Political risk, project finance, and the participation of development banks in syndicated lending

J. Finan. Intermediation 21 (2012) 287–314

Contents lists available at SciVerse ScienceDirect

J. Finan. Intermediation j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / j fi

Political risk, project finance, and the participation of development banks in syndicated lending Christa Hainz a,b,⇑, Stefanie Kleimeier c a Ifo Institute – Leibniz Institute for Economic Research at the University of Munich, Poschingerstrabe 5, 81679 Munich, Germany b CESifo, 81679 Munich, Germany c Maastricht University, Tongersestraat 53, 6211 LM Maastricht, The Netherlands

a r t i c l e

i n f o

Article history: Received 7 August 2011 Available online 19 October 2011 Keywords: Project finance Syndicated loans Political risk

a b s t r a c t How should loan contracts for financing projects in countries with high political risk be designed? We argue that non-recourse project finance loans and the participation of development banks in the loan syndicate help mitigate political risk. We test these arguments by conducting a study with a sample of 4978 loans made to borrowers in 64 countries. Our results show that if political risk is higher, then project finance loans are more likely to be used, and development banks are more likely to participate in the syndicate. We also show that the terms of the loan contract depend not only on the political risk but also on the legal and institutional environment as well. Ó 2011 Elsevier Inc. All rights reserved.

1. Introduction How should a company finance a project located in a country in which political risk is high and in which investor protection is weak? Such a project will only be realized if the risk can be reduced to a bearable level. Coasian bargaining theory and a growing number of empirical studies on law, institutions, and finance clearly show that loan contracts can be designed to mitigate deficits in the legal and institutional environment and thereby provide the project with access to finance. The following case illustrates how a loan contract can help mitigate political risk: The South African petrochemical group Sasol opts for a unique hybrid project finance structure to finance a gas field project in Mozambique. Under this hybrid structure, lenders initially have full recourse to Sasol, which assumes almost all project-related risks. The sole – but important – excep⇑ Corresponding author. Tel.: +49 89 9224 1237; fax: +49 89 9224 1462. E-mail addresses: [email protected] (C. Hainz), [email protected] (S. Kleimeier). 1042-9573/$ - see front matter Ó 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.jfi.2011.10.002

288

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

tion is the project’s political risk. The loan contract specifies that, if well-defined political risk events occur, the financing structure automatically changes from the full-recourse structure to a project finance structure. In this case lenders have recourse only to the project but no longer to Sasol. Cadwalader (2004), the project’s legal consultant, emphasizes the role of development banks in actively mitigating political risk: ‘‘Sasol would like to maximize the influence that the political risk providers [. . .] bring to the deal – their ability to exert political pressure on, in this case, the Mozambican government to prevent or cure a political risk event.’’ This case shows that there are two important features of the loan contract for managing political risk. First, the degree of recourse that the lender has helps manage political risk. In this respect, we discriminate between full-recourse loans and non-recourse project finance loans.1 Second, the participation of development banks in the loan syndicate also helps manage political risk. Development banks provide a so-called ‘‘political umbrella’’ because these banks can use their leverage to influence governmental decisions and deter adverse events that would negatively affect the project’s outcome.2 This paper analyzes the determinants of these two features of the loan contract, i.e., its recourse structure and the participation of a development bank in the loan syndicate. We base our tests on a sample of 4978 loans made to borrowers in 64 countries from 1996 to 2005. We run logit and multinomial logit regressions to explain how the two loan characteristics depend on political risk as well as the legal and institutional environment. The results clearly show that higher political risk increases the probability that project finance loans will be used and that development banks will participate in the syndicate. We also exploit major changes in political risk to validate the results of the cross-country regressions via a difference-in-difference approach. Our study demonstrates that the design of a loan contract is determined by not only the legal and institutional environment but also by political risk. Our study contributes to the growing literature on the effects of country-level risks on the design of loan contracts. Some studies investigate how law and institutions determine one particular feature of the contract. Esty and Megginson (2003) show that strong creditor rights and reliable legal enforcement are correlated with smaller and more concentrated syndicates. More recent studies emphasize the importance of studying more dimensions of the loan contract to capture its complexity. Qian and Strahan (2007) provide evidence suggesting that stronger creditor rights lead to a more concentrated syndicate structure, longer maturity, and lower interest rate. Bae and Goyal (2009) show that both creditor rights and the enforceability of a contract affect loan size, maturity, and interest rate. However, none of these papers explicitly address the role of political risk. There is evidence only for the rating and pricing of corporate bonds that political risk matters (Qi et al., 2010). As the Sasol case shows, political risk is critically important to the design of the loan contract, particularly in risky countries. Therefore, we contribute to the literature by investigating how political risk influences the choice between project finance and full-recourse loans and the syndicate structure of these loan contracts. Our paper is closely related to Subramanian and Tung’s study (2009). They show that project finance loans are more often used in countries with weak creditor rights and poor protection against insider stealing. In contrast, we investigate political risk in more detail while controlling for the legal and institutional environment. We explain not only the use of project finance loans but also the participation of development banks in the loan syndicate. This latter feature is not considered in the existing literature, with the exception of Esty and Megginson (2003). These researchers use the participation of development banks as a control variable while explaining the syndicate structure of project finance loans. Our study yields two main results. First, we show that projects in countries with higher levels of political risk are more likely to be structured as project finance loans and that development banks are more likely to participate in the syndicates. This finding suggests that the particular features of the project finance structure – i.e., separate incorporation, high leverage, and extensive contractual 1 In this study, we will use the terms ‘project finance’ and ‘non-recourse loan’ synonymously and in contrast with the term ‘fullrecourse loan’. 2 However, there are two arguments against a state-contingent financing structure, which may explain why this deal structure is not commonly used. First, how one can define specific political risk events ex ante is unclear. Second, the switching provision limits the incentives of banks to mitigate political risks in situations in which a predefined risk event occurs. Thus, the provision does not exploit the political umbrella of the development banks.

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

289

arrangements – and the ‘‘political umbrella’’ of the development banks reduce political risk. However, we discover several differences between the two risk management strategies. Our results show that the decision to use project finance is sensitive to political risk but that this sensitivity is the same across all levels of political risk. In contrast, the participation of development banks is sensitive to political risk, but this sensitivity is higher at higher levels of political risk. Second, we find evidence that the legal and institutional environment is relevant for the design of the loan contract. Specifically, the probability of using a project finance loan is lower if the legal provisions regarding corporate governance are better. Thus, project finance can, for example, reduce the risk that a manager expropriates funds from the firm. Moreover, project finance is more likely to be used as creditor rights improve. However, the participation of development banks in the syndicate depends on neither creditor rights nor corporate governance provisions. Our study proceeds as follows. In Section 2, we present our key country-level variables, i.e., political risk and law and institutions, and argue that the risks involved can be addressed by either a project finance structure or the participation of a development bank. In Section 3, we describe the data and methodology. In Section 4, we discuss the results and provide our robustness checks, and in Section 5, we conclude this paper. 2. The impact of political risk and law and institutions on the loan contract 2.1. Features of the loan contract If a firm applies for a loan, then a bank evaluates the riskiness of the to-be-financed project. Depending on the results, the bank may take different measures to reduce the risk that it must bear. As a result, each loan contract has many features that address the riskiness of the project to be financed. These risk management measures differ not only in nature but also in the way that they are related to the project’s riskiness. They can address the probability of default, the loss in the case of default, or both at the same time. One important decision that must be made is the choice between project finance and full-recourse lending. Additionally, it must be decided whether development banks participate in the loan syndicate. If the lenders use all of the risk management tools in an efficient manner, then the other features (i.e., the interest rate) will be adjusted such that the lenders are willing to bear the remaining risk and, therefore, grant the loan (Corielli et al., 2010; Nevitt and Fabozzi, 1995). 2.1.1. Project finance loans The parties involved in the loan may decide to structure it as a project finance loan. Such a structure has three important features. First, it involves the creation of a legally independent project company. This company is financed with equity from one or more of the sponsoring firms and with bank loans. The structure ensures that the liability is limited such that the lenders have no or only limited recourse to the sponsoring firms (Esty, 2004). In contrast, a traditional full-recourse loan allows the lender to access all of the assets of the project and the sponsoring firm. Second, the company is highly leveraged. Typically, its debt-to-equity ratio amounts to at least 70% (Esty and Sesia, 2009). Third, a project finance structure involves a number of contractual arrangements. For this reason, it is sometimes called ‘‘contractual finance’’ (Esty and Megginson, 2003). Through the contracts, specific project risks are allocated to those parties that can best manage them. The aim of the contract structure is to reduce the project risk and agency costs (Corielli et al., 2010; Brealey et al., 2000). One of the contracts is a ‘‘cash flow waterfall’’, which summarizes a project’s cash flows and assigns different priorities to each cash inflow and outflow (Buljevich and Park, 1999). By utilizing this contract, the lenders obtain access to and control over the company’s cash flows. Thus, the project company’s cash flows become fully verifiable (Subramanian and Tung, 2009). 2.1.2. Participation of development banks As Esty and Megginson (2003) show, the structure of the syndicate is used to manage risks. One particularly important member of the syndicate is a development bank. In contrast to commercial

290

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

banks, the development banks have a special status. Buiter and Fries (2002) argue that multilateral development banks’ ‘‘(. . .) support for private sector projects can be instrumental in mitigating risks associated with government policies and practices’’. Therefore, multilateral development banks are also known as political umbrellas (Buljevich and Park, 1999).3 Among the lenders, multilateral development banks such as the International Financial Corporation (IFC), a member of the World Bank Group, have high bargaining power derived from their involvement in the financing of a large number of projects, from their provision of financial aid, and from their frequent interactions with the government.4 Therefore, these two risk management strategies are instrumental in addressing political risk and other risks that arise on the country level.

