Project Finance as a Risk- Management Tool in International Syndicated Lending

Transcription

1 Discussion Paer No. 183 Project Finance as a Risk Management Tool in International Syndicated ending Christa ainz* Stefanie Kleimeier** December 2006 *Christa ainz, Deartment of Economics, University of Munich, Akademiestr. 1/III, Munich. [email protected] **Stefanie Kleimeier, imburg Institute of Financial Economics, FdEWB, Maastricht University, P.O. Box 616, 6200 MD Maastricht, Netherlands. [email protected] Financial suort from the Deutsche Forschungsgemeinschaft through SFB/TR 15 is gratefully acknowledged. Sonderforschungsbereich/Transregio 15 Universität Mannheim Freie Universität Berlin umboldtuniversität zu Berlin udwigmaximiliansuniversität München Rheinische FriedrichWilhelmsUniversität Bonn Zentrum für Euroäische Wirtschaftsforschung Mannheim Seaker: Prof. Konrad Stahl, Ph.D. Deartment of Economics University of Mannheim D68131 Mannheim, Phone: +49(0621) Fax: +49(0621)

2 Project Finance as a RiskManagement Tool in International Syndicated ending Christa ainz # University of Munich Stefanie Kleimeier * Maastricht University Abstract We develo a double moral hazard model that redicts that the use of roject finance increases with both the olitical risk of the country in which the roject is located and the influence of the lender over this olitical risk exosure. In contrast, the use of roject finance should decrease as the economic health and cororate governance rovisions of the borrower s home country imrove. When we test these redictions with a global samle of syndicated loans to borrowers in 139 countries, we find overall suort for our model and rovide evidence that multilateral develoment banks act as olitical umbrellas. JEClassification: Keywords: D82, F34, G21, G32 Project finance, syndicated loans, olitical risk, double moral hazard. # Christa ainz, Deartment of Economics, University of Munich, Akademiestr. 1/III, Munich, Germany, hone: , [email protected] * Stefanie Kleimeier, imburg Institute of Financial Economics, FdEWB, Maastricht University, P.O. Box 616, 6200 MD Maastricht, The Netherlands, hone: , [email protected] The authors would like to thank Franz Benstetter, Elena Carletti, Stefano Gatti, Clara Graciano, Carola Grün, Victoria Ivashina, Isabelle Kronawitter, uc aeven, William Megginson, Sven Rady, Monika Schnitzer, Koen Schoors, and articiants of the Annual Meeting of the German Economic Association Magdeburg, the 2nd Worksho of Alied Infrastructure Research at the Technical University of Berlin, the Research Seminar of the Institute of Caital Market Research and Finance at the University of Munich, the XII International Tor Vergata Conference on Banking and Finance in Rome, the First Conference of the Financial Intermediation Research Society in Cari, the EFA Meeting in Maastricht, the IFE seminar at the University of Maastricht and the Economics Worksho at the University of Tübingen for helful comments and suggestions. The usual disclaimer alies. Christa ainz gratefully acknowledges financial suort from the German Science Foundation, under SFBTransregio 15 and from FOROST.

3 1. Introduction When looking at the global distribution of syndicated loans, we observe that firms in countries with high olitical risk receive relatively few loans, and that many of these loans are structured as limitedrecourse roject finance. The low number of loans can be understood in the context of the law and finance nexus, which states that there is a ositive correlation between creditor rights and financial intermediation (Djankov, Mciesh, and Shleifer, 2007). The rominence of roject finance can be exlained by the use of rivate contracting that mitigates the risks arising from oorly functioning institutions in the host country (Qian and Strahan, 2005). Although these concets may hel to exlain the use of roject finance in general, several questions remain unanswered, such as when is it otimal to grant a roject finance loan? ow does the degree of recourse influence the incentives of lenders and borrowers to manage risks? When multilateral develoment banks are art of the syndicate, what role do they lay in mitigating risk, articularly olitical risk? In this study, we follow Esty (2004) and define "roject finance" as the creation of a legally indeendent roject comany that is financed with equity from one or more sonsoring firms, and which has non or limited recourse debt for the urose of investing in a caital asset. We investigate the relationshi between syndicated lending and the management of olitical and cororate risks. To answer the questions we ose above, we develo a double moral hazard model and test its redictions by using a global samle of syndicated and roject finance loans made to borrowers in 139 countries between 1991 and Our theoretical model analyzes the terms of a rivate loan contract in a setu in which the borrower must exert effort in order to manage the firm, and the lender is able to mitigate olitical risk. Generally, we show that there is a tradeoff between the incentives for the borrower and the lender when they determine the degree of recourse. We find that borrowers refer roject 1

4 finance loans if the cororate governance system in a country is weak, economic health is oor, and olitical risk and bank influence are high. In the emirical art of our study, we also test which tye of lender or syndicate is actually able to mitigate olitical risk in ractice. Thus, our study rovides evidence for the way olitical risk can be mitigated through a loan contract, i.e., through the inclusion of multilateral or national develoment banks in the syndicate. The gas field roject that the South African etrochemical grou Sasol is currently develoing in Mozambique illustrates how contracts can be designed to mitigate olitical risk. To finance the deal, Sasol oted for a unique hybrid roject finance structure. Initially, lenders have full recourse to Sasol, which assumes almost all roject related risks. The sole but imortant excetion is the roject s olitical risk. ere, the loan contract secifies that if welldefined olitical risk events occur, the lenders have the right to switch from the fullrecourse structure to a rojectfinancing structure. The lenders, in articular the develoment banks that fund more than half of the loan, requested that this hybrid structure be created. Cadwalader (2004), the roject s legal consultant, interrets this structure as a commitment device of the develoment banks to actively mitigate the olitical risk: Sasol would like to maximize the influence that the olitical risk roviders [ ] bring to the deal their ability to exert olitical ressure on, in this case, Mozambican government to revent or cure a olitical risk event. imiting the ability of the lenders, and effectively of the subrogated insurers, to ull the roverbial rug from underneath the roject (by limiting the lenders acceleration and enforcement rights) is arguably the most effective way of insuring that such institutions actively seek to remedy or mitigate any olitical risk event [ ]. We rovide a systematic analysis of the relationshi between olitical risk, its mitigation by lenders, and the tye of loan, e.g., the degree of recourse under the loan contract. Thus, our study relates to the emirical and theoretical literature on syndicated loans in general and on roject 2

5 finance loans in articular, and also to aers on the relationshi between law and finance. Since we study the incentive effects of different loan tyes for borrowers and lenders, our aer is also related to the theoretical literature on bank moral hazard and double moral hazard. Most emirical studies on roject finance are limited to loanricing studies or syndicatestructure analysis. 1 Desite the fact that the focus of these studies is quite different from our own, these studies rovide clear evidence for the relevance of olitical risk. All loanricing studies agree that olitical risk, either directly or indirectly via a olitical risk guarantee, is reflected in the sread of the loan. Kleimeier and Megginson s (2000) study also indicates that the higher the olitical risk of the host country the more likely is roject finance. Esty and Megginson (2003) investigate the syndicate structure of roject finance loans in the context of legal quality, which, though not identical to olitical risk, is clearly related to it. These authors find that for borrowers in countries with weak creditor rights and oor legal enforcement, syndicates must be articularly large and diffuse in order to deter strategic default. In contrast, in countries with strong and enforceable legal rights, syndicates are structured to ensure monitoring and lowcost recontracting. Thus, the syndicate structure is a direct resonse to risk. Qian and Strahan (2005) resent similar results. These authors argue that the design of rivate loan contracts is determined by the legal and institutional characteristics of a country. Among the few theoretical aers on roject finance, none addresses the issue of olitical risk in the context of borrower and lender incentives. Shah and Thakor (1987) and John and John (1991) model roject finance based on the tax advantage of debt, and Chemmanur and John (1996) focus on rivate benefits of control. abib and Johnson s (1999) model is driven by the redeloyability of assets. aux (2001) argues that roject finance can be used as a commitment devise by the headquarters to monitor the quality of the roject and to influence the roject 3

