD) and the World Bank. Firms are from a wide range of sectors in 26 transition countries.
6. The condition number is measured by the root of the largest eigenvalues of the correlation matrix divided by the smallest. Its high level (22.12), within the interval [15, 30], suggests the existence of multicollinearity between inefficiency determinants.
7. To correct for sample selection, Greene (Citation2010) developed a simulation based method for the normal-half normal usual stochastic frontier model. This procedure, which could be extended for the truncated normal case, remains an area for further research. The authors thank a referee who called attention to this point.
8. zj variables are PCIND1 and PCIND2, respectively.
9. For firms developing in an environment beyond (below) the upper (lower) quantile, adjusted and non-adjusted efficiencies measures are the same.