2.2. Country-level risks 2.2.1. Political risk Political risk in the host country is an important factor that influences the probability that a loan will be serviced as scheduled. In general, political risk can be divided into three broad categories: traditional political risk, regulatory risk, and quasi-commercial risk (Smith, 1997). The traditional political risk category addresses risks related to expropriation, to the convertibility and transferability of currency, and to political violence. The regulatory risk category covers risks arising from unanticipated regulatory changes, such as changes in taxation or foreign investment laws. The quasi-commercial risk category encompasses the risks that arise if a project contends with state-owned suppliers or customers whose abilities or willingness to fulfill their contractual obligations towards the project may become questionable. This definition shows that political risk comprises a broad range of different risks. Some of these risks, such as restrictions on transferability, are easy to identify ex post, whereas others, such as changes in the tax law that may lead to creeping expropriation, are not.5 Moreover, risks such as expropriation or regulatory changes can be closely related to the corporate sector and even to individual firms. Other risks (e.g., corruption) apply to the entire society but can negatively impact an individual firm’s performance. We expect a project finance structure to help parties address political risk by rendering a government intervention less likely or by limiting the damage that such an intervention can cause to a project. Because of the separate incorporation of the project, government intervention that causes a project to fail becomes highly visible for financial market participants. This visibility increases the political costs of government intervention. Additionally, highly leveraged projects tend to have little cash such that the temptation for the government to expropriate the project is reduced (Esty, 2003). Finally, some contractual arrangements completely avoid government interventions that may affect a project. For instance, convertibility and transfer risks disappear if off-shore accounts are used (Buljevich and Park, 1999). The participation of development banks in a loan syndicate should also help mitigate political risk. Development banks provide a political umbrella because they frequently interact with the different bodies of a government. If an adverse government intervention jeopardizes the success of a project, then a development bank can use the leverage stemming from its special status and from its repeated interactions with the government. In this case, the government will be aware of the negative effect that an intervention will have on the loan (partially) granted by the development bank and, thus,

3 According to the World Bank’s definition, multilateral development banks are ‘‘institutions that provide financial support and professional advice for economic and social development activities in developing countries’’. Multilateral financial institutions also lend capital to developing countries but have a narrower country-membership structure and/or a focus on specific sectors or activities. 4 As an example of bank influence, consider the Russian A.O. Volga project, which was financed by Dresdner Bank Kleinwort Benson and the IFC. When this project suffered because of the Russian crisis and the moratorium in 1998, the bank’s influence manifested itself in that ‘‘the IFC umbrella regarding transfer and convertibility risk has remained effective, since IFC’s loans were explicitly exempted from the moratorium’’ (Lazarus, 2001, p. 119). 5 Actually, creeping expropriation has become more common, whereas total expropriation has become relatively rare (Esty, 2003).

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

291

on the government’s reputation at the development bank. Ultimately, the threat of reputational damage might deter government intervention.6 The International Country Risk Guide (ICRG) provides various measures of political risk, including bureaucracy quality, corruption, democratic accountability, ethnic tensions, external conflict, government stability, internal conflict, investment profile, law and order, military politics, religious tensions, and socioeconomic conditions. We use the ICRG’s rating of political risk related to investment profile as our main measure of political risk because this measure captures the risks that most affect firms. The ICRG’s rating is composed of the following subcategories: contract viability/expropriation, profit repatriation, and payment delays.7 2.2.2. Law and Institutions Prior studies have highlighted the importance of legal and institutional provisions to the design of loan contracts. We argue that these risks can be managed by the use of project finance. In the design of loan contracts, the characteristics of project finance (i.e., separate incorporation, high leverage, and extensive contractual arrangements) are instrumental. First, sponsors can exploit the fact that the project is incorporated separately and heavily financed with debt to commit to monitoring the project. In this way, the project manager’s incentives are aligned (Laux, 2001). Second, in contrast to a diversified firm, the cash flows in projects cannot be easily shifted between different divisions or diverted by the management. As cash flows become verifiable, the manager’s behavior no longer depends on the legal and institutional environment (Subramanian and Tung, 2009). Therefore, these two theoretical models predict that a project finance structure can influence incentives if the legal and institutional environment cannot shape them. Thus, project finance loans should be more often used in countries with poor legal systems. However, the non-recourse structure also implies that the lender only has access to the project’s assets in the case of a default. If these assets are poorly protected, then the lender’s payoff in this case will be low. This danger may render the use of a project finance loan unattractive. In contrast to their abilities to manage political risk, development banks are less adept at addressing problems in legal and institutional environments. Legal provisions in corporate law cannot be as easily changed as, for instance, a particular tax rate, and these changes often require a reliable legal infrastructure, such as well-functioning courts. Moreover, after the ‘‘law in the book’’, the implementation of the law is important. However, development banks may find influencing the implementation of the law to be even more difficult from outside the country because the law is implemented by the judiciary, which may be independent. In our analysis, we focus on three aspects of national laws and institutions. We first analyze a country’s corporate governance system, which shapes the framework in which a firm operates. We measure this system by using Djankov et al.’s (2008) index of ex-post control over self-dealing transactions. This index reflects the degree to which minority shareholders are legally protected against expropriation from corporate insiders. Additionally, the index measures the disclosure or approval of transactions and the scope of private litigation or public enforcement. Self-dealing includes ‘‘executive perquisites, excessive compensation, transfer pricing, appropriation of corporate opportunities, self-serving financial transactions such as directed equity issuance or personal loans to insiders, and outright theft of corporate assets’’ (Djankov et al., 2008, p. 431). Overall, we expect corporate governance to be inversely related to the use of project finance. Second, we analyze a country’s creditor rights because they specify what happens in the case of a default. We measure the strength of creditor rights by using Djankov et al.’s (2007) index of aggregate creditor rights. Following La Porta et al. (1998), Djankov et al. (2007, p. 303) described their index as follows: ‘‘A score of one is assigned when each of the following rights of secured lenders is defined in laws and regulations: First, there are restrictions, such as creditor consent or minimum dividends, for a debtor to file for reorganization. Second, secured creditors are able to seize their collateral after the 6 Esty (2003, p. 9) supports this argument in the following: ‘‘The historical record shows that governments gave higher priority to obligations from multilateral lending institutions such as the World Bank Group when they could not service all of their external debt. One of the main reasons why they have received preferential treatment was because they are often the only source of new lending to countries in financial distress’’. 7 See Table A.1 in the appendix for details regarding our independent variables, dependent variables, and data sources.

292

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

reorganization petition is approved, i.e., there is no automatic stay or asset freeze. Third, secured creditors are paid first out of the proceeds of liquidating a bankrupt firm, as opposed to other creditors such as government or workers. Finally, if management does not retain administration of its property pending the resolution of the reorganization.’’ We expect weaker creditor rights to be associated with more project finance. Finally, we also measure the efficiency of the legal system using the time needed to resolve a case of insolvency from the World Development Indicators (WDI). In general, the longer a bankruptcy process lasts, the weaker the creditors’ positions become. More specifically, non-recourse project finance insulates a sponsor in the case of a default and might be beneficial if the bankruptcy process is inefficient (i.e., lasts for a long time). Therefore, we expect a slower bankruptcy process to be positively related to the use of project finance. 2.2.3. Other country-level control variables In addition to political risk, corporate governance, and creditor rights, other country features explain the workings of the financial system and may, therefore, affect the loan structure. For example, several studies on the determinants of loan contract design consider various aspects of the rule of law, shareholder rights, country risk, corruption, and financial or economic development (Bae and Goyal, 2009; Esty and Megginson, 2003; Qian and Strahan, 2007; Subramanian and Tung, 2009). Including these or other country-level variables can improve the empirical model. However, including many country-level variables can introduce multicollinearity, which in turn leads to model instability. To find a balance between these two aspects, we consider the following four country-level variables. First, we use Euromoney’s ‘‘Economic Performance Index’’ as a proxy for the economic development of the borrower’s country. This annual index is based on the country’s current GDP per capita figures and on a poll of economic projections. Therefore, the index contains both current and forward-looking information, which are especially useful when considering the medium- to long-term natures of the loans. Second, we obtain two proxies for financial development from the WDI. These proxies consist of the domestic credit provided by the banking sector as a percentage of GDP and the proportion of adults covered by a private credit bureau. Because we focus on loans, proxies based on the banking market better capture the financial developments than proxies based on the stock market. Similarly, we prefer our proxy for domestic bank credit over, for example, the alternative proxy for domestic credit to the private sector because our proxy measures the loans made by banks instead of loans made by all types of financial institutions. Moreover, our proxy captures different types of borrowers. This feature is useful because different types of borrowers emerge in our sample, where infrastructure projects are often associated with the public sector. Third, we want to control for loan demand by following Berger and Hannan (1989) and Neuberger and Zimmermann (1990), who propose using the annual growth rate of a banking product as a proxy for demand and supply conditions in this market segment. Therefore, we measure the annual growth rate of bank lending based on the WDI’s volume of domestic credit (in terms of local currency) provided by the banking sector. Finally, we include dummies for the country’s legal origin (i.e., British, French, German, and Scandinavian) in accordance with La Porta et al. (1999, 2008). La Porta et al. (2008) show that legal origin influences a country’s regulations, institutions, and economic outcomes, such as procedural formalism, corruption, unemployment, stock market development or private credit. In accordance with Qian and Strahan (2007), we include legal origin dummies as broad proxies for each country’s characteristics. 3. Data and methodology 3.1. Data We obtain our sample from the Dealscan database, which provides a comprehensive record of syndicated loan transactions around the world.8 From the Dealscan database, we collect a sample 8 Our results might be biased if the sample selection is correlated with political risk. For instance, if the coverage of either project finance loans or development bank participation in low-risk countries is higher, then the results of this study will be affected. We address this issue by exploiting major changes in political risk in a within-country identification strategy.

