A. Global Trends in PPPs

The last 30 years have seen a dramatic rise of PPPs in emerging and developing countries. After a slowdown during the 1997 Asian financial crisis, PPPs have started to increase in developing Asia and Latin America (LAC), especially after the 2008 global financial crisis. Having spread across the globe, PPP investments doubled from about $50 billion per year on average during 2000–05 to about $100 billion per year on average during 2010–15. Most projects are in large MICs, such as BRICs, Argentina, Mexico, Colombia, Philippines, Indonesia, Turkey, and Nigeria (see appendix Figure A1). Countries in developing Asia and LAC have experienced a spike in PPPs since 2004, reaching about $50–60 billion per year, while PPPs remain relatively small in other regions.

While the dispute risk peaked in the late 1990s to the early 2000s in developing Asia and LAC, it continued to rise in sub-Saharan Africa (SSA) where more contracts started to be signed (Figure 2). The dispute is more common in developing Asia, the Central and Eastern Europe & the Commonwealth Independent States (CEE-CIS) and LAC, particularly in the upper-MICs.

View Full Size

Overall Trend of PPP Investments

Citation: IMF Working Papers 2017, 243; 10.5089/9781484328286.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Evolution of Dispute Risk

Citation: IMF Working Papers 2017, 243; 10.5089/9781484328286.001.A001

Source: World Bank PPI database

• Download Figure
• Download figure as PowerPoint slide

B. Country-level Determinants of Contractual Disputes

Appendix Figure A1 shows that PPPs (a measure of a government’s past experiences in PPP schemes for infrastructure financing) and observed disputes tend to cluster in particular regions. One explanation of this cluster is that disputes are more common in countries vulnerable to financial crises. Dispute risk also tends to rise when an economy is hit by external shocks (such as commodity price fluctuations, natural disasters, and adverse shocks in growth and inflation) (Nose 2014). Another determinant of disputes relates to political regime change. Another explanation is institutional. Previous literature found that countries with authoritarian political regimes and poor economic and political institutions are more likely to initiate disputes (Acemoglu and Robinson, 2005; Eden, Kraay and Qian, 2012). Such findings are consistent with the patterns found in pre-claim data on breach of contract that were collected by World Bank’s political risk insurance agency (MIGA). In many cases, economic crisis and a political regime change exposed PPP projects to financial stress or inconsistent policy (or both), leading to the renegotiations of contract.

C. Contract Design and Fiscal Institutions

In many PPP arrangements, governments offer guarantees, direct subsidies, or the Viability Gap Funds (VGFs) to support private participation in infrastructure (see Engel, Fischer, and Galetovic (2014) for the example of VGFs in India).^{4 5} While more private investors had sought for the supports from multilateral financial institutions (MFIs) in the early 1990s, the provisions of guarantees and direct subsidy have recently become important risk mitigation instruments for PPPs (Figure 3). In regional context, PPP contracts that received government guarantees and direct subsidies were concentrated in developing Asia and LAC (Figure 4). In both regions, concessions were also common. In SSA countries, guaranteed contracts have gradually increased but are still rare, and concessions and MFI-supported projects are more common. The recent increase in government involvements in PPPs could elevate the fiscal risk as more guarantees and subsidies are provided, especially if they are granted to pursue projects which are not always financially viable.

View Full Size

Change in Government Involvements

(in percent of newly signed contracts)

Citation: IMF Working Papers 2017, 243; 10.5089/9781484328286.001.A001

Source: World Bank PPI database

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Regional Distribution of Each Instruments

(in count)

Citation: IMF Working Papers 2017, 243; 10.5089/9781484328286.001.A001

Sources world Bank PPI database

• Download Figure
• Download figure as PowerPoint slide

Indeed, PPPs were often motivated by country’s fiscal constraints on borrowing and taxation that limited the government’s financial ability to undertake public investments. PPPs are often used to circumvent budget constraints and to postpone recording the fiscal costs of investments rather than for efficiency reasons. In many countries, PPPs have not always performed better than public procurement as they have functioned as ad-hoc, off-budget arrangements without robust process for risk assessments. The success of PPPs depends on the effective procurement decisions, the economic and financial feasibility, and their efficient management and oversight under strong fiscal and legal institutions (Sabol and Puentes, 2014). It also requires transparent disclosure of fiscal risk in the medium-term budgetary frameworks to monitor and manage contingent liabilities.

A. Procurement

A government decides whether to use public procurement or PPPs based on its cost-benefit calculation and project profitability. With public procurement, the government allocates budget to make public investment z which yields project returns W_A. With PPPs, the government sets up a private company – special purpose vehicle (SPV) – and engages with the SPV to develop, operate, and maintain the infrastructure. The SPV invests k, complementing public investments, and the return π is linked to service outcomes and performance of the asset W_B. The SPV’s return depends on expected demand for services q, project financing cost r, and operational costs c₁ for SPV. PPPs are often procured through a non-competitive process (e.g., unsolicited proposals) in the pre-matured PPP markets. PPP project could thus be inefficient, if politically motivated, and creates short-term rents R.