6 manager s incentives. Our study is thus the first that models roject finance as an instrument to otimally align both borrower and lender incentives. Moral hazard aers study different tyes of bank moral hazard deending on the tasks assigned to the bank. Rajan and Winton (1995) investigate the bank's incentive to gather unverifiable information on the future rosects of the firm after credit is granted. Manove, Padilla, and Pagano (2001) study the bank s incentive to exert costly effort for screening rojects ex ante. More realistically, the bank s as well as the firm s efforts are unobservable, so there is a double moral hazard roblem. If the double moral hazard roblem exists because the bank has to monitor the effort of the firm, then it is otimal to finance firms by a mix of bank credit and external caital (Besanko and Kanatas (1993)). Double moral hazard also arises in venture caital arrangements in which the entrereneur and the venture caitalist must exert effort (Schmidt, (2003)). Neither ure equity finance nor ure debt finance solves both roblems simultaneously. Instead, a convertible security with an aroriately set rice gives both arties an incentive to exert first best effort. The emirical studies on roject finance highlight the rominent role of olitical risk. Our aer systematically analyzes how olitical risk is mitigated through rivate contracting that is designed such as to solve the double moral hazard roblem. Our study rovides emirical suort for the notation that multilateral develoment banks act as olitical umbrellas. Our study roceeds as follows. Section 2 resents our model with the double moral hazard roblem. We analyze the incentives that result from different degrees of recourse for both lenders and borrowers. In section 3, we derive emirically testable hyotheses. We interret the effort of the lender as an effort made to influence olitical risk, and the effort of the borrower as an effort made to increase the rofitability of the roject. We resent the data sources in section 4. Section 1 oan ricing studies include Sorge and Gadanecz (2004), Nguyen and Ross (2002), Kleimeier and Megginson 4

7 5 rovides a descrition of the global use of roject finance and discusses our emirical results. Section 6 concludes. 2. A Double Moral azard Model of Project Finance We discriminate different modes of bank finance, i.e., fullrecourse syndicated loan versus rojectfinance loans. A rojectfinance loan can either be a limitedrecourse loan that allows lenders some recourse to the sonsors, in the form of, for examle, additional equity or the reduction of dividend or royalty ayments; or a nonrecourse loan that gives lenders no recourse to the sonsors The Basic Model A firm wants to finance a roject, which yields a ayoff of X >0 in the case of success and zero in the case of failure. We assume that the roject s assets have a liquidation value of zero. The roject costs I. We assume that the firm finances the investment roject through credit. The robability of the roject s success is determined by the effort of the firm, denoted by e, and by the effort of the bank, denoted by b. We assume that the imact on the robability of success is indeendent of the action of the resective other agent, so we do not model the bank in its traditional role as a monitor. If the firm's manager exerts effort e, the robability of success increases from to. Similarly, if the bank decides to exert effort b, the robability of success increases from to. Accordingly, the robabilities of success can be,, or. The sonsoring firm that decides on the realization and financing of a new roject has wealth of W. This wealth includes (1998, 2000, 2002), Ivashina (2005). Esty and Megginson (2003) and Sufi (2006) analyze the syndicate structure. 5

8 all assets of the firm and the cash flows generated by all other rojects of the firm. We assume that assets are not firm secific. The time line is as follows: The firm decides whether to incororate the roject within the sonsoring firm or searately. At time 0, the bank offers a selection of credit contracts. All contracts secify the reayment R in the case of success and V in the case of failure, where V is determined by the bank s degree of recourse. The firm chooses the contract, which is then signed by the bank and the firm. Next, both the firm and the bank can exert effort. At time 1, the ayoff of investment, i.e., X or zero, is realized and the bank receives reayment R in the case of success or V in the case of failure. The decision on how to incororate the roject determines the debtor s liability and the bank s degree of recourse, and therefore the bank s ayoff if the roject fails. If the roject is incororated searately, unless the sonsor grants limited recourse it receives a nonrecourse credit. If the roject is incororated within the sonsoring firm, the bank grants a traditional fullrecourse credit. Table 1 shows that in the case of full recourse, the bank receives R even if the roject fails and generates a ayoff of zero. owever, in the case of limited or nonrecourse credit, the bank only receives V if the roject fails with X > V > 0 in the limitedrecourse case, and V=0 in the nonrecourse case. [Insert Table 1 about here] We restrict the analysis to welfareincreasing rojects characterized by the assumtion (e, b) X I e b 0. (1) In a firstbest world with symmetric information, the effort levels of both the firm and the bank can be observed and verified by the court, and therefore stiulated in a contract. In ractice, these 6

9 effort levels are not contractible. Thus, the credit contract must be designed so that both arties have an incentive to exert effort. owever, in a world with asymmetric information, the incentives of both bank and borrower deend on the terms of the credit contract. With a nonrecourse loan, the firm has an incentive roblem, because the exected net return for the firm s manager is lower than its effort costs: Ê I ˆ ( ) ÁX < e. Á Ë (2) Nevertheless, it is socially efficient to exert effort, because we assume that the increase in the exected returns exceeds the costs of effort, i.e. ( ) X > e. Furthermore, we assume that it is ossible to solve the bank s moral hazard roblem if it grants a nonrecourse loan, since the highest exected return from exerting effort covers the bank s effort costs: I. (3) ( ) b > We solve the game by backward induction. We assume that the banking sector is erfectly cometitive; thus, banks make zero exected rofit. Firms chose the contract that maximizes their ayoff from the selection of contracts. First, we must study which tye of contract solves the bank s moral hazard roblem, then we must analyze which loan contract solves the firm s moral hazard roblem. Only then can we focus on the double moral hazard case to show which tye of contract is otimal The Moral azard Problem of the Bank The bank can increase the robability of success by exerting effort b. It decides to do so if or R + ( 1 ) V b R + ( 1 ) ( )( R V) b. V (4) 7

10 The incentive comatibility constraint of the bank (ICB) in (4) is more easily fulfilled if the difference between the bank's ayoffs in the case of either the success or failure of the roject is high. Consequently, increasing R and simultaneously decreasing V imroves the bank's incentive to exert costly effort. We restrict our analysis to arameters that fulfill the bank s articiation constraint (PCB), given by: R + ( 1 ) V I b 0 (5) Proosition 1: The moral hazard roblem of the bank can always be solved by granting a nonrecourse credit, i.e. V=0. Proof: See Aendix A. By incororating the roject searately, the firm is not liable at all if the roject fails. Therefore, the difference between the bank s ayoffs for either success or failure is high, and the bank has an incentive to exert effort. If it is not ossible to induce the bank to exert effort by searately incororating the roject, then the moral hazard roblem of the bank cannot be solved. We exclude this ossibility by assumtion (3). Of course, if the moral hazard roblem of the bank is less severe, i.e., ( )( R V) b the incentive roblem., then both nonrecourse and limitedrecourse credits solve 2.3. The Moral azard Problem of the Firm The firm's manager must also decide whether to exert effort e. Therefore, the management considers the exected ayoff that is influenced by the robability of success, the reayment R in the case of success, and the reayment V in the case of failure. Effort is exerted if 8

11 ( W + X R) + ( 1 )( W V) e ( W + X R) + ( 1 )( W V) or ( )( X R + V) e. (6) By insecting the incentive constraint of the firm (ICF) of Eq. (6) in more detail, we obtain the following imlication for its liability. Proosition 2: The moral hazard roblem of the firm can always be solved by making the borrower fully liable, i.e., R=V. Proof: See Aendix A. By increasing the debtor's liability, his ayoff in the case of failure is reduced, and the difference in statecontingent ayoffs of the firm increases. Thus, the credit contract rovides an incentive to the firm to exert effort. If the moral hazard roblem of the firm is less severe, i.e.,( )( X R + V) e, then the firstbest solution can be reached with limited or fullrecourse credit Double Moral azard The emirical facts suggest that both banks and firms have to contribute to the success of an investment roject. Proositions 1 and 2 show that the debtor's liability influences both the bank's and the firm's incentive to exert costly effort. On the one hand, limited recourse increases the bank's incentive. On the other hand, limited recourse has a negative imact on the firm's incentive. 9

12 1 1 Proosition 3: If both incentive roblems are less severe, i.e., X e + b, ( )( R V) b and ( )( X R + V) e moral hazard roblem., limited recourse, i.e., R>V>0, solves the double Proof: See Aendix A. If the incentive roblems are not severe, the bank can design a credit contract that solves both incentive roblems. The exact terms of the credit contract deend on how slack the incentive comatibility constraints are. Since we assume that banking is erfectly cometitive, the contract always generates an exected rofit of zero to the bank, i.e., ( 1 ) V I b 0 R + =. (7) Therefore, the relationshi between R and V is deduced as ( 1 ) I + b V R =. (8) If the bank s incentive comatibility constraint but not the firm s incentive comatibility constraint is binding, the credit contract stiulates Ï 1 ÌR = I + b ;V = I b. (9) Ó Both the bank and the firm have an incentive to exert effort only if the condition stated in roosition 3 holds. For all other arameter constellations it is imossible to design a contract that solves both moral hazard roblems. In such cases, the contract should grant an incentive to the arty for whom the moral hazard roblem is solved most efficiently. In the following roosition we state which of the moral hazard roblems should be solved. 10