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

293

of asset-based loans signed between January 1, 1996, and December 31, 2005. Dealscan classifies loans in accordance with their purpose (e.g., trade finance, working capital, and general corporate purposes) and type (e.g., term loan, bridge loan, and credit line) and includes information about the size of each loan as well as the overall size of all of the loans associated with the same deal. We differentiate between project finance loans and full-recourse loans based on the loan purpose. To ensure that we focus on investments in which the firm can actually choose between a project finance loan and a fullrecourse loan, we only consider asset-based loans whose purposes are ‘‘project finance’’ and contrast these loans with loans whose purposes are ‘‘capital expenditure’’, ‘‘equipment purchase’’, ‘‘telecom build-out’’, ‘‘aircraft and ship finance’’, ‘‘leasing’’, ‘‘real estate’’ or ‘‘corporate purposes’’. Loans designed for the latter purpose might not necessarily be based on assets. Therefore, we impose two additional requirements for these loans. First, these loans must be of a sufficient size. That is, they must be larger than the smallest project finance loan. Second, the deal to which this loan belongs must include at least one term loan.9 Dealscan also contains detailed information about the loan syndicate. Specifically, the Dealscan database includes the names of the lenders in the syndicate. We use this information to identify whether a development bank participates in a deal. We classify lenders as development banks in accordance with the World Bank’s definition of multilateral development banks and multilateral financial institutions. However, we also consider prominent national development banks, such as export–import banks. We include these lenders if they have substantial shares in the syndicated loan market and might, therefore, have substantial influence over a host government. In particular, we select national development banks that funded asset-based loans worth at least $500 million in real terms from 1991 to 2005. As outlined above, we explain the choice of project finance and the participation of development banks by referring to each country’s characteristics. The results of Doidge et al.’s study (2007) suggest that national characteristics explain much more (i.e., 39–73%) of the variance in governance ratings than observed firm characteristics (i.e., 4–22%). Thus, our use of national characteristics is justified. Although borrower-specific proxies (e.g., the proxies related to corporate governance) might be preferable, we can only obtain some corporate governance provisions on the firm level and only for fullrecourse loans. The borrower of a project finance loan is a newly established company whose financial statements or other records are unavailable because project finance companies in general are not publicly listed. Nevertheless, we control for investment characteristics. Based on the Dealscan record of each loan, we use the total size of all of the loans that finance a given deal (in millions of real 2005 US$) as an indicator of the size of the investment. Similarly, we measure the life of the investment by using the maximum maturity in years among all of the loans that finance a given asset. To capture some additional characteristics of borrowers in an indirect manner, we apply a two-stage regression approach similar to Esty and Megginson (2003). First, we regress the loan spread on the loan features that might influence the spread (i.e., loan size, loan maturity, and dummies indicating whether the loan is priced over LIBOR, is guaranteed, has financial or general covenants, or is denominated in a currency different from the borrower’s home currency) and on our proxies for countries, investments, and alternative explanations described below. The error term of this regression reflects the unexplained, residual borrower risk. This risk is uncorrelated with the other independent variables and can be included as an indirect measure of the borrowers’ characteristics. Finally, we examine alternative explanations for the use of project finance loans and the participation of development banks. First, firms may face a debt-overhang problem (Myers, 1977). These firms can solve this problem by utilizing a project finance structure that allows them to move the project’s assets and debt off the balance sheet. Therefore, we control for the leverage ratio at the industry level, which we measure by calculating the median long-term debt to total assets ratio in the borrower’s industry. We derive this ratio at the four-digit industry-level using Compustat Global data from 1996 to 2005. Second, the use of project finance involves both an asset and a financing choice (Esty, 2003). Because the asset choice is industry-specific, we control for the borrower’s industry. Based on the borrower’s 2-digit SIC code, we create five industry groups: mining including oil and gas

9

Subramanian and Tung (2009) use a similar approach.

294

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

(10–14), construction (15–19), manufacturing (20–39), transportation and utilities (40–49), and ‘other’ for the remaining SIC codes. The ‘other’ group encompasses agriculture, trade, financial services, and public administration. Overall, we have complete information for 4978 asset-based loans signed by borrowers from 64 countries (excluding the US) from 1996 to 2005.10 Of these loans, 1549 are project finance loans and 3429 are full-recourse loans. Development banks are members of the lending syndicate for 545 of these loans. 3.2. Methodology We first model the loan and bank choice in a reduced form by estimating two separate logit regressions. We define our dependent variables as 0/1 dummies. The loan dummy takes a value of one for a project finance loan and zero for a full-recourse asset-based loan. The bank dummy takes a value of one if a development bank participates and zero otherwise. Because many of our proxies vary only by the country and year levels and not by the loan level, we cluster the standard errors at the country-year level. Doing so results in more conservative estimates of the significance of the independent variables. In accordance with Qian and Strahan (2007), we recognize that the loan and bank choice might be jointly made. In this case, our regressions are not pure, reduced forms. Next, we explicitly consider the situation in which the loan and bank choice are jointly made and estimate a multinomial logit model. Instead of defining two separate dummies, we now consider an indicator that assumes a value of one for a project finance loan in which a development bank participates in the loan syndicate, a value of two for a project finance loan in which no development banks participate, a value of three for a fullrecourse loan in which a development bank participates, and a value of four for a full-recourse loan in which no development bank participates. Although the process by which we interpret the regression coefficients is more complex, the multinomial logit model can account for situations in which the loan and bank choice are jointly made. Both the logit and multinomial logit models investigate cross-country differences in political risk. However, in cross-country studies, unobserved variables may be correlated with political risk. Therefore, the coefficient of political risk may be biased. Moreover, if the sample selection is biased with respect to political risk, then the coefficient of political risk is biased as well. To address these concerns, we employ a within-country identification strategy. We investigate the effect of a substantial change in political risk on the use of project finance loans and the participation of development banks in the loan syndicate.11 A substantial change in political risk is based on a proxy for the involvement of the military in the country’s government. An increasingly involved military increases the probability that the decisions taken by the government (e.g., increasing tax rates or changing the exchange rate regime) negatively impact investors.12 We include the country dummies and country trends to control for the time-invariant and time-variant unobserved country characteristics. These proxies also absorb the average level of sample selection biases within a country or their country-specific variations over time. In the next section, we first present the results of the logit model with respect to the decisions made regarding the project finance loan and the development banks’ participation in the loan syndicate. Next, we consider the jointness of the financing structure decision by employing a multinomial logit model. We address the endogeneity concerns and sample selection biases by applying a withincountry identification strategy, which confirms our cross-sectional results. Finally, we perform a set of robustness checks for the cross-country identification strategy.

10 In accordance with the literature, we exclude loans to US borrowers (see Qian and Strahan, 2007). Including US borrowers yields similar results that are discussed in Section 4.5 on robustness. 11 Other recent finance papers (e.g., Djankov et al., 2007; Subramanian and Tung, 2009) employ a similar difference-indifference strategy to identify changes in creditor rights and show that this within-country identification strategy allows the scholars to interpret the relationships as causal in nature. 12 We thank the referee for identifying these problems, for suggesting the within-country identification strategy and for pointing us towards a proxy of political risk that is based on a change in government.

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

295

4. Results 4.1. The use of project finance and the participation of development banks From January 1996 to December 2005, companies raised $982035 million in real terms in the global syndicated loan market to finance their asset-based projects. As Panel A of Table 1 shows, approximately one-third of these funds were project finance loans that were worth $287037 million on a real basis. The borrowers in the industrialized countries raised the most asset-based syndicated loans. Western Europe and Canada alone accounted for 56% of the total volume of loans. Looking at each country’s project finance share, we find that this financing structure is the least important in Western Europe but the most important to the borrowers from Africa and the Middle East. With regard to the development banks, Panel B shows that the development banks participated in 545 asset-based loans that were worth $124870. Approximately 85% of these asset-based loans were funded by national development banks. The Kreditanstalt für Wiederaufbau (KfW) and the Korea Development Bank (KDB) were the most active lenders. The most active multilateral financial institution was the European Investment Bank, which participated in 31 loans worth $9550 million. Among the multilateral development banks, the European Bank for Reconstruction and Development (EBRD, 41 loans worth $5598 million) and the World Bank (55 loans worth $3873 million) dominated the loan markets. Table 2 provides a closer look at the loan and bank choice in relation to the characteristics of the borrowers’ countries and investments. Panel A shows that in comparison with full-recourse loans, project finance loans fund assets in countries that are politically riskier and have weaker corporate governance systems. Project finance is associated with stronger creditor rights. Moreover, the use of project finance increases as the duration of the insolvency resolution process increases, the banking markets deteriorate, and the credit bureaus’ coverage improves. Project finance is also more widely used if a country’s economic performance is weak and loan growth is strong. Project finance loans are more prominent in countries with a French legal origin, as shown in part by the importance of project finance for African countries (Table 1). In contrast to other lenders, development banks are associated with investments in countries that have higher levels of political risk and weaker corporate governance. These banks are also more likely to participate in countries with lengthier insolvency resolution processes, poorer economic performances, less developed banking markets, and stronger growth in loans. Our findings also show that development banks lend more capital to countries of German legal origin.13 This trend is mostly driven by the KDB. A total of 80% of the KDB’s loans fund investments in countries with German legal origin, including South Korea, Taiwan, and China. To a lesser extent, the EBRD (44%) and KfW (65%) also impact these countries, as these banks have a geographic focus on Europe. We confirm these patterns when we examine the combination of loan and bank choice. In comparison with the three other financing choices, full-recourse loans without development banks fund investments in countries that have stronger governance, equal or poorer creditor rights, political stability, shorter insolvency resolution processes, stronger or equal economic performances, larger banking markets, and weaker loan growth. With 3135 loans, full-recourse loans without development banks account for the largest type of financing in our sample. The 545 loans with development banks are almost equally split into project finance and full-recourse loans. This finding indicates that as a group, development banks have no preferences for project finance loans over full-recourse loans or vice versa. With regard to investment features, Panel B shows that both project finance loans and the participation of development banks are associated with larger and longer-term investments, with many loans given to the transportation sector and utilities. Project finance loans are more frequent in the mining and construction industries but less frequent in the manufacturing sector. The evidence regarding the debt overhang problem is mixed. Development banks are associated with high-leverage industries, whereas project finance loans are marginally associated with low-leverage industries. 13 On the one hand, British common law is more oriented towards market institutions. On the other hand, French civil law is more oriented towards state intervention. German and Scandinavian civil law exist in between, though they are somewhat less oriented toward state intervention. For details, see Beck et al. (2003).

