Abstract
It must have been sometime in January or February of 1999, when Mike Mussa, sitting in his office on the tenth floor of the IMF headquarters building, remarked to us that if IMF loans were always repaid, there was obviously no way that the IMF could be a source of moral hazard.
It must have been sometime in January or February of 1999, when Mike Mussa, sitting in his office on the tenth floor of the IMF headquarters building, remarked to us that if IMF loans were always repaid, there was obviously no way that the IMF could be a source of moral hazard.
We nodded our heads respectfully but did not really believe him. Clearly, it could not be as simple as that. Whether or not they were repaid, IMF loans made crises less unpleasant for countries, so they created an incentive to be less careful about avoiding them in the first place. That was moral hazard, wasn’t it—like failing to install fire detectors in the presence of fire insurance? Mike’s observation, furthermore, seemed to relate only to the relationship between the IMF and crisis countries. But what about international investors whose reckless lending behavior had contributed to the crises? They were let off the hook with the help of IMF crisis lending. There had to be “investor moral hazard,” which Mike’s claim seemed to ignore altogether.
But who was to contradict such an authority in international economics, who also happened to be the author of a pathbreaking early contribution to the theory of moral hazard and adverse selection, not to mention our boss? So, rather than engaging him in an argument there and then, we went down to the second floor cafeteria and thought about it.
By the time we were back in his office, in one of our meetings with Mike and Alexander Swoboda on a paper that we were writing at the time, we realized not only that Mike was right, but that his “theorem” was far more meaningful than we had first thought. In particular, Mike was making a much stronger point than the usual argument that IMF lending provided desirable insurance at a justifiable cost to incentives and thus did not generate “excessive moral hazard.” Larry Summers, for example, had on many occasions compared the IMF to a fire brigade, arguing that although fire brigades were undoubtedly a source of moral hazard, everyone agreed they were a good thing. In contrast, Mike was pointing out that—under some conditions, which he believed were generally met—IMF assistance quite literally created no moral hazard. According to Mike, in the circumstances under which the IMF lends (or should be lending), there is no trade-off between insurance and incentives. To the extent that governments or investors adjust their behavior in the presence of the IMF “safety net”—say by lending more, at lower interest rates, or by monitoring less—this will be efficient and desired under the conditions of Mike’s theorem.
But what are these conditions? In essence, the Mussa theorem holds if IMF lending occurs at no one’s expense. In particular, it must not occur at the expense of IMF creditor countries and their taxpayers. This would be the case if the IMF were to provide transfers, rather than loans; hence Mike’s insistence that IMF loans take place under adequate safeguards, and that the terms of IMF lending reflect the repayment risk faced by the IMF. But IMF lending must also not occur at the expense of the taxpayers in the borrowing countries, which are ultimately the source of IMF repayment. This point had been left implicit by Mike, who had assumed that countries acted like a single agent, or were ruled by benevolent governments. Our contribution to the Mussa theorem was to persuade Mike to make this assumption explicit. In this form, the “theorem”—in words—made it into our joint paper (Mussa and others, 2000). This was an extensive treatise on international financial architecture, which turned out to be so long and mind-numbingly even-handed that Rudi Dornbusch, our discussant at the conference when it was presented, accused it of “reading like a report by the Council on Foreign Relations.”
Thus buried, it is perhaps no wonder that the Mussa theorem did not make much of an initial impact. Mike himself stated his point in two subsequent papers (Mussa, 1999 and 2004), and we for our part tried on at least one occasion (Jeanne and Zettelmeyer, 2001). But as far as we can see, it has still failed to win many converts, and basic misunderstandings on the welfare effects of IMF lending and the conditions under which it creates moral hazard continue to plague the literature. One reason may be that the Mussa theorem has never been stated as such— namely, as a tight, formal link between assumptions and results. The objective of this paper is to fill this gap. With this out of the way, one can have a useful discussion on whether the assumptions underlying the Mussa theorem are likely to be satisfied or not—and if not, how IMF lending policies would need to be structured to make them hold.
In this paper, we develop the theorem in the context of a simple model of IMF lending. We begin by showing that even in a model in which moral hazard is ruled out by construction—in the sense that outcomes are independent of any effort variables—the presence of IMF lending will lead to symptoms that are commonly associated with moral hazard—namely, an increase in capital flows and a decline in risk premiums. We then introduce policy effort and show that under the assumptions of the Mussa theorem, any reduction in effort as a result of IMF lending is efficient and hence does not constitute moral hazard in the usual sense. Finally, we discuss how moral hazard arises when the assumptions of the Mussa theorem are violated.
We also explore the implications of the theorem for the empirical literature on moral hazard and for IMF lending policy. The main contribution here is to point out that virtually all empirical “tests of moral hazard” that have been proposed in the literature miss the point, because they test for symptoms of insurance that arise—assuming that IMF lending is effective—regardless of whether moral hazard is present. Instead, we advocate an approach that focuses on testing the assumptions of the Mussa theorem directly.
I. A Simple Model of the IMF
Our analysis is based on a simple model in which the IMF may lend to countries that are unable to roll over their external debt with private investors. We look at the impact of IMF lending not only ex post (at the time of the crisis) but also ex ante, on the equilibrium of the credit market. The question is whether IMF bailouts distort the behavior of international investors or the domestic authorities in a way that could be characterized as moral hazard.
Assumptions
Consider a large number of identical emerging market (capital scarce) countries. The rest of the world is populated by investors from rich countries. We denote by W the wealth of foreign investors per emerging market country.
There are three periods, t = 0, 1, 2. The representative emerging market country invests k in period 0; this yields an output y = θk in period 2, where the productivity parameter θ is stochastic. The total cost of investing k is k+f(k), where f(.) is increasing and convex.
A signal about the level of future output is observed in period 1. For simplicity, we assume that the signal is perfectly informative, that is, agents observe θ in period 1. The probability distribution function of θ is denoted by g(·).