13 Proosition 4: If it is not ossible to solve both incentive roblems, i.e., X < 1 1 e + b (10) and the moral hazard roblem of the firm is severe, i.e., ( ) X b > ( ) X e, (11) then it is otimal to solve the incentive roblem of the bank through roject finance by searately incororating the new roject and via granting a nonrecourse credit, i.e., V=0. Proof: See Aendix A. For most arameter constellations, it is not ossible to design a contract that gives both arties aroriate incentives. Therefore, the otimal contract solves the incentive roblem of the arty whose effort has a relatively higher imact on the robability of success. Therefore, it is otimal to solve the bank s incentive roblem if ( ) X b > ( ) X e. For examle, if e and b have the same size and b increases the robability of success more than e, then it is otimal to give the bank the incentive to exert effort. To induce the bank to exert effort, the difference between the ayoffs for success and failure must be high. The credit contract can stiulate a sufficiently high difference between the statecontingent ayoffs when the new investment roject is searately incororated. Then, the bank receives no return in the case of failure, as the roject s ayoff is zero and there is no recourse on the assets of the firm that sonsors the roject. 11

14 3. Testable yothesis Our model defines the effort of both the bank and the firm only generally, and thus leaves room for different tyes of effort. Since the focus of our study is on risk management, we aly the roositions that arise from our theoretical model to the firm s and bank s effort in managing risks. As noted earlier, our only assumtion on the efforts of bank and firm is that the influence of one on the robability of success of the roject is indeendent of the actions of the other. In our model we assume that the bank exerts effort to mitigate olitical risk, and that the manager makes an effort to imrove the firm s oerational erformance. For examle, the manager determines the technical realization of the roject. The bank, which has many ways in which it can influence the robability of the roject's success, might, for examle, try to influence government decisions, or it might assist the firm in obtaining access to markets or exerts like an auditor. We base our emirical analysis on Proosition 4, because we erceive that in reality the incentive roblems are severe. We use the notation of our model to hrase the incentive roblems we intend to study. The bank s moral hazard roblem arises because a bank must exert effort, which causes costs of b in order for the bank to reduce olitical risk, which is catured by ( ). The firm s moral hazard roblem arises because a firm must exert effort, which causes costs of e, in order for the firm to increase the robability of success from to. For a more detailed derivation, we note that Proosition 4 redicts that roject finance should be the referred financing choice when ( ( )X b )X b > ( )X e 1. (12) ( )X e > An economic interretation of this roject finance reference ratio (PFP ratio) relates the use of roject finance to four factors. These factors cature the firm s and the bank incentive roblems in the following way: 12

15 The firm s moral hazard roblem is reflected by the denominator of the PFP ratio, and here by the influence of managerial effort on the robability of success, given by ( ) relation to the manager s effort costs e. We believe that in countries with a healthy economy, a given level of effort changes the success robability more than would the same effort in a country with a weak economy. In other words, the higher is ( ), in, the better the economic health and erformance of a country. As the PFP ratio decreases with an increase in ( ), we should observe a negative relation between economic health of the borrower s country and the use of roject finance. Furthermore, the effort a firm s manager must exert to manage a cororation causes costs of e. In an international comarison, we would exect the costs of effort to be higher in countries with less effective cororate governance systems. (We note that the absolute size of effort costs can be interreted as rivate benefits. In countries with oor cororate governance, effort costs are high and so are rivate benefits.) In these economies, there are far fewer restrictions and unishments if a manager deviates from the best cororate strategy. Ceteris aribus, the firm s moral hazard roblem is greater in countries with a oor cororate governance system. Given that a better cororate governance system imlies lower effort costs e, and given that the PFP ratio decreases with a decrease in e, we exect a negative relation between the quality of the cororate governance system and the use of roject finance. The numerator of the PFP ratio reflects the bank s moral hazard roblem. ere, ( ) reflects the change in the robability of success when the bank exerts effort, and b reflects the cost associated with this effort. We interret ( ) in the context of olitical risk. The more the government s actions can influence the robability of the success of a roject, the higher the difference between and will be. On the one hand, the robability of success of the roject 13

16 without any effort by the bank,, will be low in countries with high government involvement and high olitical risk. On the other hand, once the bank exerts effort, the success robability of the roject will be significantly increased. Thus, high olitical risk increases the numerator, and we exect a ositive relation between olitical risk and the use of roject finance. The bank s influence on the host government is reflected in b. The higher the bank s influence on the host government, the lower is the cost of b at which a given increase in the roject s success robability will be achieved. In other words, the lower the b, the cheaer it is for a bank to constrain olitically adverse moves. Therefore, we exect a ositive relation between bank influence (measured by a smaller b) and the use of roject finance. To illustrate the role of banks of our model, consider the following situation: If the government erceives that aart from the investment roject under consideration, a bank is imortant to the country, it might be more reluctant to engage in actions against the roject comared to a situation in which the government does not fear farranging consequences of its actions. The government could, for examle, be concerned that negative actions on one roject could sill over onto other rojects financed by the same bank. From the bank s oint of view, any effort will be less costly if the bank is relatively imortant to the government and country. Among all lenders, multilateral develoment banks such as the International Financial Cororation (IFC), a member of the World Bank Grou, or the Euroean Bank for Reconstruction and Develoment (EBRD) have high bargaining ower because they finance many rojects and also rovide financial aid. Thus, there is reeated interaction between the develoment bank and the host government. In contrast to commercial banks that might also be frequent lenders, the develoment banks (DBs) have a secial status. Buiter and Fries (2002) argue that multilateral develoment banks differ from other lenders because their ( ) suort for rivate sector rojects can be 14

17 instrumental in mitigating risks associated with government olities and ractices. Therefore, multilateral DBs are also known as olitical umbrellas (see Buljevich and Park, 1999). For these reasons, we exect that the bank s ability to exert effort and consequently its effort cost b will deend on its status as a DB as well as on its market share in the syndicated loan market. Considering all these factors, we roose the following testable hyothesis: Testable yothesis: Among all syndicated loans made to borrowers in a country, the fraction of roject finance loans is larger, the weaker the cororate governance system, the weaker the economic health, the higher the olitical risk of the borrower s country and the higher the influence of the lending bank over the host government. 4. Data Our main data source is the Dealscan database. To obtain a roxy for our deendent variable, we extract from Dealscan all syndicated loans (Ss) signed between January 1, 1991 and December 31, By controlling for the borrower s nationality and the tye of syndicated loan, we can define the roject finance share as PF PF_vol i = $ S $ i i *100 (13) where PF $ i is the total volume of roject finance loans (PFs) in $ million of all borrowers of country i and S $ i is the total volume of S in $ million of all borrowers of country i. First, we note that this measure of our deendent variable is consistent with our assumtion that the investment roject is credit financed, and with our differentiation between different tyes of bank finance. S $ i reflects all investments that managers finance with bank credit. These include fullrecourse syndicated loans as well as limited or nonrecourse PFs. 15

18 Using these factors makes it ossible for our emirical analysis to exlain how large the share of PFs is. Second, we aggregate all loans to borrowers of the same country over time. Because not all countries access the syndicated loan market every year, the number of anelobservations ranges from one to 15 er country. Furthermore, for many of the anel observations, the value of the deendent variable is zero. This is often the case for countries with a low number of Ss er year. We ot for the ooled aroach, because we have a sufficiently large number of countries in our samle. owever, we do conduct robustness checks based on a anel data set for those countries that enter the syndicated loan market every year. We obtain the roxies for our four exlanatory factors from Dealscan, Euromoney, and the World Bank. Since the firm s moral hazard roblem deends on how much an increase in the manager s effort will raise the success robability of a roject, given the manager s effort cost, we need two countrylevel variables to measure these two comonents of firm moral hazard. Earlier, we noted that a manager's influence on the robability of success deends on the economic health of a country, meaning that a given effort has a larger effect in a country with better economic ersectives. We use Euromoney s economic erformance index as our roxy for the economic health of the borrower s country. This annual index is based on the current GDP er caita figures and on a oll of economic rojections. Thus, it contains not only current but also forwardlooking information, which is esecially useful to us when we consider the medium to longterm nature of syndicated loans. To aggregate this variable over time, we use an equally weighted average over all years. Furthermore, we convert the original scale of this roxy from zero to 25 to zero to 100. A higher value indicates better economic erformance and health. We also noted earlier that the firm s effort costs deend on the country s cororate governance system, which means that the weaker the cororate governance system, the higher is 16