Higher PPP regulation and operational quality X₁ ∈ [0,1] limit politician’s rent-seeking activities. The procurement selection is determined by the majority voting under citizen-candidate model (Osborne and Slivinski, 1996; Besley and Coate, 1997). Voters are assumed to have policy bias σⁱ to PPPs which is uniformly distributed: ${\sigma }^i\sim U[-\frac{1}{2b},\frac{1}{2b}]$. PPPs will be selected when it attracts majority voting: Pr(W_A < W_B + σⁱ) > 1/2. This results in the condition below:

This condition implies that PPPs are chosen because of its higher efficiency gain than public procurement (f(k,z) ≫ f(z)); or, regardless of project’s feasibility, for the adverse selection caused by (i) overestimation of project return of PPPs Eπ ex-ante (“optimism bias” (Flyvbjerg, 2009)) and (ii) rent-seeking by politicians, both due to weak PEVI quality.

B. Investment Decision

When Eq. (1) is satisfied, the government pursues a PPP project under the build-operate-transfer (BOT) contract. The payment scheme can be: (i) government-pay (government provides a recurring subsidy (called availability payments) $\overline{\alpha }$ to firm, and firm invests in project) or (ii) user-pay (firm collects user-fees, and makes capital and operational investments). Government makes take-it-or-leave-it offer to the firm with contractual term: $A=(p,\overline{\alpha },\mu )$ where p is user fee and $\overline{\alpha }$ is availability payments, μ. takes zero for the user-pay and one for the government-pay scheme. The asset will return to government when the contract terminates at time $\overline{t}$.

Government decides the optimal commitment on PPPs which maximizes the present value of output under the following budget constraint:

where T is lump-sum tax revenues, i_t = [z_t+1 — (1 — δ)z_t] is infrastructure investment, and c₂ is operational cost for the government to maintain the infrastructure capital, pq is user-fees where q is stochastic demand for public services: q_t = ρq_t-1 + ε_t and ε_t~N(0, σ²). The government may offer direct subsidy $\overline{\alpha }(if\mu =1)$, and revenue guarantee p(q — q)z_t to ensure the minimum user-fee revenues pqz_t to the firm. The project yields financial loss if the demand for public services decreases below the break-even level. Fiscal sustainability requires the fiscal deficit to be below the fiscal limit L.

The budget constraint incorporates transparency in reporting the fiscal risk of guarantees. The budget transparency is higher as the value of X₂ approaches one, while close to zero if a contingent liability is hidden by undertaking the PPP off-budget.

Assuming the Cobb-Douglas production function $f(k_t,z_t)=k_t^{\phi }{z}_t^{\gamma }$, the return to private investment φ increases as service quality improves. The firm chooses the optimal level of private investment $k_t^{*}=\text{argmax}(k_t^{\phi }{z}_t^{\gamma }+\mu \overline{\alpha }-r{k}_t-c_1{k}_t)$ where last two terms are the firm’s financing and operational cost every period. Given k*, the dynamic optimization solves the optimal level of public investment z* as follows:

where E_tλ_t+1 is the dynamic Lagrange multiplier which gets lower as lump-sum tax increases $(\frac{\partial E_t{\lambda }_{t+1}}{\partial T_t}<0)$ and higher as fiscal costs (direct subsidy and fiscal risk of guarantees) increase$(\frac{\partial E_t{\lambda }_{t+1}}{\partial \overline{\alpha }}>0,\frac{\partial E_t{\lambda }_{t+1}}{\partial {\chi }_2}>0)$ (see appendix E)

From Eq.(3), $\frac{{\partial }_{z_t}^{*}}{\partial {\chi }_2}<0$, implying that the optimal level of PPP investment declines as budget transparency improves. In other words, the lack of budget transparency leads to over-commitment on PPPs. If combined with the adverse selection of riskier PPP investments, this leads to over-commitment on infeasible PPPs which likely trigger guarantees (Maskin and Tirole, 2008; Milesi-Ferretti, 2003).

C. Conditions for Outbreak of Disputes

For both the government-pay and the user-pay schemes, the model illustrates the conditions under which disputes are initiated either by the firm or the government.

Firm-led disputes

■ Under the government-pay scheme:

The government-pay contract (μ = 1) is financed by government’s subsidy $\overline{\alpha }$. The firm refuses to participate in the contract if the participation constraint is violated (Π_t < 0) where ${\mathrm{\Pi }}_t=\overline{\alpha }-(r+c_1)k_t^{*}$. In this case, the firm initiates disputes when subsidy is insufficient to recover financing and operational costs: $\overline{\alpha }<(r+c_1)k_t^{*}for\forall t$.

■ Under the user-pay scheme:

The use-pay contract (μ = 0) creates financial loss if user-fee revenues are lower than the minimum threshold. The loss occurs with probability F(∊) = Pr (qt < q) = Ф (q — ρq_t-1) where Ф(·) is the cumulative distribution function. Firm’s participation constraint is violated if ${\Pi }_t^{’}=p\left[F\right(\epsilon )\underset{-}{q}+(1-F\left(\epsilon \right)\left)q\right]-(r+c_1)k_r^{*}<0for\forall t$.