Emerging market countries have no wealth in period 0, so investment must be financed by borrowing abroad. We assume that k is financed by short-term debt maturing in period 1. The size of the debt repayment coming due in period 1 is denoted by d. This debt is held by a continuum of atomistic foreign investors who have liquidation rights in period 1. A creditor who is not fully repaid in period 1 can liquidate a small fraction of the investment, subject to a sequential service constraint.1
The total liquidation value of the investment is λθk with λ < 1, implying that liquidation is never efficient.
In period 1 the country has no income and must roll over its short-term debt by issuing new debt maturing in period 2. The country can pledge (credibly promise to repay) a fraction α of its output to the new lenders. This fraction is lower than 1 because of limited contract enforcement (as in the sovereign debt literature), or debtor moral hazard (as in Holmstrom and Tirole, 1998).
The utility of any agent (that is, residents of emerging market countries as well as foreign investors) is given by
The residents of an emerging market country have no income in periods 0 and 1, and consume in period 2 only. Their welfare, thus, is equal to their expected final consumption (period 2 output minus debt repayment). International lenders are risk neutral and lend provided that the expected return is nonnegative. There is a zero-return safe technology allowing investors to store their wealth between periods. We make the following assumption, which will be one of the two pillars supporting our main result on IMF-induced moral hazard (the Mussa theorem).
Assumption A1. The emerging market country is run by a benevolent policymaker who chooses k to maximize the welfare of the representative resident.
The equilibrium is characterized by a pair (k*, d*) that maximizes domestic welfare conditional on the participation constraint of foreign lenders. We define ex ante (period 0) global welfare per emerging market country as the sum of the welfare of the representative resident, U0, and of the foreign lenders, V0:
We will be interested in the question of how IMF bailouts affect the welfare of debtor countries and their lenders, taking into account their possible effect in terms of moral hazard. But first, let us look at the equilibrium when there is no IMF—that is, in a situation of complete laissez-faire.
Laissez-Faire
There is a rollover crisis if the debt coming due in period 1, d, is larger than the output that the country can pledge to private lenders, αθk We assume that in this case, a coordination problem between creditors makes them run for the exit, and the investment is completely liquidated. Hence, one of the following two things happens in period 1: either there is no liquidation (if d ≤ αθk), or there is complete liquidation (if d > αθk).
A rollover crisis occurs in period 1 if θ falls below a threshold ¯θ .2 Thus, ¯θ is defined as the minimum level of productivity at which debt d can still be serviced, given a level of investment k,
The zero-profit condition for lenders is (assuming that λ is so small that the payoff that creditors derive from liquidation can be neglected on the right-hand side)
Equation (2) says that the real interest factor, 1 + r ≡ d/k, is inversely proportional to the equilibrium probability of repayment.
Combining equations (1) and (2) to substitute out d and k, one obtains
We assume that this equation has solutions ¯θlf As shown in Figure 1, equation (3) may have several solutions, in which case the economically relevant solution is the lowest one. Importantly, this solution depends on the level of α, that is, the degree to which output can be pledged for repayment. Increasing α shifts the curve upward in Figure 1 and so decreases ¯θlf. The integral on the right-hand side of equation (3) increases, and so the equilibrium interest rate rlf= α¯lf - 1 is decreasing with α. Note that the interest rate r does not depend on k. This property, which comes from the linearity of the production function, simplifies the algebra a great deal.
The welfare-maximizing level of k, denoted klf, is the solution to
implying
In the laissez-faire equilibrium, the debtor country’s welfare is given by
Because in equilibrium investors demand a zero expected return on their wealth, their ex ante welfare is simply
implying the following expression for global welfare (per debtor country):
Global welfare is equal to the net surplus of the investment in the emerging market country plus the wealth of investors. Notice that global welfare is lower than the first-best level, defined as the level that would arise if the country could raise foreign capital through an equity contract that removes the risk of liquidation by promising foreign investors a share of domestic output rather than a fixed repayment.3 In that case, there would be no liquidation, and global welfare would be given by
with kfb given by
Comparing with equation (5) shows that the first-best level of capital is higher than the level under laissez-faire. Capital flows are inefficiently low under laissez-faire because of the liquidation risk, which reduces the expected productivity of the investment.
Effects of IMF Lending
We now introduce the IMF, an institution that collects funds from international investors (or, equivalently, rich countries that tax international investors) in period 0 and may lend them to emerging market countries with rollover crises in period 1. We assume that in period 2 the Fund’s assets are redistributed to its financiers in proportion to their contributions.
We assume that in period 1 the country can pledge a fraction β of its output to the IMF, where β is in general different from α. In the remainder of this paper, we assume that β > α. In other words, in this paper we interpret the critical difference between the IMF and private lenders as being that the maximum repayment that the IMF can extract from a country, as a fraction of output, is higher than for private creditors.4 Countries are willing or able to sacrifice a larger fraction of their output to repay the IMF than they are to repay the private sector. This could be because of IMF conditionality or because the IMF can punish defaulting countries more severely than private creditors can. As a result, the IMF may be able to lend in some rollover crises, without incurring a loss, when private lenders cannot.
We assume that although it is a monopoly, the IMF lends at the actuarially fair interest rate (that is, it does not extract any monopolistic rent from crisis countries, and its expected profit is zero). This implies that investing in the IMF provides the same return (zero) as investing in the debt market, and that the creation of the Fund leaves the welfare of rich-country investors unchanged. Whether, as an empirical matter, this is a realistic assumption will be discussed in the next section.
Assumption A2. The IMF lends at the actuarially fair interest rate.
The IMF lends under the same conditions as private lenders would if they could extract βy from a country, instead of αy. The equilibrium in period 0, hence, is the same as if private creditors had a power of extraction β. (It does not matter to period 0 lenders whether their claims will be rolled over with public or private money, provided that they are repaid.) Then the equilibrium relationships remain (1) and (2), except that α is replaced by β.
Because ¯θ is decreasing with α, β > α implies that the liquidation threshold is lower with the IMF, ¯θimf < ¯θlf. The IMF, therefore, lends in equilibrium to countries with 9 in the range [¯θimf, ¯θlf], that is, to countries that are not solvent enough to roll over their debts with private creditors but are solvent enough to borrow from the Fund.