19 the manager s oortunity cost (lost rivate benefits or erquisites) of ursuing the best cororate strategy. As a roxy for the strength of the cororate governance system, we use a measure of financial develoment that combines the develoment of the stock market and the banking system. We define cororate governance as the equally weighted average of stock market caitalization and domestic credit to the rivate sector, both as a ercentage of GDP. We obtain both values from the World Bank s World Develoment Indicators (WDI). A higher value indicates better cororate governance, because managers in countries with a more develoed financial sector are more controlled by active stockholders and bank lenders than are managers in countries with less develoed financial sectors. Thus, this measure catures the strength of the cororate governance system in terms of market forces. Similar to our economic health roxy, we average the annual values. Because we measure both comonents of the cororate governance roxy as ercentages, we do not rescale this variable. The bank s moral hazard roblem also consists of two comonents, olitical risk and bank influence, for which we need countrylevel roxies. We note that olitical risk can be divided into three broad categories: traditional olitical risk, regulatory risk, and quasicommercial risk (Smith 1997). The traditional olitical risk category addresses risks relating to exroriation, currency convertibility, and transferability, and to olitical violence. The regulatory risk category covers risks arising from unanticiated regulatory changes. These risks include taxation or foreign investment laws alicable to the whole economy, but can also be industry secific. The quasicommercial risk category reflects those risks that arise when the roject contends with stateowned suliers or customers whose ability or willingness to fulfill their contractual obligations towards the roject is questionable. We consider traditional olitical risk and regulatory risk when we interret bank moral hazard. The World Bank s Worldwide Governance Research Indicators Data Set secifies six 17

20 measures of olitical risk that fit our ercetion of these risk categories. These measures are the voice and accountability of the government, olitical stability, government effectiveness, regulatory quality, rule of law, and control of corrution. This data set covers all countries in our samle, but its time coverage is limited to 1996, 1998, 2000, 2002, and As our main roxy, we use a combined average of all six measures. We aly simle averages and rescale the resulting roxy from zero to 100. A higher value indicates more olitical risk. For a more detailed analysis, we also look at each of these six measures searately. As the second comonent of the bank s moral hazard roblem we measure the lender s influence over the host government, which, as noted earlier, deends on both the lender s status as a DB and its market share in the S market. First, we obtain national league tables for Ss signed between 1991 and 2005 for each country i from Dealscan. These tables rank all lenders based on the amount of funds they rovide during our samle eriod, which allows us to identify exactly which lender rovides how many funds to each country. Second, we classify lenders as DBs based on the World Bank s definition of multilateral develoment banks and multilateral financial institutions. We define our measure of bank influence as the aggregate market share of these DBs in the S market and as such combine lender status and market share in our roxy. We measure the market share of DBs as their total syndicated loan volume to borrowers of country i relative to the total syndicated loan volume of all lenders to borrowers of country i. A higher value indicates more bank influence. For a detailed analysis, we also measure a lender s individual market share and extend our analysis to rominent national DBs, such as exortimort banks. We include these latter lenders if they have a substantial share in the S market and might therefore have substantial influence over the host government. Based on Dealscan s league table for global Ss signed between 1991 and 2005, we select those national DBs that fund at least 100 Ss which are worth $1,000 bn. 18

21 Table 2 lists the individual DBs and indicates to which category they belong. 2 Among them, the Euroean Bank for Reconstruction and Develoment (EBRD) and Germany s Kreditanstalt für Wiederaufbau (KfW) are the most rominent lenders, followed by the Jaan Bank for International Cooeration (JBIC). There is no clear link between a DB s total loan volume and the share of PF in its loan ortfolio. For examle, the PFshare is high for the firstranked EBRD but low for the secondranked KfW. Overall, loan volume alone does not imly a reference for roject finance. [Insert Table 2 about here] 5. The Global Market for Syndicated oans, Project Finance, and Political Risk 5.1. The Use of Project Finance Between January 1991 and December 2005, comanies from 139 countries raised funds in the global syndicated loan market, amounting to 119,779 facilities worth $24,778 billion. 3 As Table 3 shows, 5,282 of these facilities are PFs worth $941 billion, which lenders allocated to borrowers from 108 countries. As exected, borrowers in industrialized countries receive most Ss. Western Euroe, North America, and Australia alone account for 87% of the global S volume. ooking at a country s PF borrowing relative to its total S borrowing indicates that PFs are not a very imortant source of funds for cororate borrowers in industrialized countries. For examle, in North America and Western Euroe, 3% or less of the total S volume is in form of PF. 2 The table also identifies which lenders we include in our bank influence roxy. When discussing emirical results related to this bank influence roxy, we will refer to these lenders in general terms as develoment banks (DBs). Only when we conduct the detailed analysis of individual lenders will we secifically refer to the different lender categories. 3 Dealscan actually contains loans to borrowers from 145 countries, but we deleted six countries (Bermuda, British Virgin Islands, Brunei, Cayman Islands, Iraq, and Taiwan) due to missing values for our four main roxies. These countries account for an additional 2,395 S facilities worth $320,566 million. 19

22 Australia is an excetion with 15%, as are individual western Euroean countries such as Greece and Portugal. Borrowers from atin America, Eastern Euroe, the Middle East, Africa, and Asia dislay a reverse attern. For most of these countries, the volume of PFs relative to all national Ss is at least three times, and at most ten times, as high as the global average. In the most extreme cases of Cameroon, Neal, and Uganda, PFs account for 100% of the national S volume, followed by iberia, Tajikistan, and Bosnia erzegovina with more than 80%. Overall, it aears that countries with high shares of PFs reresent cororate borrowers from countries with higher olitical risk, lower economic erformance, and less access to credit or equity finance. This observation gives us our first indication that risk influences the decision of banks and borrowers on PF over onbalancesheet lending. [Insert Table 3 about here] Fig. 1 and Table 4 rovide further suort for the overall relevance of risk for PF. ere we use a very general, comrehensive measure of risk, Euromoney s country risk score, which comrises both economic and olitical risk. Fig. 1 resents our individual country observations and shows a clear attern of more roject finance when there is higher country risk. In Table 4, we slit our total samle into quartiles based on this country risk score and comare the different levels of S and PF usage. These quartiles confirm the results of Fig. 1. As risk increases from low to moderately low, the relative use of PF rises from about 7.1% to 22.5%. For countries with moderately high and high olitical risk, PF becomes even more imortant as a source of financing, amounting to 32.2% and 24.8%, resectively. To illustrate the strength of our results, we note that for two of the four quartiles, the mean values for two consecutive quartiles are 20

23 significantly different based on a onesided ztest. The difference is most ronounced for the two low countryrisk quartiles and could imly a somewhat nonlinear relation between the use of PF and risk. [Insert Fig. 1 and Table 4 about here] 5.2. The Relevance of Bank and Firm Moral azard To test our hyothesis, we first use our four main roxies for bank and firm moral hazard. Regressions (1) to (5) in Panel A of Table 5 resent the results. Based on our hyothesis, we exect to find that PF is used more often when the bank influence is stronger and olitical risk and the economic health and cororate governance system are weaker. In the single regressions (1) to (4), all sloe coefficients are significant and have the exected signs. The coefficients size shows that a 10% increase in bank influence, i.e., market share of DBs, leads to 4.8% more PF, and a 10% increase in olitical risk leads to 6.8% more PF. The coefficients of our firm moralhazard roxies are somewhat smaller. ere, a 10% imrovement in economic health or cororate governance leads to 3.2% and 1.9% less PF, resectively. Combined with the exlanatory ower in terms of adjusted R 2, it aears that bank moral hazard, and articularly olitical risk, is the most imortant determinant of the use of PF. In the multile regression (5), where we use factor analysis to remove the correlation between the indeendent variables, each of our four roxies still has the exected sign. owever, our economic erformance roxy is no longer significant. Based on our observations in Fig. 1 and Table 4, we also consider nonlinear secifications (not reorted) by using the squared values of our indeendent variables. Although the coefficients have the exected sign, the exlanatory 21