When PIM quality is weak, the break-even public service demand q would be over-estimated ex-ante, increasing the likelihood of incurring financial loss F(∊).

Government-led disputes

■ Under the government-pay scheme:

The government decides whether to continue PPPs by comparing welfare levels when the project completes (W_H) vs. being terminated (W_NH). Successful PPP projects continue until period $\overline{t}$ and yields welfare W_H, which is the sum of net present value (NPV) of expected outputs and the SPV’s profit, and the residual value of the asset I at the end of the contract.

After the project starts, the government may breach the contract at period τ but pay the reputation cost H, and continue the project under public procurement afterwards. In this case, the welfare function after the period T becomes:

where $\tilde{z}*$ is the optimal public capital under traditional public procurement. The reputation cost would be higher when the government faces stronger political constraint σ with H′(σ) > 0. Contract will be terminated if:

The first square bracket of Eq. (4) captures the efficiency gains of PPP compared to public procurement, and the second square bracket captures the value for money. Both terms get larger when PPPs become more cost-effective. The government-led dispute occurs when the sum of these two terms get smaller than the right-hand side (i.e., the residual value of the asset net of reputation cost).

■ Under the user-pay scheme:

Similarly, the welfare W′_H is defined as follows for the user-pay scheme:

The third time captures the fiscal risk for the government. As W_NH remains the same, the condition for contract breach is expressed as follows:

Eq. (5) implies that the accumulation of PPP’s contingent liabilities could induce the government to initiate disputes.

From the theoretical model, testable predications can be obtained as below:

• Hypothesis 1 (Adverse selection and PIM quality): Weaker PIM quality X₁ could result in the adverse selection of high risk projects.

• Hypothesis 2 (Budget transparency): Lack of budget transparency X₂ in reporting contingent liabilities from guaranteed debt leads to over-commitment on PPPs.

• Hypothesis 3 (Effect of guarantee): The provision of guarantees encourages firm’s participation in PPPs, while increasing the government-led disputes for higher fiscal risk and insufficient returns to recover project costs.

• Hypothesis 4 (Reputation costs): The likelihood of reneging contract is lower with stronger political constraint: $\frac{\partial \Omega }{\partial H\left(\sigma \right)}>0$.⁶ (“sovereign theft” by Tomz and Wright (2010))

A. Data and Sample

Contractual dispute data come from the World Bank’s Private Participation in Infrastructure (PPI) Database. Additional data were collected from various public sources (such as Factiva, websites of multilateral agencies and sponsors) and information from project entities. While Guasch, Laffont, and Straub (2007) focused on concessions in Latin America, it covers both greenfield contracts, represented by BOT, and brownfield concessions for 113 emerging and developing countries (see Table A1 for country list and Table A2 for the summary statistics). The data include about 6,000 PPP contracts signed between 1984 and 2015.

The data examine the nature of disputes in detail, and distinguish whether they are triggered by the government or private investors. Contracts are classified as “disputed” if they were distressed, renegotiated or cancelled due to conflicts between the government and private parties. Disputes triggered by sovereign risk including economic crisis (financial crisis, sharp currency devaluations) and political risk (nationalization, expropriation, policy changes in tariff setting) are classified as “government-led”, while they are “firm-led” disputes if triggered by sponsor’s insolvency and technical problems. Even though contracts experienced distress, they are classified as “non-disputed” if the renegotiations were caused by the sponsor’s unilateral actions (e.g., change in business strategy) or uninsurable external events beyond the control of the parties (such as war and civil conflict).

B. Hazard Estimation

The reduced-form approach (the hazard model) is used to test the hypotheses derived in section III. It estimates the probability of disputes conditional on the duration of contract A_i defined as the number of years a project survives before it ends due to contract breach or termination. The data are right-censored if the project is still ongoing in 2015. The observed duration is A_i = min(A*_i, c) where c is 2015 for all censored contracts. Assuming the density of A_t follows the Weibull distribution,⁷the hazard function λ can be estimated parametrically as: $\lambda (A_i,x_{icr})=\exp \left(x_{icr}^{’}\beta \right)\alpha A_i^{\alpha -1}$ where α is the measure of duration dependence. Standard errors are clustered over industry and contract year levels.

The hazard regression evaluates fiscal commitments, and contract- and country-level characteristics in determining dispute risk. In the above hazard function, x_icr = [X_1,i; X_2,c; D_jr] are covariates for contract i in sector j, country c and region r. X_1,i is contract-level characteristics such as financial arrangements with MFIs, the government’s provision of guarantees and direct subsidies, and concession agreement. X_2,c is the country-level variables on the short-term macroeconomic fluctuations (average real GDP growth and inflation during the contractual period) and the quality of political institutions (democratic regime and the duration of national leaders).⁸ Dummies for regional and sectoral affiliations D_jr are also included to account for the regional and sectoral clustering of dispute events as shown in section II.

