Conclusions in this section are based on case studies presented in Allen and others (2002) and Rosenberg and others (2005).

See Box 2.1 of Roubini and Setser (2004) for a comparison of assumptions in different generations of models.

To use an analogy, lightning strikes might leave a house at risk of burning down and while measures can be taken to reduce that risk (e.g., installing a lightning conductor), some risk may be unavoidable. By purchasing insurance, however, the homeowner transfers the associated financial risk from his own relatively weak, undiversified balance sheet to that of the insurance company, which is much stronger in that it holds diversified risks.

This suggests that, when the government’s balance sheet is relatively weak, multilateral organizations could usefully issue debt denominated in emerging market country currencies, thus providing a domestic-currency-denominated asset to the banking sector without the corresponding default risk. Multilateral organizations would, however, assume the corresponding currency risk.

Over the past few years, the Brazilian government has gradually eliminated much of its foreign-currency-indexed debt.

For example, a bank may be closing its spot foreign exchange exposure through a derivative transaction with its parent conglomerate; such practices apparently occurred in Turkey prior to the 2000 crisis.

The IMF’s Independent Evaluation Office’s evaluation “Report on the Evaluation of the Role of the IMF in Argentina, 1991–2001” provides a discussion of related factors (see www.ieo-imf.org). Also, as pointed out in Daseking and others (2005), the exchange rate guarantee implicit in a pegged regime (or currency board) cannot simultaneously explain both asset and liability dollarization. For instance, if the peg is credible, households may want to borrow in foreign currency (since foreign exchange interest rates are typically lower and there is little risk of a devaluation) but then they would not want to hold dollar deposits. Conversely, if there are doubts about the viability of the peg, households would want to hold dollar deposits but not borrow in foreign currency. Empirically, there does not seem to be any association between pegged exchange rate regimes and dollarization of the banking system.

Among the purposes of the IMF, as listed in Article I of the Articles of Agreement, is “to give confidence to members by making the general resources of the Fund temporarily available to them” (emphasis added). As Sir Joseph Gold points out, a Stand-By Arrangement gives confidence by allowing a member “to ensure that it would be able to draw if, within a period of 6 or 12 months, the need presented itself” (see Gold, 1970, pp. 23–24).

This section draws on IMF (2006a).

Excluding Brazil (2001), the one case of exceptional access at the outset of a precautionary program and that later turned nonpre-cautionary, average access under precautionary arrangements was 31 percent of quota.

For a discussion of disbursement patterns in precautionary arrangements, as well as possible alternatives, see IMF (2003).

This allows for upper credit tranche conditionality in the arrangement, which applies once the country’s outstanding IMF credit exceeds 25 percent of quota. Only Argentina (2000) and Paraguay (2003) have received more than 25 percent of quota at the approval of a precautionary arrangement.

Capital account crisis cases are excluded from these figures because the magnitude and abruptness of capital outflows means that the behavior of these economies is different from the “classical” programs supported by the General Resources Account (GRA) (Ghosh and others, 2005), and including them in the sample of drawing programs would necessarily bias the comparison in favor of precautionary programs. They are also treated separately in the econometric analysis below. Section IV examines the special case of capital account crises.

Specifically, an analysis of current account balances relative to debt-stabilizing current account balances suggests that only one-fifth of members with precautionary arrangements are “underadjustors” in the sense that their current account balance falls short of the debt-stabilizing balance even though their external debt is relatively high (exceeding 40 percent of GDP). In general, there is a positive relationship between members’ current account balances (relative to the debt-stabilizing balance) and their external debt; this relationship is statistically identical for members with precautionary and drawing programs.

The authorities’ decision is modeled here as a simultaneous choice between requesting a precautionary program, a drawing program, or none at all. Sequential decision trees are also possible; for instance, the authorities could first decide to request IMF support, and then decide whether or not to treat the program as precautionary. For logical consistency, however, such sequential modeling structures require Independence of Irrelevant Alternatives (IIA) so that the second-stage choices are independent of the first stage (see Ben-Akiva and Lerman, 1987, for a discussion). Since the IIA assumption does not hold empirically in this dataset, the simultaneous modeling structure was adopted.

The index values of perceptions are based on assessments of political risk made by a statistical model of risk developed by the PRS Group (International Country Risk Guide indicators).

For variables that are defined in percentage terms (percent a year or percent of GDP), the coefficients represent the effects of a 1 percentage point change in the explanatory variable on the percentage change in the probability of choosing that particular option. For example, a current account deficit that is 1 percent of GDP higher than the mean value would lead to a 34 percent (not percentage point) increase in the probability of choosing a precautionary program (rather than no program). For variables that are scalars, the coefficient estimate is an elasticity so that a 20 percent decline in the index of internal conflict (which corresponds to one standard deviation) would lead to an 84 percent increase in the probability of choosing a precautionary program.

These estimates are based on the first program year. A similar choice model was also estimated for the whole program period for use in the analysis below of macroeconomic performance over the whole program period. A version of the model based on monthly data was estimated for the sovereign spreads analysis below.