As a result, in the presence of IMF lending, in period 1 a country is in one of the following three cases:
If θ ≥ ¯θlf, the country can roll over its debt with private lenders, and the IMF does not lend;
If θ ∈ [¯θimf,¯θlf], the country repays its debt with an IMF loan—the IMF bails out private.creditors; or
If θ < ¯θimf, the IMF lets the rollover crisis result in a liquidation of the investment.
It is important to note that under our assumptions, an IMF bailout does not involve a transfer of resources from the rest of the world to the debtor country or the private investors who are bailed out. The IMF lends at an interest rate that reflects the risk of default on its loans. The IMF, thus, does not take more credit risk than a perfect lender of last resort. The IMF is different, however, from a lender of last resort, since it lends to countries that are insolvent (and not only illiquid) on their debt to private creditors. These countries are solvent on their debt to the IMF because of the latter’s superior ability to extract repayment.
Replacing ¯θlf with ¯θimfin equation (5) gives the equilibrium level of capital, kimf.
It is larger than klf but remains smaller than the first-best level kfb.
The presence of the IMF produces the symptoms commonly associated with “investors’ moral hazard.” Specifically, the anticipation of IMF bailouts
Increases capital flows to emerging market countries (kimf is larger than klf); and
Lowers the interest rate at which these countries can borrow abroad (r imf is smaller than rlf).
However, there is no moral hazard stricto sensu in this model. In law and economics, moral hazard refers to the risk that one party to a contract can behave to the detriment of the other party once a contract has been concluded. Moral hazard requires a hidden action. For example, Kreps (1990, Chapter 16) defines the problem of moral hazard as one in which “one party to a transaction may undertake certain actions that (a) affect the other party’s valuation of the transaction but that (b) the second party cannot monitor/enforce perfectly.”
Because there is no hidden action in our model, there cannot be moral hazard stricto sensu. The larger capital flows and lower interest rate are the reflection of optimal insurance, not moral hazard. As noted by Tirole (2002, p. 42), “Even assuming that investors invest in expectation that official or taxpayer resources will be available for bailouts, the notion of over lending moral hazard … is, as it stands, puzzling. Investors are motivated by profit and react to the incentives given to them.…”
This point seems to be misunderstood in many discussions of IMF-induced moral hazard (in particular those that emphasize “investors’ moral hazard”). These discussions tend to confuse the nondistortive impact of insurance on the equilibrium of the credit market with the possibly (but not necessarily) distortive consequences of insurance.
Insurance that does not generate any distortion is optimal. Under the assumptions we have made so far, the presence of the IMF unambiguously increases the ex ante (period 0) welfare of all agents. It leaves the welfare of rich-country investors unchanged, and it strictly increases the welfare of the residents in emerging market countries. More formally, in the presence of the IMF global welfare per country,
is higher than under laissez-faire because
The Mussa Theorem
Having looked at the impact of IMF lending on capital flows and interest rates, let us now focus on moral hazard on the side of the borrower, that is, the incentives of the domestic policymaker to take actions that reduce the probability of a crisis.5 We now assume that the distribution of θ is a function g(e, ·), where e is a policy effort that costs h(e)k to the representative resident (with h(·) increasing and convex). Increasing the effort raises the probability that θ will be above any given threshold, that is,
One question is whether the policy effort is decided at the same time as, or after, foreign investors lend to the country—that is, whether the country can commit to the effort level in the context of its relationship with private lenders. If commitment were not possible, e would be a hidden action in the relationship between the country and private lenders, and the debt contract would be affected by debtor moral hazard. Because we are interested in moral hazard specifically generated by IMF lending, we first look at a situation in which this brand of debtor moral hazard is absent. We assume, therefore, that the country can commit to e at time 0, when it borrows from private investors.6
Under laissez-faire, the threshold in θ triggering a rollover crisis is still given by equation (3),
which, using (6), implies that ¯θlf is a decreasing function of e,
Raising e shifts the curve upward in Figure 1 and so lowers the intersection ¯θlf.
Other things being equal, increasing the effort reduces the default risk premium and, consequently, the interest rate. Higher effort thus reduces the probability of a crisis through two channels: directly, by increasing the probability of high-productivity draw θ, and indirectly, by reducing the interest rate and thus the ex post burden of the debt. Under laissez-faire the country’s ex ante welfare is given by
where we have used the fact that the expected repayment to lenders must be equal to k.7 The country sets the effort level to maximize (7). The optimal effort under laissez-faire is thus the solution to the problem
With the IMF - the country’s welfare and optimal effort would be given by (7) and (8), in which ¯θlf(e) is replaced by ¯θimf(e). Does the presence of the IMF increase or reduce the domestic policy effort? As shown by Corsetti, Guimarães, and Roubini (2003) and Cordella and Levy-Yeyati (2004), the answer to this question is ambiguous in general. The expectation of IMF bailouts could decrease or increase the level of effort.
This question can be studied in a specification of our model. For simplicity, let us assume that θ can take just three values, θ l, θm, and θ h, with
For the sake of argument, we assume that the values of the parameters are such that under laissez-faire, there is a rollover crisis if the state is low or medium, and that the superior enforcement technology of the IMF allows it to rescue a country if the productivity state is medium (but not if it is low). That is,
The probability of a particular state s = l, m, his denoted by p(e, s). First, let us assume that effort shifts probability from the medium to the high state. More specifically, the probability of the low state, pl, does not depend on e; the probability of the medium state is (1 - pt)(1 - e); and the probability of the high state is (1 - pt)e. Then, applying equation (8), the effort levels with and without the IMF are given by
from which it follows that the level of effort is lower with the IMF,
By exercising effort, the country can increase the likelihood of ending up in the high state rather than the medium state. Without the IMF, the only way of avoiding a rollover crisis and inefficient liquidation is to end up in the high state. With the IMF, however, the country is protected from a rollover crisis in the medium state. Hence the incentive to apply effort to avoid the medium state is higher without the IMF.