C. Bias Adjustment using the Propensity Score Matching (PSM)

The selection of contract type would depend on the characteristics of the government and private firms, which biases the estimates of contract-level variables. Previous literature applied the PSM (Austin, 2013; Gayat et al, 2012) or instrumental variable (Guasch, Laffont, and Straub, 2007) to minimize the selection bias. Given the non-linearity of the hazard model and the absence of valid instruments, this paper applies the PSM where the hazard estimates are adjusted by applying the inverse probability of treatment weighting (IPTW). The PSM-IPTW method offers the most parsimonious specification with least mean squared error when estimating the hazard model (Austin 2012). The selection on unobservable could still exist for unobserved omitted factors (i.e., investor’s strategic skills in drafting contracts) (Guasch, Laffont, and Straub, 2007), although bias would be minimized by including comprehensive set of country- and investor-level covariates to match the characteristics between different contract types.

A. Determinants of Contract Selection

As the first stage of the hazard regression, the probit model estimates the probability of drafting a contract with (i) MFI support (loans, equity investments, or guarantees), (ii) guarantees, (iii) direct subsidies, and (iv) concession agreement. It includes country-level variables (regulatory quality and bureaucratic efficiency as measured by the World Bank’s governance indicator, fiscal and public debt situations, and commodity exporter status), contract-level variable (investment size), and investor’s risk attitude (captured by a dummy whether the sponsor originates local or foreign). Regional and sector dummies are also controlled. Main results in Table 1 are summarized as follows:

• Gross financing needs (GFN) (in percent of GDP) is positively signed for guarantees, implying that government with large GFN tends to make suboptimal risk commitments in PPPs with guarantees to avoid immediate payments in the short-term. In column 3, the public debt is interacted with Wang et al. (2015)’s budget transparency index in terms of the coverage of fiscal reporting on stock (in column 3; see more details in section V.D). The positive coefficients of public debt variable and the negative interaction term with the budget transparency index indicate that indebted governments tend to make risky commitments with guarantees, but such suboptimal risk commitment decreases in countries with higher budget transparency. The GFN variable is negatively signed for direct subsidies. As the fiscal financing requirements increase, government has less fiscal space to subsidize the project while relying more on guarantees and concessions to implement the project.

• The contract selection also hinges on the quality of institutions. The government with better regulatory environment (central regulatory body in supervising the PPPs) can formulate large-scale PPP projects with MFI supports. On the other hand, when the bureaucracy process is more efficient, government may well improve the quality of existing infrastructure under concessions. Higher bureaucratic efficiency is also associated with a lower probability for government to offer guarantees and makes private financing feasible with direct subsidies.

• Local investor’s participation in the contract increases the likelihood that projects receive direct subsidies,⁹ reflecting private party’s strategic decision in involving local investors in the government-funded projects. Likewise, for concessions, government prefers to select private parties with local investor’s participation to offer business opportunities to local partners, to utilize their local knowledge and network, or to maintain the government’s control over the asset.

• PPP contracts signed in commodity exporting countries tend to receive more guarantees and direct subsidies than commodity importers.

Table 1.

Determinants of Government’s Financial Commitment

* Significant at 10%; ** 5%; *** 1%. Robust standard errors are presented in the square brackets. Average marginal effects of each covariate are presented.

Table 1.

Determinants of Government’s Financial Commitment

	(1)	(2)	(3)	(4)	(5)
	MFI supports	Guarantees		Direct subsidy	Concession
Ln(investment size)	0.030*** [0.002]	0.002 [0.002]	0.003 [0.002]	0.003 [0.002]	−0.002 [0.003]
Ln(GDP per capita in 2000)	−0.141 [0.171]	−0.395** [0.192]	−0.289 [0.194]	0.635*** [0.236]	−0.294 [0.201]
Ln(GDP per capita in 2000) squared	0.005 [0.010]	0.027** [0.012]	0.019 [0.012]	−0.045*** [0.014]	0.014 [0.012]
Ln(GFN/GDP)	0.055*** [0.013]	0.052*** [0.014]	0.058*** [0.014]	−0.084*** [0.018]	0.059*** [0.018]
Ln(public debt/GDP)	−0.015 [0.011]	−0.017 [0.013]	0.036** [0.017]	0.051*** [0.014]	0.031** [0.013]
Budget transparency index (stock)		−0.003 [0.010]	0.397*** [0.101]	0.053*** [0.011]	0.001 [0.011]
−— x Ln(public debt/GDP)			−0.104*** [0.026]
Regulatory quality	0.076*** [0.020]	0.068*** [0.020]	0.031 [0.022]	−0.132*** [0.025]	−0.076*** [0.024]
Bureaucratic efficiency	−0.119*** [0.025]	−0.165*** [0.023]	−0.120*** [0.026]	0.261*** [0.032]	0.064** [0.030]
Commodity exporter	−0.006 [0.011]	0.037*** [0.012]	0.041*** [0.012]	0.065*** [0.012]	−0.011 [0.012]
Sponsor with same nationality	−0.043*** [0.009]	0.001 [0.009]	0.003 [0.009]	0.049*** [0.010]	0.039*** [0.011]
Dummy: post-2000	−0.062*** [0.010]	0.176*** [0.017]	0.171*** [0.017]	0.191*** [0.017]	−0.051*** [0.012]
Observations	5,877	5,877	5,877	5,877	5,877
Log-likelihood value	−1,873.2	−1,807.5	−1799.8	−1,804.5	−2,243.4
Pseudo R-squared	0.142	0.195	0.199	0.174	0.348
Sector & region dummies	Y	Y	Y	Y	Y
Average(dependent var)	0.124	0.128	0.128	0.123	0.272

* Significant at 10%; ** 5%; *** 1%. Robust standard errors are presented in the square brackets. Average marginal effects of each covariate are presented.