Robustness tests were carried out by including the level and change in private capital flows, measures of equity market volatility derived from market prices of call options on equity futures, and a market pressure index based on a weighted average of exchange rate and reserve changes. None of these variables was statistically significant, nor did its inclusion affect the statistical significance of other variables.

This section examines the effect on secondary market spreads; other papers—such as Mody and Saravia (2003) and Eichengreen, Kletzer, and Mody (2005)—have looked at the effect on spreads of new bonds issued during IMF-supported programs. They find that spreads during these periods are lower than at other periods. Since the timing of bond issuance is endogenous, the decline in spreads could reflect authorities choosing to issue bonds at the most opportune time.

These explanatory variables do not capture all of the economic and other factors that determine spreads. Drawing programs, particularly capital account crises, are associated with higher spreads relative to nonprogram periods or members with no IMF-supported program in the sample. This may suggest omitted variables, nonlinear relationship, or, possibly, stigma.

Robustness checks also considered a dummy variable capturing the announcement date of subscription to the Special Data Dissemination Standard and measures of equity market volatility derived from market prices of call options on equity futures. Inclusion of such variables did not affect the results presented here.

Excluding precautionary arrangements that immediately followed a drawing arrangement yields similar results. Moreover, spreads were higher in countries that had a similar degree of political uncertainty as that prevailing in precautionary programs but without an IMF-supported program.

See Ghosh and others (2002) and Ghosh and others (2005).

This paper draws on two recent IMF working papers—a theoretical framework by Kim (2006) and an empirical analysis by Ramakrishnan and Zalduendo (2006).

Some researchers have examined the effects of IMF financial support on private capital flows; see Cottarelli and Giannini (2002) and Bird and Rowlands (2002) for a survey of the empirical literature. These studies find limited or no evidence of catalytic effects except on official financing sources. The IMF’s Occasional Papers No. 210 and 241 (Ghosh and others, 2002, and Ghosh and others, 2005, respectively) find that IMF-supported programs in capital account crisis cases have a much smaller catalytic effect than anticipated. Other papers have looked at the effects on spreads. Here too the evidence is mixed. Haldane (1999) argues that the existence of a program increases spreads, while Eichengreen and Mody (1998) and Mody and Saravia (2003) find evidence that IMF-supported programs reduce spreads on new issues of bonds. These various papers have not, however, examined whether the IMF may have a catalytic role in crisis prevention situations.

See, for example, Ghosh and others (2002).

See Flood and Marion (1998) for a survey of currency crisis models. Zettelmeyer (2000) shows that official crisis lending limited in size relative to potential outflows can have counterproductive short-run effects—financing, rather than forestalling, a run—a result that depends primarily on the existence of multiple equilibria. In Morris and Shin (2005), however, the “global games” framework allows for a unique equilibrium for the creditor coordination problem. By using this global games framework, Corsetti, Guimaraes, and Roubini (2003) find similar results to those of Morris and Shin; namely, IMF liquidity support has a (nonlinear) catalytic effect and, under certain conditions, can encourage stronger policies. Penalver (2002) reaches a similar conclusion but focuses on the effect on longer-term capital flows of the IMF’s subsidized liquidity support. For a model of how IMF lending can reduce the probability of a crisis through a combination of providing liquidity and supporting stronger policies, see Kim (2006). A paper by Eichengreen, Gupta, and Mody (2005) looks at the effects of IMF support in preventing sudden stops.

On the costs of holding reserves, see Rodrik (2006). Rodrik estimates the cost of holding reserves at more than 1 percent of GDP, on average, for developing countries.

Alternatively, the desired level of reserves may increase—for example, because U.S. interest rates have risen, making an exit by creditors more likely and raising the likelihood of a crisis. In either case, as with most inventory-theoretic models, the country would not, in general, find it optimal to hold such a high level of reserves that the probability that its reserves dip below the optimal level would become negligible.

This risk of “debtor moral hazard” is likely to be greater in crisis prevention programs than in crisis resolution situations. In a capital account crisis (once it has erupted), the degree of external adjustment is often determined residually, given the withdrawal of private financing and the availability of official financing; see Ghosh and others (2002). In crisis prevention situations, by contrast, since private financing has not withdrawn, national authorities have greater latitude in determining how much adjustment to undertake—which gives rise to the greater possibility of debtor moral hazard. For a comprehensive discussion of possible moral hazard effects either on borrowing members or on private creditors, see IMF (2007).

Kim (2006) shows that a program with IMF financing and stronger policies (relative to the no-program situation) will indeed be welfare enhancing for the member relative to not having an IMF-supported program, and results in a correspondingly lower likelihood of a crisis.

For a discussion see IMF (2005), paragraph 9.