Note, however, that this result can easily be reversed by changing the assumptions on the way in which the state probabilities depend on the level of effort. For example, let us assume that the probabilities of the low, medium, and high states are given by (1 -ph)(1 - e), (1 -ph)e, and ph, respectively. The effort shifts probability from the low state to the medium state, keeping the probability of the high state unchanged. Now the efforts are given by
The level of policy effort is now higher with the IMF. In the absence of the IMF the optimal effort is zero: the project is liquidated regardless of whether the low or the medium state realizes, so there is no benefit to increasing the probability of the medium state. With the IMF, by contrast, the project is liquidated in the low state but not in the medium state, so that it pays to increase the probability of the latter. IMF bailouts now act as a prize inducing more effort, rather than as a cushion reducing the punishment for low effort.
This ambiguity notwithstanding, it remains true that in our model, the anticipation of IMF bailouts may reduce the domestic policy efforts to avoid a balance of payments crisis. This phenomenon is generally interpreted as “debtor’s moral hazard.”
However, our model also shows that this interpretation is incorrect. Although effort may indeed be lower in the presence of the IMF, this is not moral hazard in the usual sense of reducing effort to a second-best level. Under the assumptions made—in particular, A1 and A2—the IMF reduces the effort from a level that is inefficiently high under laissez-faire. Without the IMF, the policy effort was too high because the discipline provided by the market—the threat of liquidation in the medium state—was excessively harsh. The presence of the IMF reduced the level of effort from a second-best level to its first-best level (given the available repayment technology β).8 In short, under the assumptions made, lower effort is not the reflection of a distortion. It is simply an efficient response to an improved repayment technology, much as faster driving may be an efficient response to better roads, or the existence of airbags.9
Country and global welfare, furthermore, are strictly increased by the IMF. Ex ante welfare is increased by the expected output loss that is avoided with IMF bailouts,
Welfare is higher with the IMF for any levels of k and e. Thus, welfare must be higher in equilibrium, irrespective of the IMF’s impact on the level of effort or capital. Because the ex ante welfare of the country’s creditors remains equal to their wealth W in equilibrium, the IMF also increases global welfare. These results are summarized in the Mussa theorem.
Mussa theorem. Under assumptions A1 and A2, the anticipation of IMF bailouts increases the volume of capital flows to emerging market countries and reduces the cost of borrowing for these countries. In addition, the anticipation of bailouts may decrease the domestic efforts to avoid a crisis. However, the IMF does not generate moral hazard stricto sensu. The expectation of IMF bailouts unambiguously increases the welfare of recipient countries at no cost to the rest of the world.
Two Doors for Moral Hazard to Come Back
The logic behind the Mussa theorem is very strong. If IMF lending increases the country’s welfare at no cost to the rest of the world, it cannot induce a distortion, even if it might produce all the symptoms of moral hazard. However, this result was obtained under two strong assumptions. We show in this section how relaxing these assumptions opens doors for moral hazard to come back.10
First, there was the assumption that the domestic decisions over k and e were taken by a benevolent policymaker. Some, including Calomiris (1998), have argued that IMF lending might aggravate the domestic distortions coming from the fact that domestic policymakers are not, in reality, benevolent. The IMF, according to this argument, might increase the scale of domestic expropriation in emerging market economies because of its ability to extract more repayment from the domestic taxpayers than private lenders could. Mike Mussa has referred to this as “indirect moral hazard” (Mussa, 2004).
There are different ways this story can be told in our model. Probably the simplest is to assume that the domestic policymaker maximizes his own utility rather than the welfare of the representative taxpayer, and that he is able to divert a given fraction of the capital inflows for his own consumption. Some of the external debt, then, becomes “odious” in the sense of Kremer and Jayachandran (2002). One can think of more indirect ways for policymakers to divert the domestic fiscal resources toward specific groups (for example, providing implicit guarantees on the liabilities of “crony capitalists”), but the logic of the argument would remain essentially the same.
The objective of the policymaker, in that case, would simply be to maximize capital inflows. Assume that the level of capital inflows is excessive because of a malevolent policymaker under laissez-faire (that is, the welfare of the representative resident decreases with capital inflows). Then the IMF, by allowing the policymaker to attract even more capital inflows, could end up decreasing domestic welfare.
Another way the IMF could induce a distortion is by lending at an excessively low interest rate, which would imply a transfer from the IMF’s financiers to the recipient countries (that is, assumption A2 is violated). In that case, the moral hazard distortion would be of the same nature as that generated by a deposit guarantee in a domestic banking context. The fact that the cost of a crisis will be subsidized by a third party (here, the financiers of the IMF) may induce emerging market countries and their lenders to indulge in excessively risky behavior. In this case, those who ultimately suffer from the distortion are not the domestic taxpayers in crisis countries but the financiers of the IMF (presumably, the international taxpayer) who lose part of their contribution to the Fund.
To make this point formally, we revisit the first specification of the three-states model (in which the effort shifts the probability from the medium state to the high state). Let us assume that an IMF bailout in the medium state involves a subsidy (a transfer from the IMF to the rescued country) of size σ ≤ dimf. (The transfer is equal to dimf if the country never repays anything to the IMF, but it cannot be larger.) Now the effort levels with and without the IMF are, respectively, given by
The subsidy reduces the effort by making the medium state more attractive relative to the high state. But in this case, it reduces the effort below the first-best level (which is achieved in the absence of subsidy). The IMF, therefore, reduces the effort from a laissez-faire level that is too high to a level that is too low. It increases the welfare of the emerging market countries but reduces the welfare of the rest of the world, and it could decrease welfare as a whole, if the subsidy σ is large enough.
There is also a third potential door by which IMF-induced moral hazard might potentially come back. If the policy effort of the government were a hidden action, IMF lending could potentially aggravate the moral hazard that is introduced in the original debt contract by the country’s inability to commit to e. However, it is easy to see that IMF lending might reduce the country’s welfare in this way only if it increased the interest rate at which the country can borrow ex ante.