296

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Table 1 Project finance and full-recourse loans. Borrower location

Loan volume (real $ mio)

Number of loans

Project finance

Project finance

Full-recourse

Full-recourse

Panel A: The geographic distribution of asset-based lending Western Europe 129102 United Kingdom 40005 Spain 26819 Italy 24987 France 5922 Germany 4874 The Netherlands 2895 Greece 3995 Finland 5506

383405 115143 57079 46957 52664 40265 20952 10336 4837

550 169 166 74 18 17 16 15 5

1159 309 237 73 164 82 76 65 16

Asia Hong Kong South Korea Australia Japan China Singapore Indonesia Malaysia India Philippines

85407 14594 13179 17480 237 6525 4300 11889 4673 2108 4257

153510 28550 24540 14797 25034 12461 12956 3321 8843 10568 6208

651 64 98 103 2 84 35 107 30 25 54

1371 215 248 90 192 144 123 45 69 104 83

The Americas and Carribean Canada Mexico Chile Brazil Argentina

37293 5823 10096 3566 5292 3479

119590 37806 29843 21632 11115 10195

143 17 30 14 25 19

656 212 152 96 57 82

Eastern Europe and Russia Russia Poland

18541 7221 4535

29814 10156 8162

110 24 33

180 53 35

Middle East Turkey

13991 7999

5266 2529

78 48

39 18

2704 374 1131 827

3413 2313 103 232

17 6 3 4

24 16 1 1

287037

694998

1549

Africa South Africa Nigeria Morocco Global

3429 (continued on next page)

Overall, the observations from Tables 1 and 2 provide the first evidence that the management of political risk and the risks stemming from the legal and institutional environment influence the preference for project finance loans and the participation of development banks. There is substantial variation in political risk, corporate governance, and creditor rights, even though these proxies only vary across the countries and years.14 To illustrate this point, we note that in 2004, Thailand reflected the average level of political risk (0.23), corporate governance (0.63), and creditor rights (2.00) relatively well. We find one standard deviation more than the average level of political risk in Sri Lanka in 2005 and Pakistan in 1997, whereas the Netherlands in 2001 and New Zealand in 2002 exhibited one standard deviation less than the average political risk. We find the highest level of political risk in Colombia, followed by Venezuela. Compared with Thailand in 2004, the countries one standard deviation away from the average were the following: the UK in 2001, which had better corporate governance; the Philippines in 2001, which had weaker corporate governance; Australia in 1999, 14

Table A.2 in the appendix provides descriptive statistics for our explanatory variables for all 4978 asset-based loans.

297

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314 Table 1 (continued) Development banks Panel B: Lending by development banks All development banks Individual development banks: Kreditanstalt für Wiederaufbau (KfW) Korea Development Bank (KDB) Export Development Canada (EDC) European Investment Bank Japan Bank for International Cooperation European Bank for Reconstruction & Development (EBRD) Nordic Investment Bank World Bank (WB, incl. IBRD and IFC) Export–Import Bank of Korea Corporacion Andina de Fomento Export–Import Bank of the Republic of China Asian Development Bank China Development Bank Inter-American Development Bank Islamic Development Bank

Category

NDB NDB NDB MFI NDB MDB MFI MDB NDB MDB (subregional) NDB MDB NDB MDB MFI

Loan volume (real $ mio)

Number of loans

124870

545

51001 30113 15512 9550 8901 5599 3976 3874 2437 1906 987 828 713 345 216

129 184 59 31 27 41 23 55 21 9 4 10 6 5 2

The table reports descriptive statistics for 4978 asset-based loans signed between January 1, 1996, and December 31, 2005. We report number of loans and loan volumes in terms of real 2005 US dollars. Panel A shows the total lending volumes by region and for individual countries with at least $10 billion in total loan volume. For African countries we report the three countries with the highest loan volume. Panel B lists the total syndicated lending activities of multilateral development banks (MDBs), multilateral financial institutions (MFIs), and those national development banks (NDBs) with more than $500 million in funding. We assign the full amount of the loan to each development bank.

which had better creditor rights; and the Philippines in 2004, which had weaker creditor rights. Therefore, when interpreting our regression results in the next section, we will focus not only on the marginal effect but also show the effect that a change of one standard deviation in the explanatory variable has on the likelihood of an investment being financed with project finance and/or development banks. In this way, the effects of our proxies become comparable, and their economic importance becomes more visible. 4.2. The relevance of political risk as well as law and institutions We argue above that project finance is more valuable if the political risk is higher and if the law and institutions are weaker. We also expect the development banks to be more likely to participate if the political risk is higher. The results in Tables 3 and 4 confirm these expectations. The correlation between project finance and political risk and our other explanatory proxies is found in Table 3. We first run a regression including only political risk. We then add all of the characteristics related to countries, industries, and investments in regression (2). Because the correlation between these proxies is substantial in some cases and because we want to minimize the problem of multicollinearity, we derive our preferred specification in regression (3). This specification includes only the country-, industry-, and investment-related characteristics that are significant in either the project finance or the development bank regressions.15 The marginal effect of political risk is significant, positive, and stable across these three specifications. This finding indicates that our results are not driven by multicollinearity. In our preferred specification (3), the predicted probability of observing a project finance loan is 27.1% for the average asset-based loan. This probability is close to the sample frequency of 31.1%.16 15 We select the same preferred model for both the project finance and development bank analysis because we want to use a common specification in the multinomial regression analysis. Note that we consider including the proxy for private credit bureau coverage because this proxy is marginally significant in regression (2) of Table 4. However, when included in regression (3) of Table 4, this proxy is no longer significant. Therefore, we exclude the proxy from our preferred specification. 16 1P ^ are the ^ and b ^¼ , where a We calculate the probability of a loan being a project finance loan by using the equation p ^ ^þ  a bZ Þ i i 1þe ð estimated coefficients of the intercept and independent variables Zi of regression (3).

298

Number of loans

Panel A: Country characteristics All loans 4978 Project finance 1549 loans Full-recourse loans 3429 Loans with DB 545 Loans without DB 4433 Project finance 251 loans with DB Project finance 1298 loans without DB Full-recourse loans 294 with DB Full-recourse loans 3135 without DB

Average country characteristics

Fraction of sample with legal origin (%)

Political risk

Corporate governance

Creditor rights

Time to resolve insolvency

Economic performance

Banking market development

Private credit bureau coverage

Annual loan growth

French

German

Scandinavian

British

0.19 0.22***

0.63 0.60***

2.20 2.36***

2.32 2.49***

0.62 0.61***

1.10 0.99***

0.06 0.07***

0.13 0.15***

41 46***

22 19***

2 1

35 34

0.18 0.27*** 0.18 0.26***

0.64 0.54*** 0.64 0.51***

2.13 2.16 2.21 2.06

2.25 2.92*** 2.25 3.05***

0.63 0.56*** 0.63 0.54***

1.15 0.81*** 1.14 0.82***

0.06 0.04*** 0.07 0.05

0.11 0.15*** 0.12 0.15***

39 44 41 50***

23 41*** 19 35***

2 1*** 2 0***

36 14*** 38 15***

0.22***

0.62***

2.42***

2.39***

0.62

1.02***

0.08***

0.16***

45***

16***

2

38

***

***

***

***

***

***

1

14***

2

38

0.28

0.57

2.24

0.17

0.65

2.12

*

2.81 2.19

***

***

0.58

0.80

0.04

0.16

39

46

0.63

1.19

0.06

0.11

39

21

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Table 2 Descriptive statistics for different financing structures.

Number of loans

Average investment characteristics Investment size ($ mio real)

Investment risk

Fraction of sample in industry (%)

Industry’s leverage ratio

Mining

Transportation and utilities

Manufacturing

Construction

Other

0.13 0.12**

7 10***

38 44***

28 17***

6 14***

21 15***

0.13 0.15*** 0.12 0.12

6 6 7 9*

35 55*** 35 48***

33 23*** 29 21***

2 8** 6 15***

24 8*** 23 7***

7.5 11.1***

0.000 0.004*

5.9 11.1*** 7.1 13.4***

0.002 0.006 0.001 0.014*

10.7***

0.003

0.12

10***

43***

16***

14***

17***

9.2***

0.000

0.18***

4

61***

25***

3

8***

0.12

6

32

34

2

25

5.6

0.002

This table describes country, industry and investment characteristics in relationship to the financing choice of loan and bank type. We differentiate asset-based loans as project finance or full-recourse loans and consider whether one or more development banks (DB) are part of the syndicate. We use a two-sided t-test to test for differences in mean between project finance and full-recourse loans, between loans with versus without development banks, and between full-recourse loans without development banks and the other three financing choices. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level.