As discussed in Kim (2006), the IMF’s signaling role is enhanced (and thus the likelihood of a crisis is further reduced) by the IMF putting its own resources on the line—especially when the IMF has an informational advantage over private creditors regarding the authorities’ policy intentions. For more general discussions, see IMF (2004) and Cottarelli and Giannini (2002).

See Ramakrishnan and Zalduendo (2006) for a more detailed discussion.

Each of these terms in the index is standardized (mean equal to zero, standard deviation equal to one). A similar approach has been used in other studies that attempt to identify currency crises (see, e.g., Kaminsky and Reinhart, 1999).

The countries are Algeria, Argentina, Brazil, Bulgaria, Chile, Colombia, the Dominican Republic, Ecuador, Hungary, Indonesia, Korea, Malaysia, Mexico, Morocco, Pakistan, Panama, Peru, the Philippines, Poland, Russia, South Africa, Thailand, Tunisia, Turkey, Ukraine, Uruguay, and República Bolivariana de Venezuela. Country coverage is based on data availability during 1994–2004.

In a nutshell, cluster analysis is a technique that minimizes differences within each cluster of data and maximizes those across different data clusters (see Everitt, 1993). While the number of clusters is arbitrary, five clusters give a reasonable span to capture a range between strengthening, neutral, and weakening pressures on the balance of payments.

The cluster analysis identifies medium capital outflows to be in the range of 10 to 20 percent of GDP and large capital outflows to be over 20 percent of GDP.

For example, in Argentina, the July 2001 market pressure event is classified as a capital account crisis (i.e., 2001 Q3 (period t) = 1); hence, in the logit estimation, the dependent variable would be specified as 2001 Q2 = 1, 2001 Q1 = 1, 2000 Q4 = 1, and 2000 Q3 = 1. In contrast, the Argentina 1998 episode enters the regression with zeros because it is a control group.

Robustness checks show that this approach has no bearing on the main results beyond facilitating convergence of the maximum likelihood estimation—that is, the 32 pressure episodes and 4 quarters of data result in a dataset of 128 observations, but the results with 32 observations are consistent.

Growth and inflation performance (prior to the crisis) appear to differ between crisis and control group cases (see Figure 4.1). However, adding these variables to the logit estimation does not alter the thrust of the conclusions presented in Table 4.3.

The estimations control for changes in terms of trade. Other international cyclical factors (e.g., U.S. interest rates) were considered, but made the convergence of the maximum likelihood estimation more difficult and in the end had no bearing on the results.

More precisely, the IMF financing variable in period t−1 is calculated as the ratio of the sum of available IMF financing from t−4 to t−1 to short-term debt at end-t−1; the value in t−2 is calculated as the ratio of available IMF financing from t−5 to t−2 to short-term debt at end-t−2; and so on for earlier periods up to t−4. Since the sample includes only two precautionary arrangements, it is not possible to distinguish econometrically between the effects of disbursed IMF resources and those that are available (but not disbursed) under on-track precautionary arrangements. However, excluding these precautionary programs from the sample yields very similar results.

Country size likely captures the country’s financing needs in relation to funds available to emerging market countries.

The improvement in the fiscal balance (median values) in the year prior to the high market pressure event is about ¼ percent of GDP in countries receiving IMF financing, compared to a deterioration of ½ percent in countries without IMF financing. In terms of monetary tightening, real interest rates increase by 75 basis points among countries with IMF financing; the increase in countries without IMF financial support is 25 basis points.

While some studies differentiate between on-track and off-track (and thus nondisbursing) programs, they generally do not take account of the amount of IMF financing disbursed (or available under an on-track precautionary program).

Fiscal adjustment and monetary tightening in the year prior to the high market pressure event is greater in countries that had on-track IMF-supported programs than in countries without such programs. These policy variables may be capturing the stronger policies associated with an IMF-supported program, contributing to the lack of statistical significance of the dummy variable for the existence of an IMF-supported program.

The results remain robust to alternative definitions, such as IMF financing as a ratio to GDP.

See IMF (2006b) for details of this calculation.

Further regressions (not reported) indicate that overvaluation in the context of a pegged exchange rate regime makes the country especially vulnerable to a crisis. While this underscores the importance of avoiding overvalued fixed exchange rates, it also means that implementing even a relatively modest correction may not be straightforward, with potentially significant costs in terms of the credibility of the regime or arising from balance sheet exposures of the private and public sectors if the exchange rate overshoots in the process of exiting the regime.

The average access to lower the probability of a crisis to 25 percent would be 345 percent of quota. Lowering this probability to 10 percent and 5 percent, respectively, would require, respectively, 410 and 460 percent of quota, revealing a nonlinear relationship between access and crisis probabilities. These calculations keep constant policies (and other covariates) although, in practice, policies would be stronger under an IMF-supported program with higher access, therefore contributing to a lower likelihood of a crisis. Within these averages, the amounts needed relative to quota vary across countries in part because quotas do not always correlate closely with the economic circumstances of the country.

In fact, fundamentals typically deteriorate significantly during the crisis (from period t onward), but these effects are not included in the econometric estimation.