24 ower of these nonlinear regression are generally lower. Only for economic erformance does the adjusted R 2 increase to 7.8% in comarison to the 6% of regression (3). owever, in a multile regression, economic erformance remains nonsignificant. Overall, our findings suort both our hyothesis and our theoretical model. We can also conclude that bank moral hazard is a relatively more imortant determinant of PF. [Insert Table 5 about here] We conduct additional analyses to test whether our main results are robust for different definitions of the deendent and indeendent variables. Table 5 also reorts these results. In Panels A to E, we aggregate the indeendent variables olitical risk, cororate governance, and economic erformance in different ways. In addition to the simle average, we either calculate a weighted average with weights based on the annual S volume or we use the first or last available annual value. We also use the initial value of these variables in combination with the absolute change in this variable between the initial and last year. Within each anel, the deendent variable remains defined as roject finance share based on the volume of PFs but we also use an alternative deendent variable based on the number of PFs. ooking at each anel searately, we see that our model is robust to the two different definitions of our deendent variable. The sign, size, and general significance of our four exlanatory factors are roughly the same. owever, we note that our model can exlain the variation in the number of PFs better than it can exlain the variation in terms of loan volume. The adjusted R 2 s of the regressions are generally higher for the roject finance share based on loan numbers. Under the multile regression secification in Panel B, where we use weighted averages as an aggregation mechanism, we can exlain almost 23% of the variation in the use of 22

25 PFs. This better fit for loan numbers might be driven by industry effects. For examle, PFs for construction, oil and gas, or transortation rojects are among the largest, with an average deal size of more than $500 million, comared to loans for utilities, telecommunication, or mining rojects, with an average size of $395, $384, or $216 million, resectively. Since we aggregate our data, we cannot control for these industrydriven demand effects. A deendent variable based on loan numbers does not suffer from this bias and the fit of our model is better. The results for the different aggregation mechanism aear to be robust. We find little difference in terms of sloe coefficient and exlanatory ower between simle averages, initial values, or last values. owever, the aggregation of weighted averages leads to higher exlanatory ower, articularly in the multile regressions. A weighted average is better able to cature the resonse of PF to the time variations in olitical risk and cororate governance. In contrast, these time variations cannot be catured with our secifications in Panel E. ere, the initial level of each determinant matters, but the change over the years does not. We also investigate whether the results are robust over time. Two imortant events took lace during our samle eriod, each of which affected global financial markets. These events were the Asian crisis of 1997 and the Russian crisis of Therefore, we slit our samle into recrises, crises, and ostcrises eriods and aggregate the deendent and indeendent variables searately for each of the three eriods. Our results are reorted in Table 6 and indicate that the crises affect only the relation between firm moral hazard and PF. The role of DBs aears to have become more imortant during the crises, but the differences are not statistically significant. owever, the sensitivity of PF to olitical risk increases significantly. Before the crises, we find that a 10% increase in olitical risk leads to a 3.9% higher volume of PF relative to S. In contrast, during and after the crises, the same 10% increase in olitical risk leads to a 4.1% and 5.5% higher volume of PF relative to S, resectively. This increased sensitivity might indicate 23

26 that due to their exeriences during the crises, lenders erceived level of olitical risk has increased. Perhas the crises changed their assessment of the benefits of PF. For examle, before the crises, lenders exected to be able to manage roblems arising from a (severe) deterioration in the olitical risk environment without resorting to PF. During the crises, they realized that this was not the case. As a consequence, they now ot for PFs in situations in which they used to choose Ss. [Insert Table 6 about here] We also conduct an analysis on a country and yearsecific basis. Doing so allows us to directly control for any crisisdriven time variations on an annual level. Furthermore, we can recover any variation over the years that might be lost when we aggregate our deendent and indeendent variables. We construct a anel that contains all 65 countries that access the syndicated loan market during each year between 1996 and The time restriction is necessary because our cororate governance roxy has missing values for the other years. The reduction from 139 to 65 countries allows us to create a balanced anel in which each crosssectional country observation has the same number of timeseries observations. Table 7 resents the results, which confirm our revious findings that bank influence, olitical risk, and cororate governance are the driving forces behind the use of PF. As the fixed timeeffects show, PF was indeed used more during the crisis years. owever, overall, the additional information included in the anel does not imrove the exlanatory ower of our model. Thus, we are confident that the ooled data set alied for our main analyses is adequate. 24

27 [Insert Table 7 about here] 5.3. InDeth Analysis of Bank Moral azard We wish to examine in more detail the relation between PF and bank influence, and PF and olitical risk. First, for bank influence we rovide a roxy by measuring the aggregate market shares of all DBs including national ones. Comaring the results of regression (1) in Panel A of Table 8 with our baseline result in regression (1) of Panel A of Table 5 shows that national DBs do indeed have influence. Adding their market shares to our determinant increases the exlanatory ower of our model from 8.2% to 15.4%. In regression (2) of Table 8 we look at the market share of each DB category searately. The significant coefficients for multilateral DBs and national DBs indicate that both grous can influence olitical risk via PF, but because of their higher coefficients, national DBs are better able to do so than are multilateral DBs. Multilateral financial institutions have a nonsignificant coefficient, which indicates that even as a grou, multilateral financial institutions can only influence olitical risk when they join forces with other DBs. Regression (3) investigates the role of the five leading DBs in terms of individual lending share. ere, we find that individually, only the World Bank can significantly imrove the influence of the syndicate. A syndicate can influence olitical risk in such a way that the use of PF increases by 3.2% for every 10% increase in lending share, but the imact of a syndicate that includes the World Bank is twice as high as the coefficient of 6.8% (=3.2%+3.6%) shows. [Insert Table 8 about here] 25

28 Panel B of Table 8 reorts the results of our indeth analysis of olitical risk by looking at the individual comonents of our olitical risk index. The ositive, significant coefficients for each of the six comonents are consistent with our earlier findings: under high olitical risk, the borrower refers PF. The fact that the regression results are similar imlies that all forms of olitical risk matter. So far, our analysis indicates that olitical risk can be managed with PF. Other studies show that rivate contracts hel to mitigate the deficiencies of the legal system (Esty and Megginson, 2003; Qian and Strahan, 2005). Therefore, we wish to disentangle the effects that a oor legal system has on the use of PF, as measured by the rule of law and regulatory quality indicators on the one hand and olitical risk on the other. To do so, we conduct a twoste rocedure. In the first ste, we run the regression (olitical risk = a + b rule of law). In the second ste, we use the resulting error term as the residual olitical risk. We roceed similarly for regulatory quality. Our results, which we reort in regressions (10) and (11), indicate that neither rule of law nor regulatory quality can exlain the whole variation in the use of PF. The influence of the residual olitical risk is significant as well, if only marginally so in case of rule of law. Together with the finding that our deendent variable varies systematically with bank influence, we interret this result as evidence that the borrower uses PF to give lenders an incentive to mitigate olitical risk. Thus, the choice of a articular contract is motivated by both the aim to cure institutional deficiencies and by the intention to mitigate olitical risk. 6. Conclusion We oen our aer with the surrising finding that banks grant relatively more PFs to borrowers in riskier countries. This observation indicates that limited recourse is an efficient choice that is made by the arties of the credit contract, the bank, and the firm. We exlain that limited 26

29 recourse rovides a bank with the incentive to exert effort in order to reduce olitical risk. Thus, our model shows that the terms of a loan contract deend on the risk characteristics of the roject to be financed. Thus, the terms of the credit contract are used to manage risks. From the emirical art of our study we conclude that roject finance rovides very flexible structures that can be adated to the needs of an individual roject, articularly to the economic and olitical environment in which the roject oerates. Moreover, we find evidence for the role of multilateral and national DBs as olitical umbrellas. For commercial lenders, but also for sonsors, this result imlies that they can use the articiation of DBs strategically to reduce olitical risks. For lenders in general, olitical risk matters in terms of legalsystem quality as well as other olitical risks, and PF is an effective tool to deal with high risk environments. This insight has imortant imlications for the riskiness of PFs and the treatment of PFs by regulators. Unless the contracting arties make other rovisions, then having no recourse can increase the loss given default. But the degree of recourse also has imortant incentive effects that influence the robability of default. Studying the role of collateral demonstrates that for borrower behavior, the incentive effect dominates (iberti and Mian, 2005). A similar argument might aly to the imact of recourse on lender behavior. So far, there is very little information about the robability of default of PFs comared to other Ss. The emirical test of this question is left for future research. 27