View Table

Table 1.

Determinants of Government’s Financial Commitment

	(1)	(2)	(3)	(4)	(5)
	MFI supports	Guarantees		Direct subsidy	Concession
Ln(investment size)	0.030*** [0.002]	0.002 [0.002]	0.003 [0.002]	0.003 [0.002]	−0.002 [0.003]
Ln(GDP per capita in 2000)	−0.141 [0.171]	−0.395** [0.192]	−0.289 [0.194]	0.635*** [0.236]	−0.294 [0.201]
Ln(GDP per capita in 2000) squared	0.005 [0.010]	0.027** [0.012]	0.019 [0.012]	−0.045*** [0.014]	0.014 [0.012]
Ln(GFN/GDP)	0.055*** [0.013]	0.052*** [0.014]	0.058*** [0.014]	−0.084*** [0.018]	0.059*** [0.018]
Ln(public debt/GDP)	−0.015 [0.011]	−0.017 [0.013]	0.036** [0.017]	0.051*** [0.014]	0.031** [0.013]
Budget transparency index (stock)		−0.003 [0.010]	0.397*** [0.101]	0.053*** [0.011]	0.001 [0.011]
−— x Ln(public debt/GDP)			−0.104*** [0.026]
Regulatory quality	0.076*** [0.020]	0.068*** [0.020]	0.031 [0.022]	−0.132*** [0.025]	−0.076*** [0.024]
Bureaucratic efficiency	−0.119*** [0.025]	−0.165*** [0.023]	−0.120*** [0.026]	0.261*** [0.032]	0.064** [0.030]
Commodity exporter	−0.006 [0.011]	0.037*** [0.012]	0.041*** [0.012]	0.065*** [0.012]	−0.011 [0.012]
Sponsor with same nationality	−0.043*** [0.009]	0.001 [0.009]	0.003 [0.009]	0.049*** [0.010]	0.039*** [0.011]
Dummy: post-2000	−0.062*** [0.010]	0.176*** [0.017]	0.171*** [0.017]	0.191*** [0.017]	−0.051*** [0.012]
Observations	5,877	5,877	5,877	5,877	5,877
Log-likelihood value	−1,873.2	−1,807.5	−1799.8	−1,804.5	−2,243.4
Pseudo R-squared	0.142	0.195	0.199	0.174	0.348
Sector & region dummies	Y	Y	Y	Y	Y
Average(dependent var)	0.124	0.128	0.128	0.123	0.272

* Significant at 10%; ** 5%; *** 1%. Robust standard errors are presented in the square brackets. Average marginal effects of each covariate are presented.

B. Hazard Regression

In Table 2 below, baseline hazard estimates show that contract duration, contract characteristics and country-level variable affect the hazard rate of PPP in expected ways:

• As reported at the bottom of Table 2, duration dependence parameter is always greater than one (a > 1), confirming the positive duration dependence. This implies that the probability of disputes increases as the contract matures and the government would have stronger incentives to renegotiate contract, as predicted by the obsolescing bargain hypothesis (Woodhouse, 2006).¹⁰

• The investment size is positively signed and significant, indicating that larger PPP projects face higher dispute risk. A dummy of MFFs supports is negatively signed, implying that involving multilateral institutions, which have more experiences with PPPs than other private sponsors or lenders, could mitigate dispute risk for foreign investors. The effect is however not significant on average.

• The country-level variables show broadly consistent results as found in the literature. The negative coefficients of democratic regime (column 1) and the duration of leaders in democratic countries (column 2) suggest that PPP contracts are more likely to survive in democratic countries with stable political regime. Real GDP growth is negatively signed while CPI inflation is positive, showing that improvements in macroeconomic conditions during the contract would create buffers to reduce the dispute risk.

• Learning from past experiences is also an important factor in implementing PPPs. From the EVIF’s investment and capital stock dataset, the value of the stock of PPPs (in percent of GDP) is added in column 4.¹¹ The coefficient is negative, although not statistically significant, suggesting that PPP contracts tend to survive longer in countries with more PPP experiences.

Table 2.

Parametric Hazard Regressions (Baseline)

* Significant at 10%; ** 5%; *** 1%. Robust standard errors are clustered at industry and contract year levels and presented in the square brackets.

Table 2.