Now the country sets its effort level after it has borrowed k at interest rate r. The welfare of a country that has borrowed k at interest rate r is given by
under laissez-faire and by
with the IMF. The country’s welfare is higher with the IMF for any k, r, e, and it is lower with r, given k and e. Assume that the country may borrow a given quantity of capital k at interest rate rlf under laissez-faire and at a lower interest rate rimf < rlf in the presence of the IMF. Then, introducing the IMF in the laissez-faire equilibrium without changing the country’s level of effort unambiguously raises welfare,
The increase in the country’s welfare is even higher when the effort is set at the optimal level given the existence of the IMF (e = eimf). Conditional on rimf < rlf, therefore, the IMF necessarily increases the country’s welfare without changing the welfare of investors. Global welfare can be lowered by the IMF only if IMF lending has the effect of raising the country’s cost of borrowing ex ante—which, paradoxically, most observers would interpret as the opposite of moral hazard.
Finally, to return to a point made above, it is worth noting that even when the assumptions of the Mussa theorem are violated and the IMF does generate moral hazard, this does not imply that this moral hazard is excessive. For example, it could be that the domestic political agency problem is mild enough that on balance, the IMF remains beneficial. Or it could be that the transfer in IMF lending is small enough that the induced distortion does not reduce global welfare. However, in the latter case, one has to worry about the redistributional consequences of IMF lending. The IMF does not induce a Pareto improvement in global welfare, since the welfare of emerging market countries is increased at the expense of rich countries. Some would say, of course, that this is redistribution in the right direction.
II. Implications of the Mussa Theorem
Implications for the Empirical Literature on IMF-Induced Moral Hazard
The Mussa theorem does not prove that there is no IMF-induced moral hazard. It merely states conditions—which may themselves be quite restrictive—under which IMF lending cannot give rise to moral hazard. However, as we have shown, even if these conditions are satisfied, and there is no IMF-induced moral hazard, IMF lending will generally have implications such as increasing capital flows or lowering the interest rates at which countries borrow—not because it creates moral hazard, but simply because it makes the world a safer place.
The implications of this insight for the empirical literature on moral hazard are quite profound. There is by now extensive literature that claims to “test for moral hazard” by looking at the effect of actual or expected IMF lending on either country risk or capital flows (see, among others, Zhang, 1999; Lane and Phillips, 2000; Dell’Ariccia, Schnabel, and Zettelmeyer, 2002; Kamin, 2002; Burkhalter, 2003; Haldane and Taylor, 2003; Bai and Wei, 2005; and Zoli, 2004; see also Dreher, 2004, for a comprehensive survey and additional references). Without exception, the tests performed in this literature are incapable of distinguishing whether the effects of the IMF on market variables (to the extent that any are found) are a sign of moral hazard or simply an indication that the IMF is doing its job. This is not to say that these papers may not be valuable—clearly, the question of whether the IMF has an impact on international capital markets in the direction predicted by standard models (including models without any moral hazard) is an important one. However, they are not tests of moral hazard.11
The same objection applies to a (much smaller) body of literature that looks at the effect of IMF lending on debtor country policies from the perspective of testing for moral hazard. For example, Dreher and Vaubel (2001) find that, controlling for the state of a country’s economy, countries that had obtained IMF credit ran more expansionary monetary and fiscal policies the less they had exhausted their borrowing potential in the IMF and the more credit they had received. But it is the intended purpose of IMF lending to allow countries experiencing balance of payments difficulties to run more expansionary fiscal and monetary policies than they otherwise could. Whether this is efficient or a source of moral hazard depends on whether in the absence of IMF lending monetary and fiscal policies would be inefficiently tight (as may well be the case if countries cannot smooth consumption by borrowing in international capital markets) or whether IMF lending accommodates, or indeed exacerbates, an expansionary bias on the part of the authorities. This distinction cannot be made based on the existing literature.
A more fruitful strategy in “testing for moral hazard” is to examine whether the assumptions of the Mussa theorem are satisfied. In particular, the Mussa theorem establishes a precise empirical standard to determine whether IMF lending contains a subsidy or not. Note that this standard differs markedly from the standard proposed by some IMF critics, which is to compare IMF lending rates with market interest rates faced by the crisis countries to which the IMF lends. As stated by assumption A2, avoiding moral hazard requires only that the IMF rate of charge be high enough to cover the repayment risk faced by the IMF, which may be substantially lower—as a result of the IMF’s political clout or its ability to directly affect a country’s solvency by imposing conditionality—than the risk borne by private lenders. By this standard, does IMF lending carry a subsidy or not?
A first step in answering this question is to look at the rate of return that the IMF has actually received in its lending to various groups of countries over the past decades. To the extent that the IMF either charged subsidized interest rates or suffered major nonrepayment problems, the rate of return should be lower than the international risk-free rate, and in this case, IMF lending would undoubtedly carry a subsidy. Zettelmeyer and Joshi (2004) show that this is indeed the case for lending to low-income countries, which carried an average annual rate of return, in the 1973–2003 period, about 400 basis points below the rate that the IMF would have received from lending the same amount to industrial countries at comparable maturities. This largely reflects a combination of deliberately concessional lending and debt relief under the Highly Indebted Poor Countries initiative. In contrast, the spread between IMF rates of return and lending rates by industrial countries has been approximately zero for upper-middle-income countries (the group usually referred to as “emerging markets”). This is not surprising given that, with the exception of a few arrears cases in the 1980s, none of these countries have ever defaulted on the IMF, and the interest rate charged by the IMF in lending to these countries is calculated in terms of a spread over short-term money market interest rates in industrial countries.
By this test, then, the IMF seems to have been lending at an actuarially fair interest rate to emerging market countries. However, it is a fairly weak test, since Zettelmeyer and Joshi’s exercise assumes that currently outstanding credit to the IMF will be repaid in line with IMF repayment projections. Is this assumption justified? According to our earlier work in this area, the answer—perhaps surprisingly—is “yes.” In Jeanne and Zettelmeyer (2001), we compared the length of currently open IMF “lending cycles” (uninterrupted periods of strictly positive obligations to the Fund) with the length of lending cycles that had been fully repaid in the past, using a maximum likelihood procedure to decide whether currently outstanding IMF loans “looked” like they would eventually be repaid.12 The upshot of the exercise was that because there have been many long lending cycles with emerging markets that eventually resulted in full repayment, the vast majority of the currently open lending cycles to these countries (about 95 percent) were predicted to be eventually repaid in full.
