ABSTRACT
This study analyses the relationship between public-private partnerships (PPPs) and corruption. Several econometric analyses were developed using a sample of 92 low- and middle-income countries over the period 1995–2018. There is research in the literature that discusses the theoretical vulnerability of PPPs to corruption, but this study adds empirical evidence, considering that developing countries have more problems with corruption than do other countries. The results suggest that corruption has a positive impact on the number of PPP arrangements and on the amount of investment commitments. This provides evidence of a positive link between corruption and PPP projects.
Acknowledgements
This work has been supported by the Consolidated Research Group EJ/GV: IT 897-16.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1. There are multiple definitions of PPP depending on the region, the sector of activity, the distribution of risk, and the distribution of ownership (Mota and Moreira Citation2015). Here, the contractual perspective is used to define PPP.
2. There is uncertainty about the physical nature of the network, or the external shocks that may affect the project in the future.
3. For a more detailed literature review, see Cui et al. (Citation2018) and De Castro e Silva Neto et al. (Citation2016).
4. Fewer than four contracts are recorded for each of these 33 countries over the entire period of 1995–2018.
5. GLS is usually preferred when the number of individuals is relatively lower than the length of the period (N < T). Here, N is larger than T, but T is relatively large for a conventional panel data model; that is, N < 100 and T > 15 (Labra and Torrecillas Citation2018). Accordingly, both methods, GLS and PCSE, are used.
6. Looking at Table 1, the variance of PPPdeals (117.3) is much higher than the mean value (3.22).
7. Although the zero-inflated negative binomial model is the most appropriate model in this case, equations (2), (3), and (4) were also initially estimated using the Poisson model, the negative binomial model, and the zero-inflated Poisson model. The results are available on request.
8. In a Poisson model, alpha is constrained to zero. Therefore, Stata finds the maximum likelihood estimate of the log of alpha and then calculates alpha from this. The alpha is therefore always greater than zero, and the negative binomial model only allows for overdispersion.