30 Aendix A: Proofs of Proositions 1 4 A.1. Proof of Proosition 1 Deending on the arameter constellation, we can distinguish two cases: Case A ( )( R V) b for V 0 Case B ( ) R b for V 0 = In Case A, nonrecourse as well as limitedrecourse credits solve the incentive roblem. In contrast to Case B, nonrecourse is not a rerequisite for solving the bank s moralhazard roblem. In Case B, the bank needs higherowered incentives, since the benefits of effort on the robability of success are not as easily reaed, either because effort b is more exensive or not as owerful in terms of increasing the robability of success, i.e., ( ). To induce the bank to exert effort, the roject must be incororated searately, and therefore the firm is not liable at all in the case of failure. If it is not ossible to induce the bank to exert effort by searately incororating the roject, then the moral hazard roblem of the bank cannot be solved. This case is excluded by assumtion (3). Q.E.D. A.2. Proof of Proosition 2 For the bank, the design of the otimal contract deends on the arameter constellation. We must distinguish the following cases: Case 1 ( )( X R + V) e for R V Case 2 ( ) X e for R = V 28

31 In Case 1, the bank exerts the firstbest effort with either limited or fullrecourse credit. Full recourse is not necessary to solve the roblem, because either the difference in statecontingent ayoffs of the roject, X 0, is high, or the influence of e on the robability of success is large, i.e., (), which rovides strong incentives to the firm s manager. In Case 2, full recourse is necessary to induce the firm's manager to exert effort e. By increasing the debtor's liability, the ayoff in the case of failure is reduced, and thus, the difference in statecontingent ayoffs increases. Q.E.D. A.3. Proof of Proosition 3 Parameter constellations as in Case A and Case 1 Case A: ( )( R V) b for V 0 > Case 1: ( )( X R + V) e for R > V (A.1) (A.2) contain the cases in which we can solve both incentive roblems simultaneously. imited liability solves the firm's roblem and the ositive ayoff for the bank in the case of failure does not destroy the bank's incentive. The otimal credit contract secifies R and V according to the solution of the following otimization: max R,V,e,b s.t. ( W + X R) + ( 1 )( W V) ( )( X R + V) e 0 ( IC F) ( )( R V) b 0 ( IC B) R + e ( 1 ) V I b 0 ( PC B) The bank s articiation constraint is strictly binding because there is erfect cometition in the banking sector. Accordingly, the terms of the contract are given by I + b R = R V ( 1 ) I + b V = V or (A.3) 29

32 From the incentive comatibility constraints, the difference between the state contingent ayoffs is determined by: R V R V X b e from (IC B) from (IC F) The condition, which has to hold to solve both roblems simultaneously, is given by 1 1 X e + b. If the bank s incentive comatibility constraint is binding, but the firm s incentive comatibility constraint is not, then the credit contract determines Ï 1 ÌR = I + b ;V = I b. Ó Q.E.D. A.4. Proof of Proosition 4 In the following cases it is always imossible to solve both incentive roblems. Case 1 and Case B: Due to the incentive roblem of the bank it is necessary that V=0. This nonrecourse structure imlies for Case 1 ( )( X R) e 0 where R = I. owever, this case is ruled out by assumtion, since in this case there would be no incentive roblem of the firm. Case 2 and Case A or Case B: Case 2 requires R=V. But this fullrecourse structure cannot solve the bank's incentive roblem, because ( ) 0 b is not fulfilled. The firm maximizes its rofit by solving the moralhazard roblem that has the higher return. (1) Return of managerial effort To solve this roblem, the contract must secify R=V. Thus, the effect of e is: 30

33 31 ( ) ( )( ) [ ] ( ) ( )( ) [ ] ( )( ) ( ) e X e V R X V W 1 R X W e V W 1 R X W = + = (2) Return of bank effort To solve this roblem, the contract must secify V=0 and R, where R is such that 0 b I R =. Thus, the effect of b is ( ) ( ) ( ) b X W 1 I X W W 1 b I X W = Í Í Î È + ˆ Á Á Ë Ê + Í Í Î È + ˆ Á Á Ë Ê + + Deending on which exression is higher, the firm's manager decides which incentive roblem to solve: If ( ) ( ) b X e X >, then the firm's incentive roblem should be solved by a credit contract that secifies R=V. For arameters like those in Case 1, the contract might also secify R>V. If ( ) ( ) b X e X <, then the bank should be granted an incentive to exert effort by a credit contract secifying V=0. In Case A, the bank also gets the firstbest incentive when it has limited recourse, V>0. Q.E.D.

34 References Besanko, D., Kanatas, G., Credit market equilibrium with bank monitoring and moral hazard. Review of Financial Studies 6, Buiter, W., Fries, S., What should the multilateral develoment bank do? EBRD Working Paer No. 74. Euroean Bank for Reconstruction and Develoment, ondon. Buljevich, E. C., Park, Y. S., Project Financing and the International Financial Markets. Kluwer Academic Publishers, Norwell, MA. and Dordrecht. Cadwalader, Clients&Friends Memo: Moving Towards ybrid Project Financing. Cadwalader, Wickersham & Taft P, New York. Chemmanur, T. J., John, K., Otimal incororation, structure of debt contracts, and limitedrecourse roject financing. Journal of Financial Intermediation 5, Djankov, S., Mciesh C., Shleifer, A., Private credit in 129 countries. Journal of Financial Economics, forthcoming. Esty, B. C., Why study large rojects? An introduction to research on roject finance. Euroean Financial Management 10, Esty, B. C., Megginson, W.., Creditor rights, enforcement, and debt ownershi structure: Evidence from the global syndicated loan market. Journal of Quantitative and Financial Analysis 38, abib, M. A., Johnsen, B. D., The financing and redeloyment of secific assets. Journal of Finance 54, olmström, B., Financing of investment in Eastern Euroe: A theoretical ersective. Industrial and Cororate Change 5, Ivashina, V., Effects of syndicate structure on loans sreads. Unublished working aer. Stern Business School, New York. John, T., John K., Otimality of roject financing: Theory and emirical imlications in finance and accounting. Review of Quantitative Finance and Accounting 1,

35 Kleimeier, S., Megginson, W.., A comarison of roject finance in Asia and the West. In: ang,.. P. (Ed.), Project Finance in Asia. Advances in Finance, Investment and Banking. Vol. 6. Elsevier North olland, Amsterdam, Kleimeier, S., Megginson, W.., Are roject finance loans different from other syndicated credits? Journal of Alied Cororate Finance 13, Kleimeier, S., Megginson, W.., An emirical analysis of limited recourse roject finance. METEOR Research Memorandum RM/02/060. Maastricht University, The Netherlands. aux, C., Projectsecific external financing and headquarters monitoring incentives. Journal of aw, Economics, and Organization 17, iberti, J., Mian, A., Uncovering collateral. Unublished working aer. University of Chicago. Manove, M., Padilla, A. J., Pagano, M., Collateral vs. roject screening: A model of lazy banks. RAND Journal of Economics 32, Nguyen,.., Ross, D. G., Project finance risk ricing decisions: Australian evidence. Unublished working aer. University of Western Sydney, Australia. Qian, J., Strahan, P. E., ow laws & institutions shae financial contracts: the case of bank loans. NBER Working Paer 11052, National Bureau of Economic Research, Cambridge, Massachusetts. Rajan, R., Winton, A., Covenants and collateral as incentives to monitor. Journal of Finance 50, Schmidt, K. M., Convertible securities and venture caital finance. Journal of Finance 58, Shah, S., Thakor, A. V., Otimal caital structure and roject financing. Journal of Economic Theory 42,

36 Smith, W., Covering olitical and regulatory risks: Issues and otions for rivate infrastructure arrangements. In: Irwin., T. (Ed.), Dealing with Pubic Risk in Private Infrastructure. The International Bank for Reconstruction and Develoment. 2 nd edition, Washington, Sorge, M., Gadanecz, B., The term structure of credit sreads in roject finance. BIS Working Paer No. 159, Bank for International Settlements, Basel, Switzerland. Sufi, A., Information Asymmetry and Financing Arrangements: Evidence from Syndicated oans. Journal of Finance, forthcoming. 34

37 100 roject finance share (%) average risk of borrower's country Fig. 1. The relation between the use of roject finance and country risk. This figure lots the roject finance share against the risk of the borrower s home country. Our samle contains 139 countrysecific observations and comrises all loans signed between January 1991 and December We measure the roject finance share as the total volume of roject finance loans of all borrowers in a given country and exress it as a ercentage of the total volume of all syndicated loans of all borrowers in this country. We measure country risk as the average across all annual Euromoney s country risk scores from 1991 to We convert Euromoney s original score to a scale of zero to 100, with higher values indicating more country risk. 35