Parametric Hazard Regressions (Baseline)

	All disputes
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Ln(investment size)	0.147*** [0.037]	0.144*** [0.036]	0.113 [0.069]	0.110 [0.069]	0.092 [0.067]	−0.100 [0.175]	0.226*** [0.050]
MFI supports	−0.151 [0.185]	−0.126 [0.193]	−0.216 [0.231]	−0.213 [0.230]	−0.996* [0.546]	−0.760 [0.829]	−0.356 [0.331]
Democracy	−0.049** [0.021]		−0.034 [0.034]	−0.035 [0.034]	−0.162*** [0.039]	−0.275*** [0.099]	−0.139*** [0.053]
Duration of national leader		0.909*** [0.298]
−--- x Democracy		−0.233* [0.137]
Real GDP growth	−0.256*** [0.052]	−0.261*** [0.054]	−0.245*** [0.089]	−0.257*** [0.089]	−0.279*** [0.092]	−0.288* [0.168]	−0.377*** [0.089]
CPI inflation	0.039*** [0.002]	0.038*** [0.002]	0.038*** [0.004]	0.038*** [0.004]	0.048*** [0.005]	0.056*** [0.016]	0.018*** [0.004]
PPP capital stock/GDP				−0.004 [0.016]
Guarantee					1.453** [0.595]
Direct subsidy						0.912 [0.593]
Concession							1.037*** [0.246]
Constant	−5.556*** [0.708]	−8.131*** [1.079]	−5.255*** [0.836]	−5.149*** [0.827]	−4.796*** [1.065]	−6.136*** [2.216]	−5.339*** [1.043]
Duration dependence
Ln(α)	0.076 [0.055]	0.080 [0.054]	0.046 [0.073]	0.042 [0.074]	0.376* [0.212]	0.566 [0.373]	0.162** [0.066]
Observations	5,976	5,975	5,861	5,595	5,861	5,861	5,861
Log-likelihood	−1,499.9	−1,486.8	−2,681.7	−2,671.3	−2,968.2	−4,124.3	−3,835.2
Sample	All	All	All	All	All	All	All
PSM-IPTW applied			Y	Y	Y	Y	Y
Sector and region dummies	Y	Y	Y	Y	Y	Y	Y

* Significant at 10%; ** 5%; *** 1%. Robust standard errors are clustered at industry and contract year levels and presented in the square brackets.

View Table

Table 2.

Parametric Hazard Regressions (Baseline)

	All disputes
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Ln(investment size)	0.147*** [0.037]	0.144*** [0.036]	0.113 [0.069]	0.110 [0.069]	0.092 [0.067]	−0.100 [0.175]	0.226*** [0.050]
MFI supports	−0.151 [0.185]	−0.126 [0.193]	−0.216 [0.231]	−0.213 [0.230]	−0.996* [0.546]	−0.760 [0.829]	−0.356 [0.331]
Democracy	−0.049** [0.021]		−0.034 [0.034]	−0.035 [0.034]	−0.162*** [0.039]	−0.275*** [0.099]	−0.139*** [0.053]
Duration of national leader		0.909*** [0.298]
−--- x Democracy		−0.233* [0.137]
Real GDP growth	−0.256*** [0.052]	−0.261*** [0.054]	−0.245*** [0.089]	−0.257*** [0.089]	−0.279*** [0.092]	−0.288* [0.168]	−0.377*** [0.089]
CPI inflation	0.039*** [0.002]	0.038*** [0.002]	0.038*** [0.004]	0.038*** [0.004]	0.048*** [0.005]	0.056*** [0.016]	0.018*** [0.004]
PPP capital stock/GDP				−0.004 [0.016]
Guarantee					1.453** [0.595]
Direct subsidy						0.912 [0.593]
Concession							1.037*** [0.246]
Constant	−5.556*** [0.708]	−8.131*** [1.079]	−5.255*** [0.836]	−5.149*** [0.827]	−4.796*** [1.065]	−6.136*** [2.216]	−5.339*** [1.043]
Duration dependence
Ln(α)	0.076 [0.055]	0.080 [0.054]	0.046 [0.073]	0.042 [0.074]	0.376* [0.212]	0.566 [0.373]	0.162** [0.066]
Observations	5,976	5,975	5,861	5,595	5,861	5,861	5,861
Log-likelihood	−1,499.9	−1,486.8	−2,681.7	−2,671.3	−2,968.2	−4,124.3	−3,835.2
Sample	All	All	All	All	All	All	All
PSM-IPTW applied			Y	Y	Y	Y	Y
Sector and region dummies	Y	Y	Y	Y	Y	Y	Y

* Significant at 10%; ** 5%; *** 1%. Robust standard errors are clustered at industry and contract year levels and presented in the square brackets.

The columns 5–7 apply the PSM-IPTW method to distinguish contracts with guarantees (column 5), direct subsidies (column 6), and concessions (column 7). In the PSM-IPTW regression, the first-stage regression in Table 1 is used to estimate the propensity score to match the characteristics of the contract between treated (those with fiscal commitments) and control group. The technical appendix table A3 shows the balancing test which supports the validity of the PSM in matching contract characteristics of two groups.

The result shows that guaranteed contracts face significantly higher hazard rate than non-guaranteed contracts (equivalent to the hazard ratio of 4.3) (column 5). Higher hazard rate results from the amplification of fiscal risks due to inadequate market test on projects’ financial viability. The effect of direct subsidy is also positive but not statistically significant (column 6). Concessions tend to be riskier than greenfield investments as the demand risk is fully taken by private parties (column 7).