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Panel B: Investment and industry characteristics All loans 4978 407.74 Project finance 1549 422.95 loans Full-recourse loans 3429 400.87 Loans with DB 545 456.12* Loans without DB 4433 401.80 Project finance 251 459.04* loans with DB Project finance 1298 415.97 loans without DB Full-recourse loans 294 453.62 with DB Full-recourse loans 3135 395.93 without DB

Investment life (years)

Average debt overhang problem

299

300

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Table 3 The determinants of the loan contract. Dependent variable

Project finance dummy

Regression

(1)

(2)

(3)

0.32*** (3.02)

0.36*** (2.96) 0.20* (1.79) 0.06*** (3.76) 0.00 (0.60) 0.07 (0.57) 0.06 (1.51) 0.02 (0.19) 0.23** (2.41) 0.01 (0.11) 0.05 (0.87) 0.00 (0.00)

0.34*** (2.99) 0.21** (2.18) 0.06*** (3.89)

0.83*** (3.36)

0.82*** (3.33)

0.16*** (2.77) 0.11*** (2.80) 0.47*** (9.18) 0.02 (0.45)

0.16*** (2.78) 0.11*** (2.80) 0.47*** (9.29) 0.02 (0.47)

0.01 (0.87) 0.04*** (8.12) 0.16* (1.79)

0.01 (0.95) 0.04*** (8.17) 0.16* (1.79)

Panel A: Regression results Country characteristics Political risk Corporate governance Creditor rights Time to resolve insolvency Economic performance Banking market development Private credit bureau coverage Annual loan growth French legal origin German legal origin Scandinavian legal origin Debt-overhang problem in industry Leverage ratio Industry dummies Mining Transportation and utilities Construction Other Investment characteristics Ln(investment size) Investment life Investment risk Observations Log likelihood Pseudo R2

4978 3050.0 0.012

4978 2294.8 0.257

0.05 (0.41) 0.06 (1.55)

0.23** (2.44) 0.00 (0.01) 0.05 (0.89)

4978 2295.4 0.256 (continued on next page)

301

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Table 3 (continued) Predicted probability of project finance (%) From Panel B: Discrete changes in the financing choice’s predicted probability One-standard deviation change in Political risk 24.2 Corporate governance 29.7 Creditor rights 23.7 Predicted probability for average loan observation 27.1 Sample frequency 31.1

To

Difference

30.2 24.6 30.7

6.0 5.0 7.0

This table shows logit regression results. The dependent variable takes the value of 1 for a project finance loan and 0 for a fullrecourse loan. Panel A displays marginal effects calculated at the sample means and, below, the z-statistics. Standard errors are heteroskedasticity robust and clustered at the country-year level. The regressions contain intercepts. Panel B reports predicted probabilities based on regression (3) of Panel A. The predicted probabilities are calculated when the independent variable is valued at a level of half a standard deviation below the mean (reported in column ‘‘from’’) compared to half a standard deviation above its mean (reported in column ‘‘to’’). The difference expresses the change in probability for this one standard deviation change. All other independent variables are held at their mean values. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level.

For this average loan, the marginal effect of an increase in political risk is 0.34, whereas a discrete increase of one standard deviation in political risk is associated with an increase of six percentage points in the probability of a loan being funded with project finance.17 Both measures indicate that political risk has an economically significant effect. Among the other country characteristics, the results imply that the use of project finance is negatively correlated with better corporate governance and positively correlated with creditor rights. An improvement of one standard deviation in corporate governance and creditor rights is associated with changes in the predicted probabilities of project finance of 5% and +7% points, respectively, for the average observation. The higher likelihood of a project finance loan being made under conditions of poor corporate governance is consistent with the argument that the contractual framework of project finance can serve as a substitute for weak legal environments. In contrast, we find a positive correlation between project finance and stronger creditor rights.18 Creditor rights determine the outcomes of a bankruptcy. For a project finance loan, the creditor’s payoff in the case of a bankruptcy stems only from the project. Therefore, in a country with weak creditor rights, this payoff may be negligible. As a result, the incentive to use a project finance loan may increase as creditor rights improve. Moreover, we find that the use of project finance loans is correlated with the growth of loans but is unrelated to the amount of time required to resolve a bankruptcy, the economic performance, the degree to which the banking market is developed or our broad indicators of legal origin. In contrast to our expectations, we find a negative relationship between the industry’s leverage ratio and the use of project finance. Thus, project finance does not seem to help reduce any debt-overhang problems. We interpret the negative marginal effect as follows. There are projects in which high leverage is desirable because of agency or tax reasons (Esty, 2003). However, an existing firm may not be able to leverage itself to a significant degree. In industries with low leverage, a firm can only attain high leverage by moving its assets off its balance sheet. It may do so by using project finance. However, if the sponsoring firm’s leverage is already high, then the assets can be financed on the balance sheet. Hence, the firm can avoid incurring the expensive setup costs of a project finance structure.

17 More specifically, we consider an increase in political risk of one standard deviation (i.e., from 0.11 to 0.28) by measuring an increase from a level half a standard deviation below the mean to a level half a standard deviation above the mean. In this case, the probability of a loan being financed as a project finance loan increases by 6.0% (i.e., from 24.2% to 30.2%). All of the other variables are held at their mean values. Panel B of Table 3 reports these detailed results. Throughout the paper, we apply this change of one standard deviation centered on the mean to continuous independent variables when we report the discrete changes in the predicted probabilities. 18 Note that Subramanian and Tung (2009) find a negative coefficient in their sample, which, unlike ours, includes US borrowers.

302

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Table 4 The determinants of development bank participation. Dependent variable

Development bank dummy

Regression

(1)

(2)

0.25*** (4.97)

0.10*** (3.55) 0.04 (1.08) 0.01 (1.55) 0.00 (0.11) 0.09*** (2.75) 0.06*** (4.43) 0.07* (1.64) 0.03 (1.11) 0.07*** (2.64) 0.17*** (3.64) 0.03 (0.53)

Panel A: Regression results Country characteristics Political risk Corporate governance Creditor rights Time to resolve insolvency Economic performance Banking market development Private credit bureau coverage Annual loan growth French legal origin German legal origin Scandinavian legal origin Debt-overhang problem in industry Leverage ratio

0.00 (0.05)

Industry dummies Mining Transportation & utilities Construction Other Investment characteristics Ln(investment size) Investment life Investment risk Observations Log likelihood Pseudo R2

4978 1658.9 0.035

(3)

0.11*** (3.82) 0.02 (0.69) 0.01 (1.46)

0.08*** (2.61) 0.06*** (4.89)

0.03 (1.01) 0.05*** (2.47) 0.16*** (3.70)

0.00 (0.05)

0.02* (1.65) 0.00 (0.28) 0.00 (0.24) 0.04*** (3.97)

0.02** (1.96) 0.00 (0.14) 0.00 (0.07) 0.04*** (3.96)

0.02*** (4.08) 0.01*** (6.36) 0.02 (0.81)

0.02*** (4.18) 0.01*** (6.19) 0.02 (0.79)

4978 1334.6 0.224

4978 1338.7 0.222 (continued on next page)

303

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314 Table 4 (continued) Predicted probability of development bank participation (%) From Panel B: Discrete changes in the financing choice’s predicted probability One-standard deviation change in Political risk 4.7 Corporate governance 5.3 Creditor rights 5.1 Predicted probability for average loan observation 5.5 Sample frequency 10.9

To

Difference

6.5 5.8 6.0

1.8 0.5 0.9

This table shows logit regression results. The dependent variable takes the value of 1 when a development bank participates in the loan syndicate and 0 otherwise. Panel A displays marginal effects calculated at the sample means and, below, the zstatistics. Standard errors are heteroskedasticity robust and clustered at the country-year level. The regressions contain intercepts. Panel B reports predicted probabilities based on regression (3) of Panel A. The predicted probabilities are calculated when the independent variable is valued at a level of half a standard deviation below the mean (reported in column ‘‘from’’) compared to half a standard deviation above its mean (reported in column ‘‘to’’). The difference expresses the change in probability for this one standard deviation change. All other independent variables are held at their mean values. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level.

Regarding the industry dummies and project-specific control variables, we find that borrowers in the mining and construction sectors strongly prefer to utilize project finance loans. This finding explains the high project finance shares in Australia (50%), Nigeria (75%), and the Middle East (67%). In all of these countries, project finance loans are used substantially more than in other countries or regions. Furthermore, borrowers are more likely to use project finance to make riskier investments with longer life spans. This pattern is consistent with Kleimeier and Megginson (2000), who report that project finance loans have a median size similar to the size of fixed asset-based loans but have longer maturity dates. Table 4 presents the results regarding the participation of development banks. Again, the results are stable across the three specifications, and regression (3) presents our preferred model. For the average asset-based loan, the predicted probability of having a development bank in the syndicate is 5.5%, which is somewhat lower than the sample frequency of 10.9%. From the country characteristics included in regression (3), we learn that the development banks are more likely to lend capital to projects in countries with higher levels of political risk. This relation is strong. An increase of one standard deviation increases the predicted probability that a development bank will participate in the syndicate from 4.7% to 6.5% for the average loan. We interpret this result as a clear indication that development banks provide a ‘‘political umbrella’’. However, the development banks’ decisions to participate are not related to the corporate governance provisions or the creditor rights. Development banks are more likely to participate in countries with weaker economic performances, weaker banking market development and in countries of French or German legal origin. For the loan-level characteristics, we observe that development banks focus less on the mining industry and other industry sectors, including agriculture, trade, financial services, and public administration. Furthermore, development banks are more associated with larger investments with longer lives. Thus, we conclude that both the project finance structure and the participation of development banks help manage political risk. Panel B of Tables 3 and 4 report the discrete changes in the predicted probabilities of the financing choices. In addition, Fig. 1 illustrates the predicted probability of each financing choice at different levels of political risk. The likelihood of using a project finance loan increases linearly as the political risk increases. In contrast, as political risk increases, the likelihood that a development bank will participate rises at an increasing rate.19 Thus, project finance seems to be 19 Beyond the maximum political risk of 0.75 for Columbia in 1999 or 2000, the increase in the probability of development banks participating in the financing loan (i.e., Panel B of Fig. 1) is based on out-of-sample projections and, therefore, is less certain (i.e., the confidence interval widens for higher levels of political risk).