38 Table 1 Payoffs at time 1 This table illustrates the cash flows to the roject, sonsoring firm, and bank under different tyes of recourse in our oneeriod, twostates of the world model. In state 1, the roject is successful and ays a cash flow of X. In state 2, the roject fails and ays zero. The loan contract secifies that the bank receives R in state 1 and V in state 2. V indicates the amount of inside collateral and deends on the recourse structure of the loan. The roject costs, which are fully loanfinanced amount to I. The sonsoring firm has cash flows unrelated to the roject in the amount of W. Inside Cash flows to Recourse structure State collateral Project Sonsor Bank lender Full recourse 1 V=R X W+X R RI 2 V=R 0 WR RI imited recourse 1 0<V<R X W+X R RI 2 0<V<R 0 WV VI Non recourse 1 V=0 X W+X R RI 2 V=0 0 W 0I 36

39 Table 2 Syndicated lending by develoment banks This table lists all financial institutions, which we caterorize according to World Bank's guidelines as multilateral develoment banks (MDBs) or multilateral financial institutions (MFIs). The table also lists all national develoment banks (NDBs) with more than $1,000 billion in funding and more than 100 syndicated loans signed between January 1, 1991 and December 31, For each financial institution we reort the total amount of syndicated loans (Ss) that it funds, the share of roject finance loans among all funded syndicate loans, and whether or not the financial institution is included in our bank influence roxy. Funded amount (in $ bn) Share of PF volume in S ortfolio (in %) Inclusion in the bank influence roxy Financial institution Category Euroean Bank for Reconstruction & Develoment MDB 31, yes Kreditanstalt fur Wiederaufbau NDB 30, no Jaan Bank for International Cooeration NDB 26, no Korea Develoment Bank NDB 19, no World Bank (including IBRD and IFC) MDB 11, yes Exort Develoment Canada NDB 10, no Euroean Investment Bank MFI 7, yes ExortImort Bank of Korea NDB 5, no Internationale Nederlanden Bank NV NDB 3, no Asian Develoment Bank MDB 3, yes ExortImort Bank of the Reublic of China NDB 1, no Nordic Investment Bank MFI yes Cororacion Andina de Fomento MDB (subregional) yes InterAmerican Develoment Bank MDB yes Islamic Develoment Bank MFI yes African Develoment Bank MDB yes Central American Bank for Economic Integration MDB (subregional) yes OPEC Fund MFI yes 37

40 Table 3 Geograhic distribution of syndicated loans The table reorts descritive statistics for all syndicated loans to borrowers in 139 countries signed between January 1, 1991 and December 31, 2005 and for the average regional characteristics during this eriod. Based on the volume of syndicate loans and roject finance loans in $ million ($m), we calculate the roject finance share as the volume of roject finance loans in ercent of syndicated loan volume. We characterize the borrower s regions in four dimensions. We measure bank influence as the volume of syndicated loans funded by multilateral develoment banks and multilateral financial institutions in ercent of the total volume of syndicated loans. We measure economic health with Euromoney s economic erformance index which includes current GDP er caita figures and on a oll of economic rojections. The scale of this roxy ranges from zero to 100. We define cororate governance as the equally weighted average of stock market caitalization and domestic credit to the rivate sector, both as a ercentage of GDP. We use the average of six measures of olitical risk voice and accountability of the government, olitical stability, government effectiveness, regulatory quality, rule of law, and control of corrution from The World Bank s Worldwide Governance Research Indicators Data Set to measure olitical risk. The scale of this roxy ranges from zero to 100. A higher value indicates better economic health, better cororate governance, or more olitical risk. Syndicate loan market characteristics Syndicated loan volume (in $m) Project finance loan volume (in $m) Project finance share Average characteristic of the borrower's region Bank Economic Cororate Political Borrower region influence health governance risk Middle East & Turkey 312, , % Eastern Euroe & CIS 312,241 66, % Asia & Pacific 2,658, , % atin America & Caribbean 410,439 50, % Africa 92,042 7, % Western Euroe 6,024, , % North America 14,968, , % Global 24,778, , %

41 Table 4 Changes in the use of roject finance at different levels of country risk This table illustrates how the use of roject finance changes for borrowers in countries with different levels of country risk. We sort our 139 countrysecific observations into four quartiles of countries with low, moderately low, moderately high, and high country risk. For each quartile we reort the average country risk and average roject finance volume. We measure the roject finance share as the total volume of roject finance loans of all borrowers in a given country as a ercentage of the total volume of all syndicated loans of all borrowers in this country. We measure country risk as the average across all annual Euromoney s country risk scores from 1991 to We convert Euromoney s original score to a scale of zero to 100, with higher values indicating more country risk. The ztest indicates whether or not the average roject finance share differs between the current and next higher quartile. *, **, and *** are based on onesided robabilities and indicate that the current mean is significantly smaller than the mean of the next higher quartile at the 0.10, 0.05 and 0.01 level, resectively. evel of country risk Average country risk Average roject finance share ztest Number of observations ow % 4.68*** 34 Moderately low % 1.59* 35 Moderately high % igh % 35 39

42 Table 5 Determinants of the use of roject finance This table shows regression results for our main roxies based on a samle of 139 countrysecific observations. We define the deendent variable, roject finance share, in two different ways. First, as the total volume of roject finance loans of all borrowers in a given country and exress it as a ercentage of the total volume of all syndicated loans of all borrowers in this country. Second, as the total number of roject finance loans of all borrowers in a given country and exress it as a ercentage of the total number of all syndicated loans of all borrowers in this country. We measure bank influence as the volume of syndicated loans funded by multilateral develoment banks and multilateral financial institutions in ercent of the total volume of syndicated loans. We measure economic health with Euromoney s economic erformance index which includes current GDP er caita figures and on a oll of economic rojections. The scale of this roxy ranges from zero to 100. We define cororate governance as the equally weighted average of stock market caitalization and domestic credit to the rivate sector, both as a ercentage of GDP. We use the average of six measures of olitical risk voice and accountability of the government, olitical stability, government effectiveness, regulatory quality, rule of law, and control of corrution from The World Bank s Worldwide Governance Research Indicators Data Set to measure olitical risk. The scale of this roxy ranges from zero to 100. A higher value indicates better economic health, better cororate governance, or more olitical risk. In each anel we use a different aggregation method for the indeendent variables olitical risk, cororate governance, and economic health. In addition to the simle average, we aggregate annual values by (A) calculating a simle, equally weighted averages over time, (B) calculating a weighted average with weights based on the relative syndicated loan volume, (C) using the first or (D) last available annual value, (E) by using the initial value in combination with the change over time. We calculate this change as absolute difference between the values in the initial and last year. The calculation of the bank influence roxy does not change and we reort the results for comarison reasons only. We estimate single regressions as OS, but use factor analysis in multile regressions to remove the correlation between the indeendent variables. The table reorts tstatistics within arentheses. Deendent variable: Project finance share based on loan volume Deendent variable: Project finance share based on loan numbers (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Panel A: Simle average Bank influence (3.66) (3.50) (4.25) (4.06) Political risk (3.57) (2.73) (3.74) (2.74) Economic health (3.16) (0.01) (3.47) (0.30) Cororate governance (3.47) (2.45) (3.71) (2.60) Constant (7.86) (1.15) (7.18) (9.59) (10.34) (8.00) (1.25) (7.60) (9.98) (10.76) Adjusted R Panel B: Weighted average Bank influence (3.66) (4.63) (4.25) (5.40) Political risk (3.44) (2.28) (3.60) (2.28) Economic health (3.37) (0.78) (3.68) (1.13) Cororate governance (3.53) (2.52) (3.72) (2.63) Constant (7.86) (1.01) (7.40) (9.64) (10.83) (8.00) (1.10) (7.81) (9.99) (11.42) Adjusted R

43 Table 5 continued Deendent variable: Project finance share based on loan volume Deendent variable: Project finance share based on loan numbers (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Panel C: First value Bank influence (3.66) (3.57) (4.25) (4.11) Political risk (3.48) (2.61) (3.55) (2.46) Economic health (2.68) (0.34) (3.03) (0.68) Cororate governance (3.35) (2.41) (3.57) (2.58) Constant (7.86) (1.10) (7.08) (9.48) (10.33) (8.00) (1.11) (7.53) (9.84) (10.74) Adjusted R Panel D: ast value Bank influence (3.66) (3.34) (4.25) (3.93) Political risk (3.59) (2.81) (3.80) (2.90) Economic health (2.90) (0.42) (3.16) (0.63) Cororate governance (3.44) (2.35) (3.66) (2.46) Constant (7.86) (1.11) (7.68) (9.59) (10.31) (8.00) (1.24) (8.06) (9.95) (10.75) Adjusted R Panel E: Mixed values Initial olitical risk (2.93) (3.57) Change in olitical risk (1.23) (1.29) Initial economic health (3.48) (3.21) Change in economic health (0.89) (1.22) Initial cororate governance (2.50) (2.68) Change in cororate (1.07) (1.11) governance Constant (7.19) (1.14) (9.53) (1.17) (7.63) (9.90) Adjusted R