Like any exercise that focuses on the IMF’s repayment record, our 2001 characterization of emerging market lending as virtually subsidy free is convincing only if one assumes that the past is a good guide to the future. It is here that the potential problem lies. On the one hand, the shift to large-scale crisis lending that began with the 1995 Mexican loan does not seem to have constituted a structural break with respect to the IMF’s repayment record: all large-scale crisis loans made between 1995 and 1999 have by now been fully repaid, or are well on their way to being repaid.13 On the other hand, as Mike Mussa has argued on several occasions, more recent large-scale loans, such as the loans to Argentina in late 2000 and 2001, might be a different matter (see also Rogoff, 2002). What differentiates the loans to Argentina is that they were made when Argentina already had substantial outstanding obligations to the Fund, and that they were not made in response to a capital account crisis (as was the case for Mexico in 1995 and the Asian crisis loans) but rather in response to fiscal problems that carried the risk of a capital account crisis. When this crisis eventually materialized, it had an impact on all of Argentina’s creditors, including the IMF. Thus, to the extent that this brand of large-scale lending carries a higher risk of default than was typical for either standard lending or large-scale crisis lending in the past, it would clearly imply a subsidy under current lending terms and hence could be a source of moral hazard.
The other assumption in Mussa’s theorem is that debtor country governments act in the best long-run interests of their taxpayers. One does not have to be a cynic to doubt that this is the case. The question is how systematic the problem is in relation to IMF lending, that is, whether large-scale IMF lending has tended to go to countries whose precrises policies, subsequent crises, adjustment, and repayment to the IMF had the joint effect of redistributing wealth at the expense of the average domestic taxpayer. One area of research related to this question is the distributional effects of IMF programs (see Vreeland, 2003, Chapter 6). However, the literature in this area has tended to focus narrowly on the effects of IMF conditionality, rather than on the broader and more difficult issue of whether the combination of financial guarantees that primarily benefit a small financial and entrepreneurial elite, although they are backed by the average taxpayer, has important distributional effects. This could be the case even without any IMF conditionality. IMF-induced moral hazard through this channel remains an underexplored and, in our view, critical area of empirical research.
Implications for IMF Lending Policies
From the perspective of avoiding moral hazard in IMF lending, the policy implications of the Mussa theorem are straightforward: avoid lending at terms and in circumstances that would lead to a violation of assumptions A1 and A2. How this can be done in practice?
As far as avoiding implicit transfers in IMF lending is concerned, Mike Mussa has argued—and we agree with him—that this would be achieved under the IMF’s current interest charges provided that lending is “temporary” and takes place “under adequate safeguards,” as envisaged in the Articles of Agreement. Indeed, the IMF’s current rules imply that lending from its general resources (as opposed to deliberately concessional lending funded through special facilities) should not contain a transfer element. In particular, the IMF is allowed to lend only to countries that will, in all likelihood, be in a position to repay the Fund in the medium term. Furthermore, according to new guidelines issued in February 2003, the IMF is allowed to lend in large amounts (exceeding the usual limit of 100 percent of quota in any given year and 300 percent of quota in total) only when a “member is experiencing exceptional balance of payments pressures on the capital account resulting in a need for Fund financing,” when “a rigorous and systematic analysis indicates that there is a high probability that the debt will remain sustainable,” and when “the member has good prospects of regaining access to private markets within the time Fund resources would be outstanding.” These guidelines are meant to ensure that large-scale crisis lending—the suspected source of moral hazard with regard to emerging market countries—remains temporary and that adequate safeguards are in place.
However, while most instances of large-scale IMF crisis lending since 1995 would have satisfied the 2003 guidelines, the most recent cases—ironically, the only instances of large-scale crisis lending that took place after the guidelines were issued—did not (see IMF, 2004). The reasons for this could reflect, in some combination, the fact that these criteria were tailored too narrowly to the capital account crises of the 1990s, a standard time-consistency problem—lending might be optimal ex post even though refusing it is optimal ex ante—and, as Mike Mussa would put it, “geopolitical concerns.” Regardless of the cause, this experience shows that criteria that are meant to limit the circumstances in which large-scale IMF crisis lending can occur could be violated. This could be a source of moral hazard.
Avoiding the “indirect moral hazard” that may arise when the IMF supports governments that fail to act in the best interests of their taxpayers is conceptually harder than avoiding a subsidy in lending. As Mussa has pointed out, it would obviously go too far to refuse financial assistance to all but the most virtuous and competent governments. From the perspective of avoiding IMF-induced moral hazard, the key issue is not so much precrisis policy mistakes per se, but rather whether the distributional implications of these mistakes—and thus the propensity of governments representing special interests to commit such “mistakes”— might be magnified by the prospect of IMF liquidity support. In those cases, the IMF should not lend.
How can this be operationalized? For one thing, as Mussa has argued, the IMF must not support “gambles for resurrection,” desperate precrisis measures that benefit the government and the ruling elite by maintaining a possibility that the crisis can be avoided, at enormous expense to the domestic taxpayer in the event that the attempt fails. Others, including Peterson, Goldstein, and Hills (1999), the International Financial Institutions Advisory Commission (2000), and ourselves in our 2001 paper, would take a broader and more ambitious view of the desirable link between precrisis policies and access to IMF lending, perhaps including the level and structure of outstanding debt, its transparency with respect to data, the quality of institutions regulating the financial sector, and its responsiveness to IMF surveillance. Again, the key challenge in creating such a link in practice is to overcome the time-consistency problem that it inevitably raises.
III. Conclusions
The Mussa theorem is a statement of conditions under which official crisis lending does not have adverse incentive effects. It is pivotal to understanding the welfare implications of crisis lending as distinct from conventional forms of insurance involving conditional transfers. Whether it holds empirically is an open issue—quite possibly it does not. But this does not lessen its importance. Like the Modigliani-Miller theorem, which undoubtedly does not hold, the contribution of the Mussa theorem to the debate on IMF-induced moral hazard consists in defining an intellectual benchmark. By defining the conditions under which IMF-induced moral hazard cannot exist, it establishes a set of potential channels or mechanisms through which moral hazard could in fact arise in practice.