304

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Fig. 1. The choice of project finance and development bank participation at different levels of political risk. This figure shows the predicted probability of project finance (a) and development bank participation (b) at different levels of political risk based on the logit models estimated in regression (3) of Tables 3 and 4, respectively. The x-axis shows the level of political risk with higher numbers indicating higher risk. The y-axis shows the predicted probability of the financing choice being used.

equally effective at all levels of political risk, whereas the participation of development banks is especially valuable at high levels of political risk.20 4.3. The combined choice of project finance and the participation of development banks The financing structure decision regarding project finance and development banks is jointly taken. As a result, we observe four combinations: project finance with development banks in the syndicate, project finance without development banks, full-recourse loans with development banks and full-recourse loans without development banks. Some of the effects of political risk observed in the loan-type regressions above might be at least partly driven by the development banks’ decision to participate in these loans, or vice versa. To disentangle these effects, we consider the four financing choices at the same time in a multinomial logit framework. Table 5 reports the results. We take full-recourse loans without development banks in the syndicate as the baseline choice. Our specification is based on our preferred model, which is given in regression (3) in Tables 3 and 4. Panel A of Table 5 indicates that a higher level of political risk is associated with a higher likelihood of all three financing choices relative to full-recourse loans without development banks. With regard to the other explanatory variables, the results are generally robust. 4.4. The within-country variation of political risk Our cross-country analyses indicate that the use of project finance and the participation of development banks are related to the level of political risk in the borrower’s country and other observed characteristics of the countries, industries, and investments. To derive a causal interpretation and to address the concerns related to omitted variables and sample selection, we now consider a specification that focuses on the within-country variation in political risk. We identify countries with a substantial change in political risk based on the degree to which the military is involved in a country’s government. In one of its political risk sub-categories, the ICRG measures the involvement of the military on a scale from 0 to 12, where higher values indicate a lower degree of involvement of the military. Among the 64 countries in our sample, Israel and Kenya experienced drastic increases in political risk (i.e., a greater degree of involvement from the military in the government’s affairs). Hong Kong, Panama, Peru, and Turkey experienced drastic decreases in political risk (i.e., a lesser extent of involvement from the military in the government’s affairs). We select these 20 This non-linear relationship between political risk and development bank participation also explains why the predicted probability of development bank participation for the average loan in Panel B of Table 4 appears small relative to the sample frequency.

305

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314 Table 5 The joint choice of loan and bank type. Dependent variable

Project finance loan without development bank participation

Full-recourse loan with development bank participation

0.36*** (3.37) 0.13 (1.38) 0.06*** (4.06) 0.02 (0.21) 0.06* (1.64) 0.28*** (2.94) 0.01 (0.18) 0.06 (1.05)

0.08*** (3.43) 0.05* (1.67) 0.01* (1.92) 0.06*** (2.47) 0.06*** (4.66) 0.01 (0.48) 0.04*** (2.48) 0.12*** (3.58)

0.11** (2.29)

0.58*** (3.01)

0.11** (2.07)

0.00 (0.86) 0.00 (0.77) 0.01 (1.04) 0.02*** (3.52)

0.19*** (3.38) 0.10*** (2.62) 0.48*** (10.12) 0.04 (0.99)

0.02*** (2.61) 0.01 (0.71) 0.01 (0.99) 0.02** (2.04)

0.00*** (2.53) 0.00*** (4.18) 0.02* (1.63)

0.01 (0.93) 0.04*** (10.53) 0.14 (1.52)

0.01*** (4.10) 0.00*** (3.37) 0.01 (0.28)

Project finance loan with development bank participation

Panel A: Multinomial logit regression results Country characteristics Political risk 0.05*** (2.75) Corporate governance 0.02 (1.34) Creditor rights 0.00 (1.09) Economic performance 0.03* (1.71) Banking market development 0.02*** (2.79) Annual loan growth 0.00 (0.35) French legal origin 0.01 (1.15) German legal origin 0.04** (2.07) Debt-overhang problem in industry Leverage ratio Industry dummies Mining Transportation and utilities Construction Other Investment characteristics ln(investment size) Investment life Investment risk Observations Log likelihood Pseudo R2

4978 3563.1 0.254 (continued on next page)

countries because the changes in their political risk levels are substantial in scale and shift from a low to a high level of military involvement, or vice versa. These six countries fulfill the following criteria: They experience changes in political risk levels of at least 2.2 points on a scale from 0 to 12. Additionally, their political risk levels switch between the ranges of [0.0, 6.0] and [6.1, 12.0].21 To measure the change in political risk levels, we construct an index ‘‘Dpolitical risk1’’ that we set to 1 for the year in which the country moves to a lower level of political risk and all of the following years. We set the index to +1 for the year in which the country moves to a higher level of political risk and all of the following years. Otherwise, we set the index to zero. For the remaining 58 countries in our sample the index is set to zero for all years.

21

See Table A.3 in the appendix for details.

306

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Table 5 (continued) Predicted probability (%)

Panel B: Discrete changes in the financing One-standard deviation change in Political risk Corporate governance Creditor rights Predicted probability for average loan observation Sample frequency

Project finance loan with development bank participation

Project finance loan without development bank participation

Full-recourse loan with development bank participation

From

From

To

Difference

To

Difference

From

To

Difference

29.2 24.4 29.9

6.2 3.2 7.5

3.3 3.4 3.4 3.9

4.6 4.5 4.4

1.4 1.1 1.0

choice’s predicted probability 1.4 2.0 1.6 1.8

2.2 1.5 1.9

5.0

0.8 0.5 0.4

23.0 27.6 22.4 26.0 26.1

5.9

This table shows multinomial logit regression results. The dependent variable reflects four types of financing structures: project finance loans with development banks in the syndicate, project finance loans without development banks in the syndicate, fullrecourse loans with development banks in the syndicate, and full-recourse loans without development banks in the syndicate (unreported baseline). Panel A displays marginal effects calculated at the sample means relative to the baseline financing choice and, below, the z-statistics. Standard errors are heteroskedasticity robust and clustered at the country-year level. The regressions contain intercepts. Panel B reports predicted probabilities of the respective financing structure that are calculated when the independent variable is valued at a level of half a standard deviation below the mean (reported in column ‘‘from’’) compared to half a standard deviation above its mean (reported in column ‘‘to’’). The difference expresses the change in probability for this one standard deviation change. All other independent variables are held at their mean values. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level.

In Table 6, we include this proxy of change in political risk. We estimate a logit model in accordance with our preferred specification from regression (3) of Tables 3 and 4. We add country dummies and country trends to our specification to capture any time-invariant or time-variant unobserved factors at

Table 6 A within-country specification for changes in political risk. Dependent variable

Project finance dummy

Development bank dummy

Regression

(1)

(2)

(3)

(4)

Dpolitical risk1

0.39*** (5.06) Yes Yes Yes Yes Yes No 4957 2111.1 0.314

0.26** (2.22) Yes Yes Yes Yes No Yes 4957 2056.9 0.332

0.08*** (2.89) Yes Yes Yes Yes Yes No 4846 1215.8 0.287

0.09*** (2.85) Yes Yes Yes Yes No Yes 4846 1212.5 0.289

Country characteristics Debt-overhang problem in industry Industry dummies Investment characteristics Country dummies Country trends Observations Log likelihood Pseudo R2

This table shows logit regression results. In regressions (1) and (2) the dependent variable takes the value of 1 for a project finance loan and 0 for a full-recourse loan. In regressions (3) and (4) the dependent variable takes the value of 1 when a development bank participates in the loan syndicate and 0 otherwise. In addition to country dummies and country trends, the regressions contain the same set of independent variables as regression (3) of Tables 3 and 4. Due to collinearity, the corporate governance proxy drops out of regressions (1) and (3). The table focusses on the within-country change in political risk. For this independent variable, the table displays marginal effects calculated at the sample means and, below, the z-statistics. Standard errors are heteroskedasticity robust and clustered at the country-year level. The regressions contain intercepts. There are some countries without any project finance or development bank participation. All loans to borrowers in these countries are excluded from the samples because these country dummies perfectly predict the dependent variable to be zero. *** Significant at the 1% level. ** Significant at the 5% level. ⁄ Significant at the 10% level.

Table 7 Robustness checks regarding the determinants of the loan contract and development bank participation. (1) Baseline regression

Dependent variable

Project finance dummy

Panel A: The determinants of the loan contract Country characteristics Political risk 0.34*** (2.99) Corporate governance 0.21** (2.18) Creditor rights 0.06*** (3.89)

(2) Regression: exclude investment characteristics

(3) Regression: alternative industry dummies

(4) Regression: combination of (2) and (3)

(5) Sample: exclude real estate loans

(6) Sample: include loans to US borrowers

0.18* (1.82) 0.17* (1.85) 0.08*** (4.89)

0.31*** (2.88) 0.23** (2.36) 0.05*** (3.62)

0.16* (1.66) 0.18** (1.98) 0.07*** (4.68)

0.55*** (5.27) 0.09 (0.83) 0.07*** (4.57)

0.11*** (3.89) 0.05 (1.47) 0.02*** (6.01)

Other country characteristics Debt-overhang problem in industry Industry dummies Alternative industry dummies Investment characteristics

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes No

Yes No

No Yes

No Yes

Yes No

Yes No

Yes

No

Yes

No

Yes

Yes

Observations Log likelihood Pseudo R2

4978 2295.4 0.256

4978 2741.5 0.112

4978 2248.2 0.272

4978 2717.4 0.120

4651 2032.2 0.313

17,789 4431.6 0.309

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Regression Robustness check

(continued on next page)

307

308

Table 7 (continued) Regression Robustness check

(1) Baseline regression

Dependent variable

Development bank dummy

(3) Regression: alternative industry dummies

(4) Regression: combination of (2) and (3)

(5) Sample: exclude real estate loans

(6) Sample: include loans to US borrowers

0.08*** 0.02 (0.62) 0.02** (2.28)

0.09*** 0.02 (0.75) 0.01 (1.50)

0.07** 0.02 (0.65) 0.01⁄⁄ (2.24)

0.09*** 0.00 (0.03) 0.01 (1.24)

0.01*** 0.00 (0.77) 0.00 (1.28)

Other country characteristics Debt-overhang problem in industry Industry dummies Alternative industry dummies Investment characteristics

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes No

Yes No

No Yes

No Yes

Yes No

Yes No

Yes

No

Yes

No

Yes

Yes

Observations Log likelihood Pseudo R2

4978 1338.7 0.222

4978 1481.4 0.138

4978 1324.9 0.230

4978 1447.9 0.158

4651 1188.2 0.205

17,789 1669.1 0.367

This table shows logit regression results. In Panel A, regression (3) of Table 3 is our baseline specification for which we provide a set of robustness checks with respect to the regression specification or the sample. Correspondingly, in Panel B the baseline specification is regression (3) of Table 4. In Panel A, the dependent variable takes the value of 1 for a project finance loan and 0 for a full-recourse loan. In Panel B, the dependent variable takes the value of 1 when a development bank participates in the loan syndicate and 0 otherwise. The table displays marginal effects calculated at the sample means and, below, the z-statistics. Standard errors are heteroskedasticity robust and clustered at the country-year level. The regressions contain intercepts. The other country characteristics include our proxies for economic performance, banking market development, annual loan growth, French and German legal origin. The industry dummies differentiate between mining, transportation & utilities, construction, and other with manufacturing as the excluded benchmark industry. The alternative industry dummies are based on 2-digit SIC codes and differentiate between agriculture & forestry & fishing, mining, construction, wholesale & retail trade, finance & insurance & real estate, services, public administration, transportation & public utitilities, and manufacturing as the excluded benchmark industry. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level.