44 Table 6 Robustness checks regarding the effects of the Asian and Russian crises We aggregate the data into three searate eriods: a recrises eriod from 1991 to 1997, a crises eriod from 1998 to 1999, and a ostcrises eriod from 2000 to For each eriod, we define the deendent variable as the total volume of roject finance loans of all borrowers in a given country and exress it as a ercentage of the total volume of all syndicated loans of all borrowers in this country. We measure bank influence as the volume of syndicated loans funded by multilateral develoment banks and multilateral financial institutions in ercent of the total volume of syndicated loans. We measure economic health with Euromoney s economic erformance index which includes current GDP er caita figures and on a oll of economic rojections. The scale of this roxy ranges from zero to 100. We define cororate governance as the equally weighted average of stock market caitalization and domestic credit to the rivate sector, both as a ercentage of GDP. We use the average of six measures of olitical risk voice and accountability of the government, olitical stability, government effectiveness, regulatory quality, rule of law, and control of corrution from The World Bank s Worldwide Governance Research Indicators Data Set to measure olitical risk. The scale of this roxy ranges from zero to 100. A higher value indicates better economic health, better cororate governance, or more olitical risk. For all indeendent variables, we calculate simle averages across each suberiod. We estimate regressions as OS. The table reorts t statistics within arentheses. The Ftest indicates whether or not the sloe coefficients differ across eriods. *, **, and *** indicate significant differences at the 0.10, 0.05 and 0.01 level, resectively. Deendent variable: Project finance share (1) (2) (3) (4) Bank influence * recrises dummy 1.14 (2.22) Bank influence * crises dummy 1.29 (2.56) Bank influence * ostcrises dummy 0.58 (5.15) Political risk * recrises dummy 0.39 (2.90) Political risk * crises dummy 0.41 (3.11) Political risk * ostcrises dummy 0.55 (4.37) Economic health * recrises dummy 0.24 (3.62) Economic health * crises dummy 0.19 (2.55) Economic health * ostcrises dummy 0.22 (2.50) Cororate governance * recrises dummy 0.16 (2.92) Cororate governance * crises dummy 0.09 (2.04) Cororate governance * ostcrises dummy 0.11 (2.71) Constant (10.55) (0.51) (8.37) (11.60) Ftest Precrises versus crises Precrises versus ostcrises ** Crises versus ostcrises *** Number of observations Adjusted R

45 Table 7 Panel data robustness checks regarding the determinants of the use of roject finance This table shows regression results for our main roxies based on a anel of 585 observations taken for 65 countries and nine years. We define the deendent variable as the total volume of roject finance loans of all borrowers in a given country and year and exress it as a ercentage of the total volume of all syndicated loans of all borrowers in this country and year. Second, as the total number of roject finance loans of all borrowers in a given country and exress it as a ercentage of the total number of all syndicated loans of all borrowers in this country. We measure bank influence as the volume of syndicated loans funded by multilateral develoment banks and multilateral financial institutions in ercent of the total volume of syndicated loans. We measure economic health with Euromoney s economic erformance index which includes current GDP er caita figures and on a oll of economic rojections. The scale of this roxy ranges from zero to 100. We define cororate governance as the equally weighted average of stock market caitalization and domestic credit to the rivate sector, both as a ercentage of GDP. We use the average of six measures of olitical risk voice and accountability of the government, olitical stability, government effectiveness, regulatory quality, rule of law, and control of corrution from The World Bank s Worldwide Governance Research Indicators Data Set to measure olitical risk. The scale of this roxy ranges from zero to 100. A higher value indicates better economic health, better cororate governance, or more olitical risk. In Panel A, we estimate a fixed effects model which contains yearsecific intercets. We estimate regressions as OS. In anel B, we estimate single regressions as OS, but use factor analysis in multile regressions to remove the correlation between the indeendent variables. The table reorts tstatistics within arentheses. Deendent variable: Project finance share (1) (2) (3) (4) (5) Panel A: Fixed effects model Bank influence (2.09) (5.95) Political risk (7.41) (1.64) Economic health (6.67) (0.46) Cororate governance (6.56) (5.07) 1996dummy (4.56) (4.24) (7.52) (6.28) (2.76) 1997dummy (5.79) (3.60) (8.59) (7.70) (4.73) 1998dummy (5.89) (2.41) (8.64) (7.91) (5.40) 1999dummy (5.55) (1.88) (7.83) (7.16) (6.50) 2000dummy (5.27) (3.60) (7.67) (6.99) (5.99) 2001dummy (5.13) (2.30) (7.51) (6.95) (6.20) 2002dummy (4.68) (2.93) (7.68) (6.89) (5.71) 2003dummy (3.82) (3.23) (7.10) (6.31) (4.87) 2004dummy (4.94) (2.30) (7.25) (6.60) (5.56) Adjusted R

46 Table 7 continued Deendent variable: Project finance share (6) (7) (8) (9) (10) Panel B: Random effects model Bank influence (3.06) (5.89) Political risk (7.32) (2.37) Economic health (5.81) (2.95) Cororate governance (5.13) (3.10) Constant (14.27) (3.13) (11.27) (14.45) (16.67) Adjusted R

47 Table 8 An indeth analysis of the role of bank moral hazard This table shows regression results for our samle of 139 countrysecific observations. We define the deendent variable as the total volume of roject finance loans of all borrowers in a given country and exress it as a ercentage of the total volume of all syndicated loans of all borrowers in this country. In Panel A, we measure bank influence as the volume of syndicated loans funded by develoment banks in ercent of the total volume of syndicated loans. We distinguish between three different categories of develoment banks: multilateral develoment banks (MDBs), multilateral financial institutions (MFIs), and national develoment banks (NDBs). In regressions (1) and (3) we aggregate all three categories whereas we investigate them searately in regression (2). In regression (3) we also emloy banksecific dummy variables in combination with our market share roxy. These dummies are coded as one if the resective develoment bank is art of the syndicate and zero otherwise. This allows us to investigate whether the addition of a articular develoment bank increases the influence of the grou as a whole. Only those develoment banks are considered that lend to at least 30 of our 139 countries. These are the Euroean Bank for Reconstruction and Develoment (EBRD), World Bank (WB), Kreditanstalt für Wiederaufbau (KfW), Jaan Bank for International Cooeration (JBIC), and Exort Develoment Canada (ExCan). In Panel B, we measure olitical risk using the six individual measures rovided by The World Bank s Worldwide Governance Research Indicators Data Set. Political risk as used in regressions (10) and (11) is defined as the simle average of these six individual measures. To avoid multicollinearity roblems we do not use this roxy directly but regress it on our roxies for rule of law and regulatory quality, resectively. We use the regression residuals as the residual olitical risk roxy in regressions (10) and (11), resectively. We estimate all regressions as OS. The table reorts tstatistics within arentheses. Deendent variable: Project finance share (1) (2) (3) Panel A: Bank influence Aggregate market share of all MDBs, MFIs, and NDBs 0.56 (5.11) Aggregate market share of all MDBs 0.48 (3.79) Aggregate market share of all MFIs 0.92 (1.03) Aggregate market share of all NDBs 0.78 (3.60) Aggregate market share of all MDBs, MFIs, and NDBs 0.32 (1.35) Aggregate market share of all MDBs, MFIs, and NDBs * EBRD dummy 0.08 (0.35) Aggregate market share of all MDBs, MFIs, and NDBs * WB dummy 0.36 (1.68) Aggregate market share of all MDBs, MFIs, and NDBs * KfW dummy 0.41 (0.85) Aggregate market share of all MDBs, MFIs, and NDBs * JBIC dummy 0.29 (1.25) Aggregate market share of all MDBs, MFIs, and NDBs * ExCan dummy 0.10 (0.27) Constant (6.95) (6.81) (6.10) Adjusted R

48 Table 8 continued Deendent variable: Project finance share (4) (5) (6) (7) (8) (9) (10) (11) Panel B: Political risk Control of corrution 0.66 (4.06) Government effectiveness 0.58 (3.40) Political stability 0.58 (2.96) Voice and accountability 0.69 (3.75) Regulatory quality (2.46) (2.56) Rule of law (3.30) (3.32) Residual olitical risk (1.55) (3.52) Constant (1.22) (0.70) (0.65) (1.30) (0.10) (0.63) (0.64) (0.10) Adjusted R