C. Robustness Checks in Alternative Specifications

Figure 5-1 compares the hazard ratio estimated under the baseline with results under alternative specifications: when (i) the sample is split by the maturity of PPP market (countries which have large representation (top 5 percent) in terms of the number of PPP contracts against less matured PPP countries), (ii) 1,954 contracts signed in LAC countries, one of the epicenters of disputes, are excluded,¹² and (iii) the sample is restricted to contracts signed in SSA. Figure 5-2 restricts the sample to user-pay contracts, or dependent variable is replaced with the government-initiated dispute cases.¹³

View Full Size

Change in Hazard Ratio: by Country Groups

Citation: IMF Working Papers 2017, 243; 10.5089/9781484328286.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Change in Hazard Ratio: by Contract and Dispute Types

Citation: IMF Working Papers 2017, 243; 10.5089/9781484328286.001.A001

Source: World bank PPI database* Significant at 10%; **5%; *** 1%. PSM-IPTW regression is applied with region and sector dummies. Robust standard errors are clustered at industry and contract year levels. The first bar shows the hazard ratio of the baseline case using the whole sample in table 2. The dependent variable is the hazard of all disputes except the last bar in Figure 5-2 which is only for the government-led dispute case. Hazard ratio of 1 (horizontal red line) means that the relative dispute risk is same between two types of contract.

• Download Figure
• Download figure as PowerPoint slide

The hazard ratios of guarantees and concessions remain above one and are significant in all specifications except the SSA sample (the upper panel of Figure 5-1). The hazard ratio of guarantee is significantly higher in matured PPP markets compared to less matured markets, and is slightly higher when LAC countries are excluded. In contrast, the hazard ratio of concession is significantly higher in less matured markets. In SSA countries, PPPs with guarantees are increasing but still not common (see Figure 4). The result suggests that guaranteed contracts face lower dispute risk while concessions are riskier in the SSA compared with the baseline, reflecting larger demand risk taken by private sector for business operations in SSA countries. The upper panel of Figure 5-2 shows that guaranteed contracts face slightly higher dispute risk when the sample is restricted to user-pay scheme, while the risk triples in case of the government-led disputes. This suggests that disputes for guaranteed contracts are mostly initiated by the government.

The lower panel in Figures 5-1 and 5-2 shows that the effects of MFI supports and democratic regime on the hazard ratios remain below one throughout all specifications, showing robust results in reducing disputes. The MFI support dummy mitigates the dispute risk in any sub-samples, but the effect is particularly large in matured PPP markets and when LAC countries are excluded. The negative effects also become significantly larger when the dispute type is restricted to the government-led disputes. This shows that the government-led dispute risk could be significantly lower when MFIs are involved due to higher reputation cost that the government must incur in breaching the contract. Democratic regime, as opposed to autocracy, could also significantly reduce the government-led disputes.

D. Role of Public Financial Management

Hypotheses 1 and 2 predict that stronger fiscal management capacity would increase the survival rate of contracts. To test this, a PEVI quality measure is constructed from three difference data sources. The main data come from the EIU’s Infrascope index, which is the weighted score of the quality of PPP regulations, the institutional framework, and the operational capacity in conducting PPPs. The Infrascope reports for Asia, CEE-CIS, LAC, and SSA regions currently cover 75 countries. The sample coverage is expanded by imputing missing observations using the IMF (2015)’s public investment management assessment (PIMA) index on PPPs and World Bank (2016)’s scores on PPP preparation, procurement, and contract management. The new index measures the PIM quality directly on PPPs (scaled from 0 (low) to 100 (high)), a more relevant measure than public investment management index (PIMI) (Dabla-Norris et al, 2012) in case of PPPs.¹⁴

As shown in Figure 6, the PPP PIM quality index (y-axis) and country’s budget transparency index (x-axis) are positively correlated. The left panel defines the budget transparency for the coverage of stock (i.e., reporting of liabilities, financial and nonfinancial assets in the government’s balance sheet), while the right panel refers to the coverage of flows (i.e., government’s operational statement). The positive correlation suggests that fiscally transparent countries tend to score higher on PPP’s contract management. In regional perspective, the PIM quality shows large variance especially in Latin America. The mean comparison t-test shows that budget transparency, both in stock and flow, is significantly lower in LICs (p-value=0.00).

View Full Size

Distribution of the Quality Scores on Fiscal Institutions

Citation: IMF Working Papers 2017, 243; 10.5089/9781484328286.001.A001

Note: The budget transparency index is the average of Wang et al (2015)’s score of the fiscal coverage of institutions and the reporting on stocks for the left figure, and the average score of the fiscal coverage of institutions and the reporting on Sows for the right figure.

• Download Figure
• Download figure as PowerPoint slide

Improvements in PIM quality and budget transparency would address the adverse selection and the accumulation of hidden debt respectively, contributing to the survival of PPPs. Table 3 presents the estimates of hazard regression with additional variables on a country’s overall PPP management quality (columns 1–2), operational maturity in PPPs (columns 3–4), and budget transparency in reporting fiscal stocks (columns 5–6).¹⁵ Both PPP’s PIM quality index and the budget transparency index are standardized around the mean.

Table 3.