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Panel B: The determinants of development bank participation Country characteristics Political risk 0.11*** Corporate governance 0.02 (0.69) Creditor rights 0.01 (1.46)

(2) Regression: exclude investment characteristics

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

309

the country level. The results are robust and show that an increase in political risk indeed leads to more project finance loans and increases the participation of development banks. In Table A.4 of the appendix, we apply an even stricter definition of a change in political risk. For this alternative index of change in political risk we only consider Israel, Peru, and Turkey and again find that our results regarding political risk are robust. 4.5. Robustness checks We perform additional analyses to test whether our main results in Tables 3 and 4 are robust. Panels A and B of Tables 7 depict these robustness checks for our analysis of the project finance loans and development banks, respectively. For the sake of comparison, we replicate our preferred specifications from Tables 3 and 4 as regression (1). First, we investigate whether our results are robust to the choice of independent variables. In regression (2), we exclude the investment characteristics because they might not be fully exogenous. In regression (3), we use more detailed industry dummies because the asset type is likely to be a major driver of the decision to choose a project finance loan. In regression (4), we combine both specifications. Our results are robust (i.e., the marginal effect of political risk is significant and positive in all of the specifications).22 Next, we investigate the robustness of our results to two variations in our sample. In regression (5), we consider a reduced sample that does not include loans marked for the purpose of ‘real estate’. Thus, this sample excludes 327 loans. Although project finance has been used to finance real estate projects, full-recourse loans are preferable because physical real estate assets provide excellent collateral in the event of a default (Beidleman et al., 1990). Therefore, loans marked for this purpose might bias our original sample against the use of project finance. The results obtained from this reduced sample are consistent with the results presented in this paper. The marginal effect of the proxy for political risk is even slightly higher in the regression on project finance. In regression (6), we use a sample that includes US loans. Thereby, our sample size increases by 12811 cases, of which 529 are project finance loans and 59 contain development banks in their syndicates. With 12240 cases, 95.5% of the US loans are full-recourse loans without development banks, whereas only 63% of our original sample consists of similar loans. Therefore, we must conclude that the US market is substantially different from other markets with respect to the use of project finance and the participation of development banks. In the extended sample, US loans account for 72% of the observations. Thus, our results might be strongly driven by these additional loans. Although the results are generally robust, the marginal effects of political risk are smaller. 5. Conclusions In this paper, we aim to determine how a firm can finance a project in a politically risky country. We investigate the use of a project finance structure versus a full recourse loan and the participation of a development bank. These two features may be employed to manage political risk and thereby facilitate the financing of projects in high-risk environments. Our results clearly show that in politically risky countries, project finance loans are more likely to be used, and development banks are more likely to participate in loan syndicates. Based on the results of our study, we derive the following suggestions for the parties involved in financing an investment. If the political risk level (e.g., risks related to expropriation or profit repatriation) is high, then the parties can utilize a project finance structure or invite development banks to participate in the loan syndicate to compensate for the high risk level. At high levels of political risk, the participation of development banks is especially effective at reducing risk. Therefore, our results support the notion that development banks stretch their political umbrellas over projects. 22 In Panel A of Table 7, we find that the marginal effect of political risk is only significant at the 10% level in regressions (2) and (4). Therefore, we replicate these regressions for the sample excluding real estate loans and the sample including loans to US borrowers. In both cases, we find positive marginal effects that are significant at the 5% or 1% level. This finding strengthens our confidence in the robustness of our results. The results are available upon request.

310

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Our paper also contributes to the literature by showing that not only law and institutions but also political risk influence the design of loan contracts. Prior studies have focused on the impact of law and institutions on the loan contract. By using political risk as well as law and institutions as our explanatory variables, we find that the influence of political risk is actually more important than that of law and institutions. With respect to law and institutions, we show that the use of project finance Table A.1 Variable definitions and sources. Variable Dependent variables Project finance (PF) Development bank (DB) participation Loan type

Independent variables Country characteristics Political risk

Dpolitical risk1

Dpolitical risk2 Corporate governance

Creditor rights Time to resolve insolvency Economic performance

Banking market development

Description

Source/own calculations based on

Dummy = 1 if loan is structured as project finance loan, 0 if loan is otherwise structured as full-recourse loan Dummy = 1 if development bank participates in at least one of the tranches belonging to the same deal Indicator variable coded as (1) for project finance loans with development bank participation, (2) for project finance loans without development bank participation, (3) for full-recourse loans with development participation, and (4) for full-recourse loans without development participation

Dealscan

Investment-related political risk based on the ICRG’s political risk proxy inv_prof, scaled from 0 to 1 with larger values indicating higher political risk. Note that inv_prof is originally scaled from 0 to 12 with higher values indicating lower risk. For ease of interpretation we invert and rescale this proxy by calculating (12-inv_prof)/12 Indicator variable coded as 1 for year in which a country moves to a lower political risk level and all following years, +1 for all years in which a country moves to a higher political risk level and all following years, 0 otherwise. Political risk is measured by the ICRG political risk category for ‘military in government’ (milit) which ranges from 0 to 12 with higher values indicating less military in government, e.g. lower political risk. A move to a lower political risk level is considered to exist when a country’s milit proxy increases by at least 2.2 points and shifts from the range of [0.0, 6.0] to [6.1, 12.0] from one year to the next. A move to a higher political risk level is considered to exist when a country’s milit proxy decreases by at least 2.2 points and shifts from the range of [6.1,12.0] to [0.0,6.0] from one year to the next As Dpolitical risk1 but a minimum change of 2.0 and ranges from [0.0,5.8] and [6.3,12.0] are considered Index of ex-post control over self dealing transactions ranging from 0 to 1 with higher values indicating more control over self-dealing transaction, e.g. better corporate governance An index aggregating creditor right ranging from 0 to 4 with higher values indicating stronger creditor rights Time to resolve insolvency (years)

ICRG

An index measuring a country’s economic performance ranging from 0 to 1 with higher values indicating better economic performance. It is based on a country’s per capita GNI and of a Euromoney poll of economic projections Domestic credit provided by banking sector (% of GDP, 1.0 indicates 100%)

Euromoney

Dealscan Dealscan

ICRG

ICRG Djankov et al. (2008)

Djankov et al. (2007) WDI–World Bank (2010)

WDI–World Bank (2010) (continued on next page)

311

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314 Table A.1 (continued) Variable Private credit bureau coverage Annual loan growth British legal origin

French legal origin German legal origin Scandinavian legal origin Investment characteristics Ln(investment size)

Investment life Investment risk

Description

Source/own calculations based on

Private credit bureau coverage (% of adults, 1.0 indicates 100%). Time-invariant measure, average over available years is used Annual % change in net domestic credit (measured in local currency, 1.0 indicates 100%) Time-invariant dummy = 1 for countries of British legal origin, 0 otherwise. Baseline dummy excluded from regressions Time-invariant dummy = 1 for countries of French legal origin, 0 otherwise Time-invariant dummy = 1 for countries of German legal origin, 0 otherwise Time-invariant dummy = 1 for countries of Scandinavian legal origin, 0 otherwise

WDI–World Bank (2010)

Natural logarism of the size of the deal to which the tranche belongs. The size is measured in millions of real US dollars in 2005 values using the US GDP deflator series USY99BIRH Maximum maturity among all tranches in the deal measured in years Residual from loan pricing regression, divided by 1000.

Dealscan, IFS

Debt-overhang problem in industry Leverage ratio Median leverage ratio (long-term debt/total assets) for 4-digit SIC industry, median across global firms 1996–2005 Industry dummies Mining Construction Manufacturing

Transportation and utilities Other

Dummy = 1 if borrower belongs to an industry with 10 6 SIC 6 14, 0 otherwise Dummy = 1 if borrower belongs to an industry with 15 6 SIC 6 19, 0 otherwise Dummy = 1 if borrower belongs to an industry with 20 6 SIC 6 39, 0 otherwise. Baseline dummy excluded from regressions Dummy = 1 if borrower belongs to an industry with 40 6 SIC 6 49, 0 otherwise Dummy = 1 if borrower does not belong to any of the industries defined above

Alternative industry dummies Agriculture, forestry, Dummy = 1 if borrower belongs to an industry with fishing 01 6 SIC 6 09, 0 otherwise Mining Dummy = 1 if borrower belongs to an industry with 10 6 SIC 6 14, 0 otherwise Construction Dummy = 1 if borrower belongs to an industry with 15 6 SIC 6 17, 0 otherwise Manufacturing Dummy = 1 if borrower belongs to an industry with 20 6 SIC 6 39, 0 otherwise. Baseline dummy excluded from regressions Transportation and Dummy = 1 if borrower belongs to an industry with utilities 40 6 SIC 6 49, 0 otherwise Wholesale and retail Dummy = 1 if borrower belongs to an industry with trade 50 6 SIC 6 59, 0 otherwise Finance, insurance, real Dummy = 1 if borrower belongs to an industry with estate 60 6 SIC 6 67, 0 otherwise Services Dummy = 1 if borrower belongs to an industry with 70 6 SIC 6 89, 0 otherwise Public administration Dummy = 1 if borrower belongs to an industry with 91 6 SIC 6 97, 0 otherwise

WDI–World Bank (2010) La Porta et al. (1999, 2008)

La Porta et al. (1999, 2008) La Porta et al. (1999, 2008) La Porta et al. (1999, 2008)

Dealscan Dealscan Compustat Global

Dealscan Dealscan Dealscan

Dealscan Dealscan

Dealscan Dealscan Dealscan Dealscan

Dealscan Dealscan Dealscan Dealscan Dealscan

Unless otherwise indicated all country characteristics are time varying, e.g. they are measured on an annual level and are matched to the year of loan signing.