The Mussa theorem has important implications for the empirical literature on IMF-induced moral hazard. By pointing out that lower borrowing costs or larger capital flows are the expected consequence of any crisis-lending mechanism, regardless of whether it leads to moral hazard, it implies that the standard “test for moral hazard” used in the empirical literature is not meaningful. What this literature does is test whether IMF crisis lending has effects on international capital markets in the direction that most models of the IMF would predict—with or without moral hazard. The same applies to papers that observe that IMF lending allows countries to run more expansionary policies than they otherwise could. This may be the intended consequence of IMF lending and is not proof of moral hazard. However, the Mussa theorem is also constructive: it suggests an alternative empirical approach—namely, to test for the validity of its conditions directly.
In the policy arena, perhaps the most important contribution of the Mussa theorem is to define, in precise terms, the meaning of a moral-hazard-inducing “subsidy” in IMF lending. Time and time again, this has been a subject of misunderstanding, with IMF critics arguing that IMF lending incorporates a subsidy at the expense of creditor-country taxpayers to the extent that its interest charges are lower than those charged by private borrowers (see Higginbotham and Schuler, 2002, for a recent example). The Mussa theorem shows that this is an incorrect standard, except in circumstances when the IMF faces the same default risk as private borrowers. From the perspective of avoiding moral hazard, what matters is whether IMF interest rates are actuarially fair in the sense that they cover the risk faced by the IMF. This will generally be lower than that faced by private borrowers.
Finally, the Mussa theorem suggests that the IMF could be a source of moral hazard to the extent that its lending amplifies preexisting moral hazard problems in borrowing countries. In avoiding this source of moral hazard, the general principle must be to relate access to IMF financing to the quality of country policies and institutions prior to the crisis. This is easier said than done, however, particularly because it raises a time-consistency problem. Whether this problem can be dealt with satisfactorily will determine the feasibility of incentive-friendly large-scale lending in the future.
REFERENCES
Bai, Jushan, and Wei, Shang-Jin 2005, “Induced Imprudence? A New Test for Moral Hazard in International Lending” (unpublished; Washington: International Monetary Fund).
Borensztein, Eduardo, and Mauro, Paolo 2004, “The Case for GDP-Indexed Bonds,” Economic Policy, Vol. 19 (April), pp. 165 –218.
Burkhalter, Erika, 2003, “Argentina: A Case for Moral Hazard,” (unpublished; senior honors thesis; Pittsburgh, Pennsylvania: Carnegie-Mellon University).
Calomiris, Charles W., 1998, “The IMF’s Imprudent Role as Lender of Last Resort,” Cato Journal, Vol. 17, No. 3, pp. 275 –94.
Cordella, Tito, and Levy-Yeyati, Eduardo 2004, “Country Insurance,” paper presented at IMF conference in honor of Michael Mussa, Washington, D.C., June.
Corsetti, Giancarlo, Guimarães, Bernardo and Roubini, Nouriel 2003, “International Lending of Last Resort and Moral Hazard: A Model of IMF’s Catalytic Finance,” NBER Working Paper No. 10125 (Cambridge, Massachusetts: National Bureau of Economic Research).
Dell’Ariccia, Giovanni, Schnabel, Isabel and Zettelmeyer, Jeromin 2002, “Moral Hazard and International Crisis Lending: A Test,” IMF Working Paper 02/181 (Washington: International Monetary Fund).
Dreher, Axel, 2004, “Does the IMF Cause Moral Hazard? A Critical Review of the Evidence.” Available via the Internet: http://ssrn.com/abstract=505782.
Dreher, Axel, and Vaubel, Roland 2001, “Does the IMF Cause Moral Hazard and Political Business Cycles? Evidence from Panel Data,” Beiträge zur angewandten Wirtschaftsforschung No. 598–01 (Mannheim: University of Mannheim).
Haldane, Andrew, and Taylor, Ashley 2003, “Moral Hazard: How Does IMF Lending Affect Debtor and Creditor Incentives?” Financial Stability Review (June) (London: Bank of England), pp. 122–33.
Higginbotham, Brian, and Schuler, Kurt 2002, “The Subsidy in IMF Lending,” A Joint Economic Committee Study (Washington: Joint Economic Committee, United States Congress). Available via the Internet: http://www.house.gov/jec/imf/imfpage.htm.
Holmstrom, Bengt, and Tirole, Jean 1998, “Private and Public Supply of Liquidity,” Journal of Political Economy, Vol. 106, No. 1, pp. 1–40.
International Financial Institutions Advisory Commission, 2000, “Report of the International Financial Institution Advisory Commission.” Available via the Internet: http://www.house.gov/jec/imf/imfpage.htm.
International Monetary Fund, 2004, Review of Exceptional Access Policy (Washington).
Jeanne, Olivier, 2004, “Debt Maturity and the International Financial Architecture,” IMF Working Paper 04/137 (Washington: International Monetary Fund).
Jeanne, Olivier, and Zettelmeyer, Jeromin 2001, “International Bailouts, Moral Hazard, and Conditionality,” Economic Policy, Vol. 33 (October), pp. 409–32.
Jeanne, Olivier, and Zettelmeyer, Jeromin 2005, “Original Sin, Balance Sheet Crises, and the Roles of International Lending,” in Other Peoples’ Money: Debt Denomination and Financial Instability in Emerging Market Economies, ed. by Eichengreen Barry and Hausmann Ricardo (Chicago: University of Chicago Press), pp. 95–121. Also available as IMF Working Paper 02/234 (Washington: International Monetary Fund).
Kamin, Steven B., 2002, “Identifying the Role of Moral Hazard in International Financial Markets,”International Finance Discussion Paper No. 736 (Washington: Federal Reserve Board).
Kremer, Michael, and Jayachandran, Seema 2002, “Odious Debt,” NBER Working Paper No. 8953 (Cambridge, Massachusetts: National Bureau of Economic Research).
Kreps, David M., 1990, A Course in Microeconomic Theory (Princeton, New Jersey, and Oxford: Princeton University Press).