Effect of Guarantees on Disputes: by the Quality of Fiscal Institutions

Note: Regression includes In (investment size), MFI supports, democracy, real GDP growth, and CPI inflation as controlled in the baseline specification. * Significant at 10%; ** 5%; *** 1%. Robust standard errors are clustered at industry and contract year levels and are reported in the square brackets.

^1/High group is for contracts signed in countries with above average overall PPP PIM quality, PPP operational quality, and budget transparency on stocks; while low group is for contracts signed in countries with below average scores.

Table 3.

Effect of Guarantees on Disputes: by the Quality of Fiscal Institutions

	(1)	(2)		(3)	(4)		(5)	(6)
Guarantee	1.569*** [0.585]	0.084 [0.604]	2.140*** [0.503]	1.482** [0.581]	0.302 [0.667]	2.101*** [0.490]	1.461** [0.576]	−0.210 [0.548]	1.924*** [0.605]
PPP overall PIM quality index	−0.516*** [0.106]	−3.076** [1.131]	−0.578*** [0.159]
PPP operational quality index				−0.350*** [0.126]	−0.869** [0.276]	−0.262** [0.127]
Budget transparency index (stock)							−0.161 [0.130]	0.054 [0.183]	−1.052*** [0.296]
Duration dependence
Ln(α)	0.372* [0.211]	0.297 [0.261]	0.543** [0.235]	0.368* [0.212]	0.279 [0.256]	0.531** [0.235]	0.358* [0.204]	0.314 [0.251]	0.585** [0.240]
Equality of coefficient F test (p-value)
Guarantee	…	11.86(0.001)***		…	7.68 (0.006)***		…	11.56(0.001)***
Observations	5,741	2,450	3,291	5,741	2,450	3,291	5,861	2,223	3,638
Sample 1/	All	High	Low	All	High	Low	All	High	Low
PSM-IPTW applied	Y	Y	Y	Y	Y	Y	Y	Y	Y
Sector and region dummies	Y	Y	Y	Y	Y	Y	Y	Y	Y

Note: Regression includes In (investment size), MFI supports, democracy, real GDP growth, and CPI inflation as controlled in the baseline specification. * Significant at 10%; ** 5%; *** 1%. Robust standard errors are clustered at industry and contract year levels and are reported in the square brackets.

^1/High group is for contracts signed in countries with above average overall PPP PIM quality, PPP operational quality, and budget transparency on stocks; while low group is for contracts signed in countries with below average scores.

View Table

Table 3.

Effect of Guarantees on Disputes: by the Quality of Fiscal Institutions

	(1)	(2)		(3)	(4)		(5)	(6)
Guarantee	1.569*** [0.585]	0.084 [0.604]	2.140*** [0.503]	1.482** [0.581]	0.302 [0.667]	2.101*** [0.490]	1.461** [0.576]	−0.210 [0.548]	1.924*** [0.605]
PPP overall PIM quality index	−0.516*** [0.106]	−3.076** [1.131]	−0.578*** [0.159]
PPP operational quality index				−0.350*** [0.126]	−0.869** [0.276]	−0.262** [0.127]
Budget transparency index (stock)							−0.161 [0.130]	0.054 [0.183]	−1.052*** [0.296]
Duration dependence
Ln(α)	0.372* [0.211]	0.297 [0.261]	0.543** [0.235]	0.368* [0.212]	0.279 [0.256]	0.531** [0.235]	0.358* [0.204]	0.314 [0.251]	0.585** [0.240]
Equality of coefficient F test (p-value)
Guarantee	…	11.86(0.001)***		…	7.68 (0.006)***		…	11.56(0.001)***
Observations	5,741	2,450	3,291	5,741	2,450	3,291	5,861	2,223	3,638
Sample 1/	All	High	Low	All	High	Low	All	High	Low
PSM-IPTW applied	Y	Y	Y	Y	Y	Y	Y	Y	Y
Sector and region dummies	Y	Y	Y	Y	Y	Y	Y	Y	Y

Note: Regression includes In (investment size), MFI supports, democracy, real GDP growth, and CPI inflation as controlled in the baseline specification. * Significant at 10%; ** 5%; *** 1%. Robust standard errors are clustered at industry and contract year levels and are reported in the square brackets.

^1/High group is for contracts signed in countries with above average overall PPP PIM quality, PPP operational quality, and budget transparency on stocks; while low group is for contracts signed in countries with below average scores.

Columns 1, 3 and 5 show that contracts that receive government guarantees face higher disputes risk, while higher PPP PIM quality contributes to significantly reduce disputes. In columns 2, 4, and 6, the sample is split into two groups: a group with above average scores on PIM quality or budget transparency vs. those below the average scores.

First, columns 2 and 4 show that the coefficient of guarantee dummy increases to 2.1 for lower PIM group, reflecting significant negative adverse selection effect of PPP contracts that increases disputes for low PIM group.¹⁶ Column 6 shows that the coefficient of guarantees rise to 1.9 for contracts signed in countries with non-transparent budget reporting system. This identifies pure contingent liability effect. The equality of coefficient F test confirms significant differences in coefficient of guarantees between two groups categorized by the PIM quality and budget transparency.¹⁷