312

Country characteristics

Mean Median Minimum Maximum Standard deviation

Debt overhang problem

Investment characteristics Life Risk (years)

Political risk

Corporate Creditor governance rights

Time to resolve insolvency

Economic Banking performance market development

Private credit bureau coverage

Annual loan growth

Industry’s leverage ratio

Size ($ mio real)

0.19 0.17 0.00 0.75 0.17

0.63 0.68 0.03 1.00 0.24

2.32 1.50 0.40 10.00 2.20

0.62 0.67 0.18 1.00 0.17

0.06 0.00 0.00 0.64 0.14

0.13 0.10 0.16 1.34 0.15

0.13 0.11 0.00 0.43 0.09

407.7 7.5 171.9 5.1 0.6 0.3 19492.9 40.0 915.0 5.4

2.20 2.00 0.00 4.00 1.18

1.10 1.04 0.09 3.13 0.59

This table provides descriptive statistics for the main independent variables. The descriptive statistics are based on our sample of 4978 loans.

0.000 0.019 0.376 1.983 0.121

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

Table A.2 Descriptive statistics.

313

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314 Table A.3 Within-country changes in political risk. Country

Change in the military’s involvement in the government From

Hongkong Israel Kenya Panama Peru Turkey

To

Year

Military in government

Year

Military in government

2003 2001 1997 1996 2000 2003

6.0 7.8 9.0 6.0 4.7 5.8

2004 2002 1998 1997 2001 2004

10.0 5.1 6.0 8.3 8.7 8.0

Change in political risk considered for proxy Change in political risk

Dpolitical risk1

Dpolitical risk2

Decrease Increase Increase Decrease Decrease Decrease

Yes Yes Yes Yes Yes Yes

No Yes No No Yes Yes

This table shows the countries with a substantial change in political risk. Political risk is measured as the involvement of the military in the country’s government. The proxy ranges from 0 to 12 with higher values indicating lower involvement of the military and thus lower political risk.

Table A.4 Robustness check regarding the within-country specification for changes in political risk. Dependent variable

Project finance dummy

Development bank dummy

Regression

(1)

(2)

(3)

(4)

Dpolitical risk2 Country characteristics Debt-overhang problem in industry Industry dummies Investment characteristics Country dummies Country trends

0.37*** (2.65) Yes Yes Yes Yes Yes No

0.43*** (2.89) Yes Yes Yes Yes No Yes

0.08** (2.40) Yes Yes Yes Yes Yes No

0.08** (2.37) Yes Yes Yes Yes No Yes

Observations Log likelihood Pseudo R2

4957 2120.3 0.311

4957 2056.7 0.332

4846 1217.3 0.286

4846 1214.5 0.287

This table shows logit regression results. In regressions (1) and (2) the dependent variable takes the value of 1 for a project finance loan and 0 for a full-recourse loan. In regressions (3) and (4) the dependent variable takes the value of 1 when a development bank participates in the loan syndicate and 0 otherwise. In addition to country dummies and country trends, the regressions contain the same set of independent variables as regression (3) of Tables 3 and 4. Due to collinearity, the corporate governance proxy drops out of regressions (1) and (3). The table focusses on the within-country change in political risk. For this independent variable, the table displays marginal effects calculated at the sample means and, below, the z-statistics. Standard errors are heteroskedasticity robust and clustered at the country-year level. The regressions contain intercepts. There are some countries without any project finance or development bank participation. All loans to borrowers in these countries are excluded from the samples because these country dummies perfectly predict the dependent variable to be zero. *** Significant at the 1% level. ** Significant at the 5% level. ⁄ Significant at the 10% level.

increases with better creditor rights. In a country with weak creditor rights, a project’s payoff may be negligible, which renders a project finance structure unattractive for creditors because they do not have any additional recourse. However, weaker corporate governance provisions increase the probability that a project finance loan will be used. Taken together our findings suggest that the complexity of a project finance loan’s contractual framework can serve as a substitute for both a weak legal environment and a lack of reliable politics. Acknowledgments We warmly thank Philip Strahan (the editor) and two anonymous referees for very constructive comments on a previous version of the paper. We also wish to thank Elena Carletti, Stefano Gatti, Clara Graci-

314

C. Hainz, S. Kleimeier / J. Finan. Intermediation 21 (2012) 287–314

ano, Victoria Ivashina, Luc Laeven, Maria Lehner, Michael Lemmon, William Megginson, Sven Rady, Monika Schnitzer, Drew Winters, Bernard Yeung as well as seminar and conference participants for helpful comments and suggestions. The usual disclaimer applies. Christa Hainz gratefully acknowledges financial support by the German Science Foundation, under SFB-Transregio 15, and LMUMentoring. Appendix A See Tables A.1–A.4. References Bae, K.-H., Goyal, V., 2009. Creditor rights, enforcement and costs of loan finance. J. Finance 64, 823–860. Beck, T., Demirgüç-Kunt, A., Levine, R., 2003. Law and finance: why does legal orgin matter? J. Compar. Econ. 31, 653–675. Beidleman, C.R., Fletcher, D., Veshosky, D., 1990. On allocating risk: the essence of project finance. Sloan Manage. Rev. 31, 47–55. Berger, A.N., Hannan, T.H., 1989. The price–concentration relationship in banking. Rev. Econ. Statist. 72, 291–299. Brealey, R., Cooper, I., Habib, M., 2000. Financing of large engineering projects. In: Miller, R., Lessard, D. (Eds.), Strategic Management of Large Engineering Projects. MIT Press, Cambridge, Massachusetts, pp. 165–180. Buiter, W., Fries, S., 2002. What Should the Multilateral Development Bank Do? Working Paper No. 74. European Bank for Reconstruction and Development, London. Buljevich, E.C., Park, Y.S., 1999. Project Financing and the International Financial Markets. Kluwer Academic Publishers, Norwell, MA. Cadwalader, 2004. Clients & Friends Memo: Moving Towards Hybrid Project Financing. Cadwalader, Wickersham & Taft LLP, New York. Corielli, F., Gatti, S., Steffanoni, A., 2010. Risk shifting through nonfinancial contracts: effects on loan spreads and capital structure of project finance deals. J. Money, Credit, Banking 42, 1295–1320. Djankov, S., McLiesh, C., Shleifer, A., 2007. Private credit in 129 countries. J. Finan. Econ. 84, 299–329. Djankov, S., LaPorta, R., Lopez-de-Silanes, F., Shleifer, A., 2008. The law and economics of self-dealing. J. Finan. Econ. 88, 430– 465. Doidge, C., Karolyi, G.A., Stulz, R.M., 2007. Why do countries matter so much for corporate governance? J. Finan. Econ. 86, 1–39. Esty, B.C., 2003. Economic Motivations for Using Project Finance. Harvard Business School, Mimeo. Esty, B.C., 2004. Why study large projects? An introduction to research on project finance. Europ. Finan. Manage. 10, 213–224. Esty, B.C., Megginson, W.L., 2003. Creditor rights, enforcement, and debt ownership structure: evidence from the global syndicated loan market. J. Finan. Quant. Anal. 38, 37–59. Esty, B.C., Sesia, A., 2009. An overview of project finance and infrastructure finance – 2009 update. Harvard Business School. Kleimeier, S., Megginson, W.L., 2000. Are project finance loans different from other syndicated credits? J. Appl. Corp. Finan. 13, 75–87. La Porta, R., Shleifer, A., Lopez-de-Silanes, F., Vishny, R., 1998. Law and finance. J. Polit. Economy 106, 1113–1155. La Porta, R., Shleifer, A., Lopez-de-Silanes, F., Vishny, R., 1999. The quality of government. J. Law, Econ., Organ. 15, 222–279. La Porta, R., Shleifer, A., Lopez-de-Silanes, F., Vishny, R., 2008. The economics consequences of legal origins. J. Econ. Lit. 46, 285– 332. Laux, C., 2001. Project-specific external financing and headquarters monitoring incentives. J. Law, Econ., Organ. 17, 397–412. Lazarus, S., 2001. IFC and its Role in Globalization: Highlights from IFC’s Participants Meeting Washington, DC, June 6–7, 2001. International Finance Corporation, Washington, DC. Myers, S., 1977. The determinants of corporate borrowing. J. Finan. Econ. 5, 147–175. Neuberger, J.A., Zimmermann, G.C., 1990. Bank pricing of retail deposit account and ‘‘The California Rate Mystery’’. Fed. Reserve Bank San Francisco Rev. (2), 3–16. Nevitt, P.K., Fabozzi, F., 1995. Project Financing. Euromoney Publications, Nestor House, UK. Qi, Y., Roth, L., Wald, J.K., 2010. Political rights and the cost of debt. J. Finan. Econ. 95, 202–226. Qian, J., Strahan, P.E., 2007. How laws & institutions shape financial contracts: the case of bank loans. J. Finance 62, 2803–2834. Smith, W., 1997. Covering political and regulatory risks: issues and options for private infrastructure arrangements. In: Irwin, T. (Ed.), Dealing with Pubic Risk in Private Infrastructure. International Bank for Reconstruction and Development, Washington, DC, pp. 45–88. Subramanian, K.V., Tung, R., 2009. Law and project finance. Unpublished manuscript. Boston University School of Law. World Bank, 2010. World Development Indicators. World Bank, Washington, DC.