Lane, Timothy, and Phillips, Steven 2000, “Does IMF Financing Result in Moral Hazard?” IMF Working Paper 00/168 (Washington: International Monetary Fund).
Mussa, Michael, 1999, “Reforming the International Financial Architecture: Limiting Moral Hazard and Containing Real Hazard,” in Capital Flows and the International Financial System, ed. by Gruen David and Gower Luke (Sydney: Reserve Bank of Australia), pp. 216–36.
Mussa, Michael, 2004, “Reflections on Moral Hazard and Private Sector Involvement in the Resolution of Emerging Market Financial Crises,” in Fixing Financial Crises in the Twenty-First Century, ed. by Haldane Andrew (London: Routledge), pp. 33–51.
Mussa, Michael, Swoboda, Alexander Zettelmeyer, Jeromin and Jeanne, Olivier 2000, “Moderating Fluctuations in Capital Flows to Emerging Market Economies,” in Reforming the International Monetary and Financial System, ed. by Kenen Peter B. and Swoboda Alexander K. (Washington: International Monetary Fund), pp. 75–142.
Peterson, Peter G., Goldstein, Morris and Hills, Carla A. 1999, Safeguarding Prosperity in a Global Financial System: The Future International Financial Architecture, Council on Foreign Relations Independent Task Force Report (Washington: Institute for International Economics).
Rogoff, Kenneth S., 2002, “Moral Hazard in IMF Loans: How Big a Concern?” Finance and Development, Vol. 39, No. 3, pp. 56–7.
Tirole, Jean, 2002, Financial Crises, Liquidity, and the International Monetary System (Princeton, New Jersey, and Oxford: Princeton University Press).
Vreeland, James, 2003, The IMF and Economic Development (Cambridge: Cambridge University Press).
Weithöner, Thomas, 2003, “How Can IMF Policy Eliminate Country Moral Hazard and Account for Externalities?” (unpublished; Tübingen: University of Tübingen). Available via the Internet: www.thomas.weithoener.org.
Zettelmeyer, Jeromin, and Joshi, Priyadarshani 2005, “Implicit Transfers in IMF Lending, 1973–2003,” IMF Working Paper 05/8 (Washington: International Monetary Fund).
Zhang, Xiaoming Alan, 1999, “Testing for ‘Moral Hazard’ in Emerging Markets Lending,” Institute of International Finance Research Paper No. 99-1 (Washington: Institute of International Finance).
Zoli, Edda, 2004, “Credit Rationing in Emerging Economies’ Access to Global Capital Markets,” IMF Working Paper 04/70 (Washington: International Monetary Fund).
Olivier Jeanne and Jeromin Zettelmeyer are Deputy Division Chief and Assistant to the Director, respectively, in the Research Department of the International Monetary Fund. The authors are grateful to Michael Mussa and Alexander Swoboda for discussions on the topic, and to Tito Cordella and Carmen Reinhart for comments on earlier drafts. An earlier draft of this paper was presented at an IMF conference honoring Michael Mussa, Washington, D.C., June 4–5, 2004.
This debt structure can be endogenized (see, for example, Jeanne, 2004). We take it as given in this paper.
In addition there could be self-fulfilling rollover crises. For simplicity we leave this possibility aside. It could be added, however. Considering liquidity crises in which the IMF lends in last resort would strengthen the results, by adding a motive through which IMF lending helps avoid inefficient outcomes.
There are two sources of inefficiency in this model. First, there is the risk of liquidation, which can be removed by equity-like contracts in which creditors are promised a share of the final output y (for example, long-term GDP-indexed bonds, as proposed by Borensztein and Mauro, 2004). Second, there is the fact that the country cannot pledge all its output but only a share lower than α. This could result in credit rationing in the international financial market even if the equity-like contracts mentioned above were possible. The share of output pledged to foreign investors may have to be larger than α to avoid credit rationing and reach the first best level.
Another critical difference between the IMF and individual private lenders might simply be its size, which enables it to overcome coordination problems (see Jeanne and Zettelmeyer, 2005). We do not highlight this because we ignore self-fulfilling crises in this paper.
This could be extended to a setting in which we also introduce some crisis prevention action by investors.
As we shall see below, the results can be generalized to the case without commitment at the cost of a plausible additional assumption.
Here we use the fact that the country can commit to its effort level at the time the debt contract is signed. Thus, the nominal repayment d is adjusted to e at time 0 to keep the expected repayment equal to k.
Another way of seeing that there is no moral hazard stricto sensu in this model is that there is no hidden action and, accordingly, no strict benefit to endowing the country with the ability to commit to action e (as would be the case in a model with moral hazard).
That is, assuming that the costs of having an accident are fully borne by the driver. This is the equivalent of assumption A2, which rules out a transfer from the IMF to the country.
We focus on the two assumptions that were highlighted as A1 and A2. In addition to these, it is possible to relax other assumptions in our basic model with the effect that the presence of the IMF might magnify a previously existing distortion. For example, negative output externalities across countries might imply that effort could be suboptimally low under laissez-faire. This might be aggravated by the IMF even if IMF lending does not contain a subsidy (see Weithöner, 2003). However, in a broader sense, this example confirms the logic of the Mussa theorem—namely, that IMF lending cannot cause moral hazard unless lending occurs at the expense of a third party that is not involved in the actual lending decision. In this case, there would be an additional third party to consider.
In recognition of this point, Dell’Ariccia, Schnabel, and Zettelmeyer, 2002, as well as Bai and Wei, 2005, argue that they are only testing a necessary condition for moral hazard.
More technically, we formulated a duration model in which the probability of observing an IMF lending cycle of length t was modeled in terms of the probability of either observing an “infinite lending” cycle that would never be repaid or the probability of observing a “finite” cycle that had not been repaid yet but would be repaid with some hazard rate. The probability of being in an “infinite” cycle and the parameters of the hazard function were then estimated using the available data on the length of both completed and incomplete cycles.
Brazil’s 1998–99 loan had been largely paid off before it received new assistance from the IMF in mid-2001. Indonesia’s obligations to the Fund peaked in September 2000 and have since declined by about 20 percent.
